doi
stringlengths 10
10
| chunk-id
int64 0
936
| chunk
stringlengths 401
2.02k
| id
stringlengths 12
14
| title
stringlengths 8
162
| summary
stringlengths 228
1.92k
| source
stringlengths 31
31
| authors
stringlengths 7
6.97k
| categories
stringlengths 5
107
| comment
stringlengths 4
398
⌀ | journal_ref
stringlengths 8
194
⌀ | primary_category
stringlengths 5
17
| published
stringlengths 8
8
| updated
stringlengths 8
8
| references
list |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
2307.14430 | 40 | RedPajama We use SKILL-IT to produce a data mixture on the RedPajama dataset. The training skills are the data sources comprising the dataset, and the evaluation skills are several tasks from the Language Model Evaluation Harness [14]. SKILL-IT with T = 1 (i.e. a static, graph-based mixture) yields the mixture in Figure 7 (right). We continually pre-train a 3B parameter model trained on one trillion tokens for three billion additional tokens using this mixture, and see that it outperforms uniform sampling over the data sources (Figure 7 left). In particular, SKILL-IT achieves higher accuracy with 1B additional tokens than uniform with 3B additional tokens. | 2307.14430#40 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14225 | 41 | RecSys â23, September 18â22, 2023, Singapore, Singapore
Scott Sanner, Krisztian Balog, Filip Radlinski, Ben Wedin, and Lucas Dixon
explainable and scrutable language-based preference representation, thus providing a path forward for eï¬ective and novel LLM-based recommenders using language-based preferences.
# REFERENCES
[1] Jacob Austin, Augustus Odena, Maxwell Nye, Maarten Bosma, Henryk Michalewski, David Dohan, Ellen Jiang, Carrie Cai, Michael Terry, Quoc Le, and Charles Sutton. 2021. Program Synthesis with Large Language Models. arXiv:2108.07732 [cs.PL]
[2] Krisztian Balog, Filip Radlinski, and Shushan Arakelyan. 2019. Transparent, Scrutable and Explainable User Models for Personalized Recommen- dation. In Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR â19). 265â274. | 2307.14225#41 | Large Language Models are Competitive Near Cold-start Recommenders for Language- and Item-based Preferences | Traditional recommender systems leverage users' item preference history to
recommend novel content that users may like. However, modern dialog interfaces
that allow users to express language-based preferences offer a fundamentally
different modality for preference input. Inspired by recent successes of
prompting paradigms for large language models (LLMs), we study their use for
making recommendations from both item-based and language-based preferences in
comparison to state-of-the-art item-based collaborative filtering (CF) methods.
To support this investigation, we collect a new dataset consisting of both
item-based and language-based preferences elicited from users along with their
ratings on a variety of (biased) recommended items and (unbiased) random items.
Among numerous experimental results, we find that LLMs provide competitive
recommendation performance for pure language-based preferences (no item
preferences) in the near cold-start case in comparison to item-based CF
methods, despite having no supervised training for this specific task
(zero-shot) or only a few labels (few-shot). This is particularly promising as
language-based preference representations are more explainable and scrutable
than item-based or vector-based representations. | http://arxiv.org/pdf/2307.14225 | Scott Sanner, Krisztian Balog, Filip Radlinski, Ben Wedin, Lucas Dixon | cs.IR, cs.LG | To appear at RecSys'23 | null | cs.IR | 20230726 | 20230726 | [
{
"id": "2305.06474"
},
{
"id": "2305.07961"
},
{
"id": "2009.13292"
},
{
"id": "2204.02311"
},
{
"id": "2210.06280"
},
{
"id": "2005.14165"
},
{
"id": "2108.07732"
},
{
"id": "2306.02250"
},
{
"id": "2304.04250"
},
{
"id": "2305.08845"
},
{
"id": "2109.01652"
}
] |
2307.14430 | 41 | 5 Related work Data selection for LMs There have been several studies of large-scale data selection for LMs. Data deduplication [1, 22, 32], in which identical or nearly identical samples are removed, is a method that enables LMs to be trained on smaller, cleaned datasets and has been increasingly used as a pre-processing step for training data [4, 59, 71]. Other methods applied at scale involve ensuring high quality of data by explicitly ï¬ltering out samples or comparing the training dataset with a cleaned reference dataset [7, 31, 59]. Importance reweighting approaches have also been proposed for identifying training data from a large corpus that best approximates a smaller target distribution [69], and inï¬uence functions have been used to select a subset of training data to improve performance on downstream tasks [61]. These approaches can identify data
9
pertaining to a particular target distribution or ï¬lter out low quality data according to some heuristic, while our work aims to understand how the choice of data is related to the numerous skills that LMs learn. | 2307.14430#41 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14225 | 42 | [3] Krisztian Balog, Filip Radlinski, and Alexandros Karatzoglou. 2021. On Interpretation and Measurement of Soft Attributes for Recommendation. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR â21). 890â899.
[4] Toine Bogers and Marijn Koolen. 2017. Deï¬ning and Supporting Narrative-Driven Recommendation. In Proceedings of the Eleventh ACM Conference on Recommender Systems (RecSys â17). 238â242.
[5] Vadim Borisov, Kathrin SeÃler, Tobias Leemann, Martin Pawelczyk, and Gjergji Kasneci. 2023. Language Models are Realistic Tabular Data Gener- ators. arXiv:2210.06280 [cs.LG] | 2307.14225#42 | Large Language Models are Competitive Near Cold-start Recommenders for Language- and Item-based Preferences | Traditional recommender systems leverage users' item preference history to
recommend novel content that users may like. However, modern dialog interfaces
that allow users to express language-based preferences offer a fundamentally
different modality for preference input. Inspired by recent successes of
prompting paradigms for large language models (LLMs), we study their use for
making recommendations from both item-based and language-based preferences in
comparison to state-of-the-art item-based collaborative filtering (CF) methods.
To support this investigation, we collect a new dataset consisting of both
item-based and language-based preferences elicited from users along with their
ratings on a variety of (biased) recommended items and (unbiased) random items.
Among numerous experimental results, we find that LLMs provide competitive
recommendation performance for pure language-based preferences (no item
preferences) in the near cold-start case in comparison to item-based CF
methods, despite having no supervised training for this specific task
(zero-shot) or only a few labels (few-shot). This is particularly promising as
language-based preference representations are more explainable and scrutable
than item-based or vector-based representations. | http://arxiv.org/pdf/2307.14225 | Scott Sanner, Krisztian Balog, Filip Radlinski, Ben Wedin, Lucas Dixon | cs.IR, cs.LG | To appear at RecSys'23 | null | cs.IR | 20230726 | 20230726 | [
{
"id": "2305.06474"
},
{
"id": "2305.07961"
},
{
"id": "2009.13292"
},
{
"id": "2204.02311"
},
{
"id": "2210.06280"
},
{
"id": "2005.14165"
},
{
"id": "2108.07732"
},
{
"id": "2306.02250"
},
{
"id": "2304.04250"
},
{
"id": "2305.08845"
},
{
"id": "2109.01652"
}
] |
2307.14430 | 42 | Recent development of LMs has shifted focus from emphasizing the scale of the model to prioritizing the training data utilized. For example, models like Alpaca [56], Vicuna [9], and Koala [15] are all based on the LLaMA model combined with instruction data generated by an existing LM. Palm 2âs technical report states that the data mixture was a critical component of the ï¬nal model [17], and Mosaic MLâs recent MPT model was trained on a hand-engineered mixture of the RedPajama dataset [42]. However, these works lack rigorous explanation for why their training datasets were constructed in this way.
Finally, perhaps most related to our approach is the contemporary work DoReMi [68], which uses group distributionally robust optimization on a smaller LM to select data source mixtures for training a larger LM. Their approach focuses on selecting data at the data source level for optimizing worst-case performance across the training data sources, rather than at the more general skills level for a variety of target skill sets. Furthermore, we focus on understanding how skills are related to each other and induce some order in how LMs learn by explicitly modeling skill graph structure, which we ï¬nd to be important for data-efï¬cient LM training (see ablations in Appendix D.5). | 2307.14430#42 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14225 | 43 | [6] Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeï¬rey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz Litwin, Scott Gray, Benjamin Chess, Jack Clark, Christopher Berner, Sam McCandlish, Alec Radford, Ilya Sutskever, and Dario Amodei. 2020. Language Models are Few-Shot Learners. arXiv:2005.14165 [cs.CL]
[7] Arun Tejasvi Chaganty, Megan Leszczynski, Shu Zhang, Ravi Ganti, Krisztian Balog, and Filip Radlinski. 2023. Beyond Single Items: Exploring User Preferences in Item Sets with the Conversational Playlist Curation Dataset. In Proceedings of the ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR â23). 2754â2764. | 2307.14225#43 | Large Language Models are Competitive Near Cold-start Recommenders for Language- and Item-based Preferences | Traditional recommender systems leverage users' item preference history to
recommend novel content that users may like. However, modern dialog interfaces
that allow users to express language-based preferences offer a fundamentally
different modality for preference input. Inspired by recent successes of
prompting paradigms for large language models (LLMs), we study their use for
making recommendations from both item-based and language-based preferences in
comparison to state-of-the-art item-based collaborative filtering (CF) methods.
To support this investigation, we collect a new dataset consisting of both
item-based and language-based preferences elicited from users along with their
ratings on a variety of (biased) recommended items and (unbiased) random items.
Among numerous experimental results, we find that LLMs provide competitive
recommendation performance for pure language-based preferences (no item
preferences) in the near cold-start case in comparison to item-based CF
methods, despite having no supervised training for this specific task
(zero-shot) or only a few labels (few-shot). This is particularly promising as
language-based preference representations are more explainable and scrutable
than item-based or vector-based representations. | http://arxiv.org/pdf/2307.14225 | Scott Sanner, Krisztian Balog, Filip Radlinski, Ben Wedin, Lucas Dixon | cs.IR, cs.LG | To appear at RecSys'23 | null | cs.IR | 20230726 | 20230726 | [
{
"id": "2305.06474"
},
{
"id": "2305.07961"
},
{
"id": "2009.13292"
},
{
"id": "2204.02311"
},
{
"id": "2210.06280"
},
{
"id": "2005.14165"
},
{
"id": "2108.07732"
},
{
"id": "2306.02250"
},
{
"id": "2304.04250"
},
{
"id": "2305.08845"
},
{
"id": "2109.01652"
}
] |
2307.14430 | 43 | Data selection methods Many data selection methods have been proposed for supervised, task-speciï¬c settings. In this setting, the most typical objective is dataset condensation, which aims to identify a small subset of data that captures the larger datasetâs properties with respect to the model. Some approaches include constructing coresets [30, 47], identifying samples that the model forgets during training [58]; identifying samples with the largest gradients [46] or gradients that approximate the overall gradient [39]; clustering in embedding space and selecting points farthest from cluster centers [53]; and selecting samples with the highest uncertainty or entropy [33]. These approaches have also been shown to transfer from smaller models to larger models [10]. Unlike these methods, we study how to select data for learning one or many skills at the mixture level for LMs instead of the instance level. | 2307.14430#43 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 44 | Another area of interest is data selection for domain adaptation and multitask learning. For domain adaptation, there are a wide range of methods that select data to best match the target distribution. For example, the Moore-Lewis method matches data based on the difference in cross-entropy using a model trained on the target versus a model trained on the source data [41]. Several other approaches suggest training a model to distinguish between source and target and selecting points with high uncertainty [50], or selecting points based on some divergence in an embedding space [51]. In comparison to these approaches, our work focuses on learning one or many skills and also ï¬nds that embedding-based heuristics do not fully identify skills. | 2307.14430#44 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14225 | 45 | [9] Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, Parker Schuh, Kensen Shi, Sasha Tsvyashchenko, Joshua Maynez, Abhishek Rao, Parker Barnes, Yi Tay, Noam Shazeer, Vinodkumar Prabhakaran, Emily Reif, Nan Du, Ben Hutchinson, Reiner Pope, James Bradbury, Jacob Austin, Michael Isard, Guy Gur-Ari, Pengcheng Yin, Toju Duke, Anselm Levskaya, Sanjay Ghemawat, Sunipa Dev, Henryk Michalewski, Xavier Garcia, Vedant Misra, Kevin Robinson, Liam Fedus, Denny Zhou, Daphne Ippolito, David Luan, Hyeontaek Lim, Barret Zoph, Alexander Spiridonov, Ryan Sepassi, David Dohan, Shivani Agrawal, Mark Omernick, Andrew M. Dai, Thanumalayan Sankaranarayana Pillai, Marie Pellat, Aitor Lewkowycz, Erica | 2307.14225#45 | Large Language Models are Competitive Near Cold-start Recommenders for Language- and Item-based Preferences | Traditional recommender systems leverage users' item preference history to
recommend novel content that users may like. However, modern dialog interfaces
that allow users to express language-based preferences offer a fundamentally
different modality for preference input. Inspired by recent successes of
prompting paradigms for large language models (LLMs), we study their use for
making recommendations from both item-based and language-based preferences in
comparison to state-of-the-art item-based collaborative filtering (CF) methods.
To support this investigation, we collect a new dataset consisting of both
item-based and language-based preferences elicited from users along with their
ratings on a variety of (biased) recommended items and (unbiased) random items.
Among numerous experimental results, we find that LLMs provide competitive
recommendation performance for pure language-based preferences (no item
preferences) in the near cold-start case in comparison to item-based CF
methods, despite having no supervised training for this specific task
(zero-shot) or only a few labels (few-shot). This is particularly promising as
language-based preference representations are more explainable and scrutable
than item-based or vector-based representations. | http://arxiv.org/pdf/2307.14225 | Scott Sanner, Krisztian Balog, Filip Radlinski, Ben Wedin, Lucas Dixon | cs.IR, cs.LG | To appear at RecSys'23 | null | cs.IR | 20230726 | 20230726 | [
{
"id": "2305.06474"
},
{
"id": "2305.07961"
},
{
"id": "2009.13292"
},
{
"id": "2204.02311"
},
{
"id": "2210.06280"
},
{
"id": "2005.14165"
},
{
"id": "2108.07732"
},
{
"id": "2306.02250"
},
{
"id": "2304.04250"
},
{
"id": "2305.08845"
},
{
"id": "2109.01652"
}
] |
2307.14430 | 45 | Data attribution Another perspective on understanding training data is data attribution, which seeks to identify what data is responsible for particular model behaviors. Inï¬uence functions [28] and shapley values [16] are two ways to quantify the role of individual samples. Datamodels [23] ï¬t a model to predict behavior given a subset of training data, providing a framework for understanding individual samples as well as dataset counterfactuals. Simï¬uence [20] ï¬ts a Markov process to a set of training trajectories for ï¬ner-grained understanding of how data impacts training. We focus on understanding how groups of data associated with skills elicit broader model capabilities, and utilize this understanding to select data for more efï¬cient training.
Curriculum learning Curriculum learning [3] proposes to show the model data in order from easy samples to hard ones. Various criteria have been used to determine hardness, and anticurriculum as well as various pacing functions and mixing rates have been explored [54]. Curriculum learning can also be performed at the group level [60]. More sophisticated approaches include parametrizing each sample with a dynamic importance [52], and also accounting for irrelevant and noisy data [38]. Our approach similarly utilizes a curriculum, but it is deï¬ned over a skills graph and does not necessarily align with training on easiest to hardest skills. | 2307.14430#45 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14225 | 46 | Omernick, Andrew M. Dai, Thanumalayan Sankaranarayana Pillai, Marie Pellat, Aitor Lewkowycz, Erica Moreira, Rewon Child, Oleksandr Polozov, Katherine Lee, Zongwei Zhou, Xuezhi Wang, Brennan Saeta, Mark Diaz, Orhan Firat, Michele Catasta, Jason Wei, Kathy Meier-Hellstern, Douglas Eck, Jeï¬ Dean, Slav Petrov, and Noah Fiedel. 2022. PaLM: Scaling Language Modeling with Pathways. arXiv:2204.02311 [cs.CL] | 2307.14225#46 | Large Language Models are Competitive Near Cold-start Recommenders for Language- and Item-based Preferences | Traditional recommender systems leverage users' item preference history to
recommend novel content that users may like. However, modern dialog interfaces
that allow users to express language-based preferences offer a fundamentally
different modality for preference input. Inspired by recent successes of
prompting paradigms for large language models (LLMs), we study their use for
making recommendations from both item-based and language-based preferences in
comparison to state-of-the-art item-based collaborative filtering (CF) methods.
To support this investigation, we collect a new dataset consisting of both
item-based and language-based preferences elicited from users along with their
ratings on a variety of (biased) recommended items and (unbiased) random items.
Among numerous experimental results, we find that LLMs provide competitive
recommendation performance for pure language-based preferences (no item
preferences) in the near cold-start case in comparison to item-based CF
methods, despite having no supervised training for this specific task
(zero-shot) or only a few labels (few-shot). This is particularly promising as
language-based preference representations are more explainable and scrutable
than item-based or vector-based representations. | http://arxiv.org/pdf/2307.14225 | Scott Sanner, Krisztian Balog, Filip Radlinski, Ben Wedin, Lucas Dixon | cs.IR, cs.LG | To appear at RecSys'23 | null | cs.IR | 20230726 | 20230726 | [
{
"id": "2305.06474"
},
{
"id": "2305.07961"
},
{
"id": "2009.13292"
},
{
"id": "2204.02311"
},
{
"id": "2210.06280"
},
{
"id": "2005.14165"
},
{
"id": "2108.07732"
},
{
"id": "2306.02250"
},
{
"id": "2304.04250"
},
{
"id": "2305.08845"
},
{
"id": "2109.01652"
}
] |
2307.14430 | 46 | How LMs learn Many different explanations for how LMs learn from data have been proposed. One hypothesis is that there exist discrete, universal building blocks of LM knowledge called quanta, and power law scaling emerges from a learning over a particular distribution of quanta in the right order [37]. Another is that chain of thought reasoning emerges due to local clusters of latent variables that inï¬uence each other, which can be validated by studying the LMâs ability to do conditional inference given intermediate variables [48]. Others have provided theoretical analysis of how transformers learn topics by studying co-occurrences of words in the training data [34]. Empirically, how models learn is still a mysteryâfor instance, models trained on code are found to perform fairly well at commensense reasoning [36]. Our work initiates a study on how LMs learn various skills and how to exploit this for better data selection.
10 | 2307.14430#46 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14225 | 47 | [10] Konstantina Christakopoulou, Filip Radlinski, and Katja Hofmann. 2016. Towards Conversational Recommender Systems. In Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD â16). 815â824.
[11] Maurizio Ferrari Dacrema, Paolo Cremonesi, and Dietmar Jannach. 2019. Are We Really Making Much Progress? A Worrying Analysis of Recent Neural Recommendation Approaches. In Proceedings of the 13th ACM Conference on Recommender Systems (RecSys â19). 101â109.
[12] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Un- derstanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) (NAACL â19). 4171â4186. | 2307.14225#47 | Large Language Models are Competitive Near Cold-start Recommenders for Language- and Item-based Preferences | Traditional recommender systems leverage users' item preference history to
recommend novel content that users may like. However, modern dialog interfaces
that allow users to express language-based preferences offer a fundamentally
different modality for preference input. Inspired by recent successes of
prompting paradigms for large language models (LLMs), we study their use for
making recommendations from both item-based and language-based preferences in
comparison to state-of-the-art item-based collaborative filtering (CF) methods.
To support this investigation, we collect a new dataset consisting of both
item-based and language-based preferences elicited from users along with their
ratings on a variety of (biased) recommended items and (unbiased) random items.
Among numerous experimental results, we find that LLMs provide competitive
recommendation performance for pure language-based preferences (no item
preferences) in the near cold-start case in comparison to item-based CF
methods, despite having no supervised training for this specific task
(zero-shot) or only a few labels (few-shot). This is particularly promising as
language-based preference representations are more explainable and scrutable
than item-based or vector-based representations. | http://arxiv.org/pdf/2307.14225 | Scott Sanner, Krisztian Balog, Filip Radlinski, Ben Wedin, Lucas Dixon | cs.IR, cs.LG | To appear at RecSys'23 | null | cs.IR | 20230726 | 20230726 | [
{
"id": "2305.06474"
},
{
"id": "2305.07961"
},
{
"id": "2009.13292"
},
{
"id": "2204.02311"
},
{
"id": "2210.06280"
},
{
"id": "2005.14165"
},
{
"id": "2108.07732"
},
{
"id": "2306.02250"
},
{
"id": "2304.04250"
},
{
"id": "2305.08845"
},
{
"id": "2109.01652"
}
] |
2307.14430 | 47 | 10
Task selection In multitask auxiliary learning, the goal is to train a model to perform well on a target task(s) by selecting the most beneï¬cial source tasks to train on. One can use feature similarity to select tasks [29], but we ï¬nd in our synthetics that feature similarity does not always recover skills. In Taskonomy [70], a hypergraph over a set of tasks is learned and used to select tasks. The methods used to develop the taxonomy can be applied to further expand our graph learning (e.g., studying transitive and higher-order properties). However, their focus is on task selection in computer vision rather than data selection for LMs to learn skills. Lastly, the contemporary work of TaskWeb [24] builds a graph among 22 common NLP tasks in order to determine what the best source tasks are for a target task. Their deï¬nition of an edge in the task graph is less strict than ours (their comparison is on if training on additional data from si helps with sj, while we ï¬x the overall amount of training data over both si and sj). Overall, our approach is similar in use of the skills graph, but we incorporate it into a dynamic sampling algorithm. Furthermore, we look more broadly at skills, rather than tasks, and characterize when we expect using the skills graph to improve model performance. | 2307.14430#47 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14225 | 48 | [13] Luke Friedman, Sameer Ahuja, David Allen, Zhenning Tan, Hakim Sidahmed, Changbo Long, Jun Xie, Gabriel Schubiner, Ajay Patel, Harsh Leveraging Large Language Models in Conversational Recommender Systems. Lara, Brian Chu, Zexi Chen, and Manoj Tiwari. 2023. arXiv:2305.07961 [cs.IR]
[14] Zeno Gantner, Steï¬en Rendle, Christoph Freudenthaler, and Lars Schmidt-Thieme. 2011. MyMediaLite: A Free Recommender System Library. In Proceedings of the Fifth ACM Conference on Recommender Systems (RecSys â11). 305â308.
[15] Chongming Gao, Wenqiang Lei, Xiangnan He, Maarten de Rijke, and Tat-Seng Chua. 2021. Advances and Challenges in Conversational Recom- mender Systems: A Survey. AI Open 2 (2021), 100â126. | 2307.14225#48 | Large Language Models are Competitive Near Cold-start Recommenders for Language- and Item-based Preferences | Traditional recommender systems leverage users' item preference history to
recommend novel content that users may like. However, modern dialog interfaces
that allow users to express language-based preferences offer a fundamentally
different modality for preference input. Inspired by recent successes of
prompting paradigms for large language models (LLMs), we study their use for
making recommendations from both item-based and language-based preferences in
comparison to state-of-the-art item-based collaborative filtering (CF) methods.
To support this investigation, we collect a new dataset consisting of both
item-based and language-based preferences elicited from users along with their
ratings on a variety of (biased) recommended items and (unbiased) random items.
Among numerous experimental results, we find that LLMs provide competitive
recommendation performance for pure language-based preferences (no item
preferences) in the near cold-start case in comparison to item-based CF
methods, despite having no supervised training for this specific task
(zero-shot) or only a few labels (few-shot). This is particularly promising as
language-based preference representations are more explainable and scrutable
than item-based or vector-based representations. | http://arxiv.org/pdf/2307.14225 | Scott Sanner, Krisztian Balog, Filip Radlinski, Ben Wedin, Lucas Dixon | cs.IR, cs.LG | To appear at RecSys'23 | null | cs.IR | 20230726 | 20230726 | [
{
"id": "2305.06474"
},
{
"id": "2305.07961"
},
{
"id": "2009.13292"
},
{
"id": "2204.02311"
},
{
"id": "2210.06280"
},
{
"id": "2005.14165"
},
{
"id": "2108.07732"
},
{
"id": "2306.02250"
},
{
"id": "2304.04250"
},
{
"id": "2305.08845"
},
{
"id": "2109.01652"
}
] |
2307.14430 | 48 | Education The notion of skill has been studied in education. Classical research on learning hierarchies [66] identify sets of skills that make up subordinate capabilities for students. For instance, [12] identiï¬ed that in order for students to solve linear equations, there were many prerequisite skills, ranging from the simplest being symbol recognition to the most complex being the ability to add, subtract, multiple, and divide from both sides of the equation. More recently, decision-making over lesson sequences based on skills, e.g., what the student already knows versus what the lesson teaches, has become an area of interest in personalized learning [49].
# 6 Conclusion
Given a ï¬xed budget of data, knowing what data to train on to induce various capabilities in an LM is challenging. As LMs continue to improve, it will become increasingly important to extract as much signal as possible from the data and to direct that signal towards acquiring a broad variety of capabilities. In this paper, we introduce a skills-based framework for understanding how LMs learn and for selecting training data. We hope our study invites others to build on such a notion of skill and further explore how to align skills with data.
# Acknowledgements | 2307.14430#48 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14225 | 49 | [16] Shijie Geng, Shuchang Liu, Zuohui Fu, Yingqiang Ge, and Yongfeng Zhang. 2022. Recommendation as Language Processing (RLP): A Uniï¬ed Pretrain, Personalized Prompt & Predict Paradigm (P5). In Proceedings of the 16th ACM Conference on Recommender Systems (RecSys â22). 299â315. [17] Deepesh V Hada and Shirish K Shevade. 2021. ReXPlug: Explainable Recommendation using Plug-and-Play Language Model. In Proceedings of the
44th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR â21). 81â91.
[18] F. Maxwell Harper and Joseph A. Konstan. 2015. The MovieLens Datasets: History and Context. ACM Transactions on Interactive Intelligent Systems 5, 4, Article 19 (2015).
[19] Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia Hu, and Tat-Seng Chua. 2017. Neural Collaborative Filtering. In Proceedings of the 26th International Conference on World Wide Web (WWW â17). 173â182.
10
LLMs are Competitive Near Cold-start Recommenders
RecSys â23, September 18â22, 2023, Singapore, Singapore | 2307.14225#49 | Large Language Models are Competitive Near Cold-start Recommenders for Language- and Item-based Preferences | Traditional recommender systems leverage users' item preference history to
recommend novel content that users may like. However, modern dialog interfaces
that allow users to express language-based preferences offer a fundamentally
different modality for preference input. Inspired by recent successes of
prompting paradigms for large language models (LLMs), we study their use for
making recommendations from both item-based and language-based preferences in
comparison to state-of-the-art item-based collaborative filtering (CF) methods.
To support this investigation, we collect a new dataset consisting of both
item-based and language-based preferences elicited from users along with their
ratings on a variety of (biased) recommended items and (unbiased) random items.
Among numerous experimental results, we find that LLMs provide competitive
recommendation performance for pure language-based preferences (no item
preferences) in the near cold-start case in comparison to item-based CF
methods, despite having no supervised training for this specific task
(zero-shot) or only a few labels (few-shot). This is particularly promising as
language-based preference representations are more explainable and scrutable
than item-based or vector-based representations. | http://arxiv.org/pdf/2307.14225 | Scott Sanner, Krisztian Balog, Filip Radlinski, Ben Wedin, Lucas Dixon | cs.IR, cs.LG | To appear at RecSys'23 | null | cs.IR | 20230726 | 20230726 | [
{
"id": "2305.06474"
},
{
"id": "2305.07961"
},
{
"id": "2009.13292"
},
{
"id": "2204.02311"
},
{
"id": "2210.06280"
},
{
"id": "2005.14165"
},
{
"id": "2108.07732"
},
{
"id": "2306.02250"
},
{
"id": "2304.04250"
},
{
"id": "2305.08845"
},
{
"id": "2109.01652"
}
] |
2307.14225 | 50 | 10
LLMs are Competitive Near Cold-start Recommenders
RecSys â23, September 18â22, 2023, Singapore, Singapore
[20] Yupeng Hou, Shanlei Mu, Wayne Xin Zhao, Yaliang Li, Bolin Ding, and Ji-Rong Wen. 2022. Towards Universal Sequence Representation Learning for Recommender Systems. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD â22). 585â593. [21] Yupeng Hou, Junjie Zhang, Zihan Lin, Hongyu Lu, Ruobing Xie, Julian McAuley, and Wayne Xin Zhao. 2023. Large Language Models are Zero-Shot
Rankers for Recommender Systems. arXiv:2305.08845 [cs.IR]
[22] Fangwei Hu and Yong Yu. 2013. Interview Process Learning for Top-N Recommendation. In Proceedings of the ACM Conference on Recommender Systems (RecSys â13). 331â334.
[23] Yifan Hu, Yehuda Koren, and Chris Volinsky. 2008. Collaborative Filtering for Implicit Feedback Datasets. In Proceedings of the 2008 Eighth IEEE International Conference on Data Mining (ICDM â08). 263â272. | 2307.14225#50 | Large Language Models are Competitive Near Cold-start Recommenders for Language- and Item-based Preferences | Traditional recommender systems leverage users' item preference history to
recommend novel content that users may like. However, modern dialog interfaces
that allow users to express language-based preferences offer a fundamentally
different modality for preference input. Inspired by recent successes of
prompting paradigms for large language models (LLMs), we study their use for
making recommendations from both item-based and language-based preferences in
comparison to state-of-the-art item-based collaborative filtering (CF) methods.
To support this investigation, we collect a new dataset consisting of both
item-based and language-based preferences elicited from users along with their
ratings on a variety of (biased) recommended items and (unbiased) random items.
Among numerous experimental results, we find that LLMs provide competitive
recommendation performance for pure language-based preferences (no item
preferences) in the near cold-start case in comparison to item-based CF
methods, despite having no supervised training for this specific task
(zero-shot) or only a few labels (few-shot). This is particularly promising as
language-based preference representations are more explainable and scrutable
than item-based or vector-based representations. | http://arxiv.org/pdf/2307.14225 | Scott Sanner, Krisztian Balog, Filip Radlinski, Ben Wedin, Lucas Dixon | cs.IR, cs.LG | To appear at RecSys'23 | null | cs.IR | 20230726 | 20230726 | [
{
"id": "2305.06474"
},
{
"id": "2305.07961"
},
{
"id": "2009.13292"
},
{
"id": "2204.02311"
},
{
"id": "2210.06280"
},
{
"id": "2005.14165"
},
{
"id": "2108.07732"
},
{
"id": "2306.02250"
},
{
"id": "2304.04250"
},
{
"id": "2305.08845"
},
{
"id": "2109.01652"
}
] |
2307.14430 | 50 | We gratefully acknowledge the support of NIH under No. U54EB020405 (Mobilize), NSF under Nos. CCF1763315 (Beyond Sparsity), CCF1563078 (Volume to Velocity), and 1937301 (RTML); US DEVCOM ARL under No. W911NF-21- 2-0251 (Interactive Human-AI Teaming); ONR under No. N000141712266 (Unifying Weak Supervision); ONR N00014-20- 1-2480: Understanding and Applying Non-Euclidean Geometry in Machine Learning; N000142012275 (NEPTUNE); NXP, Xilinx, LETI-CEA, Intel, IBM, Microsoft, NEC, Toshiba, TSMC, ARM, Hitachi, BASF, Accenture, Ericsson, Qualcomm, Analog Devices, Google Cloud, Salesforce, Total, the HAI-GCP Cloud Credits for Research program, the Stanford Data Science Initiative (SDSI), and members of the Stanford DAWN project: Facebook, Google, and VMWare. FS is supported by NSF CCF2106707 and the Wisconsin Alumni Research Foundation (WARF). The U.S. Government is authorized to reproduce and distribute reprints for Governmental purposes notwithstanding any copyright | 2307.14430#50 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14225 | 51 | [24] Dietmar Jannach, Ahtsham Manzoor, Wanling Cai, and Li Chen. 2021. A Survey on Conversational Recommender Systems. Comput. Surveys 54, 5 (2021).
[25] Marius Kaminskas and Derek Bridge. 2016. Diversity, Serendipity, Novelty, and Coverage: A Survey and Empirical Analysis of Beyond-Accuracy Objectives in Recommender Systems. ACM Transactions on Interactive Intelligent Systems 7, 1 (2016), 1â42.
[26] Jie Kang, Kyle Condiï¬, Shuo Chang, Joseph A. Konstan, Loren Terveen, and F. Maxwell Harper. 2017. Understanding How People Use Natural Language to Ask for Recommendations. In Proceedings of the Eleventh ACM Conference on Recommender Systems (RecSys â17). 229â237.
[27] Wang-Cheng Kang, Jianmo Ni, Nikhil Mehta, Maheswaran Sathiamoorthy, Lichan Hong, Ed Chi, and Derek Zhiyuan Cheng. 2023. Do LLMs Understand User Preferences? Evaluating LLMs On User Rating Prediction. arXiv:2305.06474 [cs.IR] | 2307.14225#51 | Large Language Models are Competitive Near Cold-start Recommenders for Language- and Item-based Preferences | Traditional recommender systems leverage users' item preference history to
recommend novel content that users may like. However, modern dialog interfaces
that allow users to express language-based preferences offer a fundamentally
different modality for preference input. Inspired by recent successes of
prompting paradigms for large language models (LLMs), we study their use for
making recommendations from both item-based and language-based preferences in
comparison to state-of-the-art item-based collaborative filtering (CF) methods.
To support this investigation, we collect a new dataset consisting of both
item-based and language-based preferences elicited from users along with their
ratings on a variety of (biased) recommended items and (unbiased) random items.
Among numerous experimental results, we find that LLMs provide competitive
recommendation performance for pure language-based preferences (no item
preferences) in the near cold-start case in comparison to item-based CF
methods, despite having no supervised training for this specific task
(zero-shot) or only a few labels (few-shot). This is particularly promising as
language-based preference representations are more explainable and scrutable
than item-based or vector-based representations. | http://arxiv.org/pdf/2307.14225 | Scott Sanner, Krisztian Balog, Filip Radlinski, Ben Wedin, Lucas Dixon | cs.IR, cs.LG | To appear at RecSys'23 | null | cs.IR | 20230726 | 20230726 | [
{
"id": "2305.06474"
},
{
"id": "2305.07961"
},
{
"id": "2009.13292"
},
{
"id": "2204.02311"
},
{
"id": "2210.06280"
},
{
"id": "2005.14165"
},
{
"id": "2108.07732"
},
{
"id": "2306.02250"
},
{
"id": "2304.04250"
},
{
"id": "2305.08845"
},
{
"id": "2109.01652"
}
] |
2307.14430 | 51 | and the Wisconsin Alumni Research Foundation (WARF). The U.S. Government is authorized to reproduce and distribute reprints for Governmental purposes notwithstanding any copyright notation thereon. Any opinions, ï¬ndings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reï¬ect the views, policies, or endorsements, either expressed or implied, of NIH, ONR, or the U.S. Government. | 2307.14430#51 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14225 | 52 | [28] Dawen Liang, Rahul G. Krishnan, Matthew D. Hoï¬man, and Tony Jebara. 2018. Variational Autoencoders for Collaborative Filtering. In Proceedings of the 2018 World Wide Web Conference (WWW â18). 689â698.
[29] Pasquale Lops, Marco De Gemmis, and Giovanni Semeraro. 2011. Content-based Recommender Systems: State of the Art and Trends. In Recom- mender Systems Handbook. Springer, 73â105.
[30] Itzik Malkiel, Oren Barkan, Avi Caciularu, Noam Razin, Ori Katz, and Noam Koenigstein. 2020. RecoBERT: A Catalog Language Model for Text- Based Recommendations. arXiv:2009.13292 [cs.IR]
[31] Sheshera Mysore, Mahmood Jasim, Andrew McCallum, and Hamed Zamani. 2023. Editable User Proï¬les for Controllable Text Recommendation. arXiv:2304.04250 [cs.IR]
[32] Sheshera Mysore, Andrew McCallum, and Hamed Zamani. 2023. Large Language Model Augmented Narrative Driven Recommendations. arXiv:2306.02250 [cs.IR] | 2307.14225#52 | Large Language Models are Competitive Near Cold-start Recommenders for Language- and Item-based Preferences | Traditional recommender systems leverage users' item preference history to
recommend novel content that users may like. However, modern dialog interfaces
that allow users to express language-based preferences offer a fundamentally
different modality for preference input. Inspired by recent successes of
prompting paradigms for large language models (LLMs), we study their use for
making recommendations from both item-based and language-based preferences in
comparison to state-of-the-art item-based collaborative filtering (CF) methods.
To support this investigation, we collect a new dataset consisting of both
item-based and language-based preferences elicited from users along with their
ratings on a variety of (biased) recommended items and (unbiased) random items.
Among numerous experimental results, we find that LLMs provide competitive
recommendation performance for pure language-based preferences (no item
preferences) in the near cold-start case in comparison to item-based CF
methods, despite having no supervised training for this specific task
(zero-shot) or only a few labels (few-shot). This is particularly promising as
language-based preference representations are more explainable and scrutable
than item-based or vector-based representations. | http://arxiv.org/pdf/2307.14225 | Scott Sanner, Krisztian Balog, Filip Radlinski, Ben Wedin, Lucas Dixon | cs.IR, cs.LG | To appear at RecSys'23 | null | cs.IR | 20230726 | 20230726 | [
{
"id": "2305.06474"
},
{
"id": "2305.07961"
},
{
"id": "2009.13292"
},
{
"id": "2204.02311"
},
{
"id": "2210.06280"
},
{
"id": "2005.14165"
},
{
"id": "2108.07732"
},
{
"id": "2306.02250"
},
{
"id": "2304.04250"
},
{
"id": "2305.08845"
},
{
"id": "2109.01652"
}
] |
2307.14430 | 52 | 11
References [1] Amro Abbas, Kushal Tirumala, Dániel Simig, Surya Ganguli, and Ari S. Morcos. Semdedup: Data-efï¬cient learning
at web-scale through semantic deduplication, 2023.
[2] Yuntao Bai, Andy Jones, et al. Training a helpful and harmless assistant with reinforcement learning from human feedback, 2022.
[3] Yoshua Bengio, Jérôme Louradour, Ronan Collobert, and Jason Weston. Curriculum learning. In Proceedings of the 26th annual international conference on machine learning, pages 41â48, 2009.
[4] Stella Biderman, Hailey Schoelkopf, Quentin Anthony, Herbie Bradley, Kyle OâBrien, Eric Hallahan, Mohammad Aï¬ah Khan, Shivanshu Purohit, USVSN Sai Prashanth, Edward Raff, Aviya Skowron, Lintang Sutawika, and Oskar van der Wal. Pythia: A suite for analyzing large language models across training and scaling, 2023. | 2307.14430#52 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14225 | 53 | [33] Zahra Nazari, Praveen Chandar, Ghazal Fazelnia, Catherine M. Edwards, Benjamin Carterette, and Mounia Lalmas. 2022. Choice of Implicit Signal Matters: Accounting for User Aspirations in Podcast Recommendations. In Proceedings of the ACM Web Conference 2022 (WWW â22). 2433â2441.
[34] Xia Ning and George Karypis. 2011. SLIM: Sparse Linear Methods for Top-N Recommender Systems. In Proceedings of the 2011 IEEE 11th Interna- tional Conference on Data Mining (ICDM â11). 497â506.
[35] Roberto Pellegrini, Wenjie Zhao, and Iain Murray. 2022. Donât Recommend the Obvious: Estimate Probability Ratios. In Proceedings of the 16th ACM Conference on Recommender Systems (RecSys â22). 188â197.
[36] Gustavo Penha and Claudia Hauï¬. 2020. What does BERT know about books, movies and music? Probing BERT for Conversational Recommenda- tion. In Fourteenth ACM Conference on Recommender Systems (RecSys â20). 388â397. | 2307.14225#53 | Large Language Models are Competitive Near Cold-start Recommenders for Language- and Item-based Preferences | Traditional recommender systems leverage users' item preference history to
recommend novel content that users may like. However, modern dialog interfaces
that allow users to express language-based preferences offer a fundamentally
different modality for preference input. Inspired by recent successes of
prompting paradigms for large language models (LLMs), we study their use for
making recommendations from both item-based and language-based preferences in
comparison to state-of-the-art item-based collaborative filtering (CF) methods.
To support this investigation, we collect a new dataset consisting of both
item-based and language-based preferences elicited from users along with their
ratings on a variety of (biased) recommended items and (unbiased) random items.
Among numerous experimental results, we find that LLMs provide competitive
recommendation performance for pure language-based preferences (no item
preferences) in the near cold-start case in comparison to item-based CF
methods, despite having no supervised training for this specific task
(zero-shot) or only a few labels (few-shot). This is particularly promising as
language-based preference representations are more explainable and scrutable
than item-based or vector-based representations. | http://arxiv.org/pdf/2307.14225 | Scott Sanner, Krisztian Balog, Filip Radlinski, Ben Wedin, Lucas Dixon | cs.IR, cs.LG | To appear at RecSys'23 | null | cs.IR | 20230726 | 20230726 | [
{
"id": "2305.06474"
},
{
"id": "2305.07961"
},
{
"id": "2009.13292"
},
{
"id": "2204.02311"
},
{
"id": "2210.06280"
},
{
"id": "2005.14165"
},
{
"id": "2108.07732"
},
{
"id": "2306.02250"
},
{
"id": "2304.04250"
},
{
"id": "2305.08845"
},
{
"id": "2109.01652"
}
] |
2307.14430 | 53 | [5] Sid Black, Gao Leo, Phil Wang, Connor Leahy, and Stella Biderman. GPT-Neo: Large Scale Autoregressive Language Modeling with Mesh-Tensorï¬ow, March 2021. If you use this software, please cite it using these metadata.
[6] Rishi Bommasani, Percy Liang, et al. On the opportunities and risks of foundation models, 2021.
[7] Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakan- tan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. arXiv preprint arXiv:2005.14165, 2020.
[8] Mark Chen, Jerry Tworek, et al. Evaluating large language models trained on code, 2021.
[9] Wei-Lin Chiang, Zhuohan Li, Zi Lin, Ying Sheng, Zhanghao Wu, Hao Zhang, Lianmin Zheng, Siyuan Zhuang, Yonghao Zhuang, Joseph E. Gonzalez, Ion Stoica, and Eric P. Xing. Vicuna: An open-source chatbot impressing gpt-4 with 90%* chatgpt quality, March 2023. | 2307.14430#53 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14225 | 54 | [37] Filip Radlinski, Krisztian Balog, Fernando Diaz, Lucas Dixon, and Ben Wedin. 2022. On Natural Language User Proï¬les for Transparent and Scrutable Recommendation. In Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR â22). 2863â2874.
[38] Steï¬en Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt-Thieme. 2009. BPR: Bayesian Personalized Ranking from Implicit Feed- back. In Proceedings of the Twenty-Fifth Conference on Uncertainty in Artiï¬cial Intelligence (UAI â09). 452â461.
[39] Lior Rokach and Slava Kisilevich. 2012. Initial Proï¬le Generation in Recommender Systems Using Pairwise Comparison. IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews) 42, 6 (2012), 1854â1859.
[40] Badrul Sarwar, George Karypis, Joseph Konstan, and John Riedl. 2001. Item-based Collaborative Filtering Recommendation Algorithms. In Proceed- ings of the 10th International Conference on World Wide Web (WWW â01). 285â295. | 2307.14225#54 | Large Language Models are Competitive Near Cold-start Recommenders for Language- and Item-based Preferences | Traditional recommender systems leverage users' item preference history to
recommend novel content that users may like. However, modern dialog interfaces
that allow users to express language-based preferences offer a fundamentally
different modality for preference input. Inspired by recent successes of
prompting paradigms for large language models (LLMs), we study their use for
making recommendations from both item-based and language-based preferences in
comparison to state-of-the-art item-based collaborative filtering (CF) methods.
To support this investigation, we collect a new dataset consisting of both
item-based and language-based preferences elicited from users along with their
ratings on a variety of (biased) recommended items and (unbiased) random items.
Among numerous experimental results, we find that LLMs provide competitive
recommendation performance for pure language-based preferences (no item
preferences) in the near cold-start case in comparison to item-based CF
methods, despite having no supervised training for this specific task
(zero-shot) or only a few labels (few-shot). This is particularly promising as
language-based preference representations are more explainable and scrutable
than item-based or vector-based representations. | http://arxiv.org/pdf/2307.14225 | Scott Sanner, Krisztian Balog, Filip Radlinski, Ben Wedin, Lucas Dixon | cs.IR, cs.LG | To appear at RecSys'23 | null | cs.IR | 20230726 | 20230726 | [
{
"id": "2305.06474"
},
{
"id": "2305.07961"
},
{
"id": "2009.13292"
},
{
"id": "2204.02311"
},
{
"id": "2210.06280"
},
{
"id": "2005.14165"
},
{
"id": "2108.07732"
},
{
"id": "2306.02250"
},
{
"id": "2304.04250"
},
{
"id": "2305.08845"
},
{
"id": "2109.01652"
}
] |
2307.14430 | 54 | [10] Cody Coleman, Christopher Yeh, Stephen Mussmann, Baharan Mirzasoleiman, Peter Bailis, Percy Liang, Jure Leskovec, and Matei Zaharia. Selection via proxy: Efï¬cient data selection for deep learning, 2019.
[11] Robert M Gagne. The acquisition of knowledge. Psychological review, 69(4):355, 1962.
[12] Robert M Gagne and Noel E Paradise. Abilities and learning sets in knowledge acquisition. Psychological Monographs: General and Applied, 75(14):1, 1961.
[13] Leo Gao, Stella Biderman, Sid Black, Laurence Golding, Travis Hoppe, Charles Foster, Jason Phang, Horace He, Anish Thite, Noa Nabeshima, et al. The pile: An 800gb dataset of diverse text for language modeling. arXiv preprint arXiv:2101.00027, 2020.
[14] Leo Gao, Jonathan Tow, Stella Biderman, Sid Black, Anthony DiPoï¬, Charles Foster, Laurence Golding, Jeffrey Hsu, Kyle McDonell, Niklas Muennighoff, Jason Phang, Laria Reynolds, Eric Tang, Anish Thite, Ben Wang, Kevin Wang, and Andy Zou. A framework for few-shot language model evaluation, September 2021. | 2307.14430#54 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14225 | 55 | [41] Anna Sepliarskaia, Julia Kiseleva, Filip Radlinski, and Maarten de Rijke. 2018. Preference Elicitation as an Optimization Problem. In Proceedings of the ACM Conference on Recommender Systems (RecSys â18). 172â180.
[42] Harald Steck. 2019. Embarrassingly Shallow Autoencoders for Sparse Data. In The World Wide Web Conference (WWW â19). 3251â3257. [43] Bas Verplanken and Suzanne Faes. 1999. Good Intentions, Bad Habits, and Eï¬ects of Forming Implementation Intentions on Healthy Eating.
European Journal of Social Psychology 29, 5-6 (1999), 591â604.
[44] Jason Wei, Maarten Bosma, Vincent Y. Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M. Dai, and Quoc V. Le. 2022. Finetuned Language Models Are Zero-Shot Learners. arXiv:2109.01652 [cs.CL] | 2307.14225#55 | Large Language Models are Competitive Near Cold-start Recommenders for Language- and Item-based Preferences | Traditional recommender systems leverage users' item preference history to
recommend novel content that users may like. However, modern dialog interfaces
that allow users to express language-based preferences offer a fundamentally
different modality for preference input. Inspired by recent successes of
prompting paradigms for large language models (LLMs), we study their use for
making recommendations from both item-based and language-based preferences in
comparison to state-of-the-art item-based collaborative filtering (CF) methods.
To support this investigation, we collect a new dataset consisting of both
item-based and language-based preferences elicited from users along with their
ratings on a variety of (biased) recommended items and (unbiased) random items.
Among numerous experimental results, we find that LLMs provide competitive
recommendation performance for pure language-based preferences (no item
preferences) in the near cold-start case in comparison to item-based CF
methods, despite having no supervised training for this specific task
(zero-shot) or only a few labels (few-shot). This is particularly promising as
language-based preference representations are more explainable and scrutable
than item-based or vector-based representations. | http://arxiv.org/pdf/2307.14225 | Scott Sanner, Krisztian Balog, Filip Radlinski, Ben Wedin, Lucas Dixon | cs.IR, cs.LG | To appear at RecSys'23 | null | cs.IR | 20230726 | 20230726 | [
{
"id": "2305.06474"
},
{
"id": "2305.07961"
},
{
"id": "2009.13292"
},
{
"id": "2204.02311"
},
{
"id": "2210.06280"
},
{
"id": "2005.14165"
},
{
"id": "2108.07732"
},
{
"id": "2306.02250"
},
{
"id": "2304.04250"
},
{
"id": "2305.08845"
},
{
"id": "2109.01652"
}
] |
2307.14430 | 55 | [15] Xinyang Geng, Arnav Gudibande, Hao Liu, Eric Wallace, Pieter Abbeel, Sergey Levine, and Dawn Song. Koala: A dialogue model for academic research. Blog post, April 2023.
[16] Amirata Ghorbani and James Zou. Data shapley: Equitable valuation of data for machine learning. In International Conference on Machine Learning, pages 2242â2251. PMLR, 2019.
[17] Google. Palm2 technical report. Technical report, 2023.
[18] Anupam Gupta. Advanced algorithms: Notes for cmu 15-850 (fall 2020), 2020.
[19] Suchin Gururangan, Ana Marasovi´c, Swabha Swayamdipta, Kyle Lo, Iz Beltagy, Doug Downey, and Noah A. Smith. Donât stop pretraining: Adapt language models to domains and tasks. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 2020.
[20] Kelvin Guu, Albert Webson, Ellie Pavlick, Lucas Dixon, Ian Tenney, and Tolga Bolukbasi. Simï¬uence: Modeling the inï¬uence of individual training examples by simulating training runs, 2023. | 2307.14430#55 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 56 | [21] Peter Henderson*, Mark S. Krass*, Lucia Zheng, Neel Guha, Christopher D. Manning, Dan Jurafsky, and Daniel E. Ho. Pile of law: Learning responsible data ï¬ltering from the law and a 256gb open-source legal dataset, 2022.
12
[22] Danny Hernandez, Tom Brown, Tom Conerly, Nova DasSarma, Dawn Drain, Sheer El-Showk, Nelson Elhage, Zac Hatï¬eld-Dodds, Tom Henighan, Tristan Hume, Scott Johnston, Ben Mann, Chris Olah, Catherine Olsson, Dario Amodei, Nicholas Joseph, Jared Kaplan, and Sam McCandlish. Scaling laws and interpretability of learning from repeated data, 2022.
[23] Andrew Ilyas, Sung Min Park, Logan Engstrom, Guillaume Leclerc, and Aleksander Madry. Datamodels: Predicting predictions from training data, 2022.
[24] Joongwon Kim, Akari Asai, Gabriel Ilharco, and Hannaneh Hajishirzi. Taskweb: Selecting better source tasks for multi-task nlp, 2023. | 2307.14430#56 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 57 | [25] Hannah Rose Kirk, Yennie Jun, Filippo Volpin, Haider Iqbal, Elias Benussi, Frederic Dreyer, Aleksandar Shtedritski, and Yuki Asano. Bias out-of-the-box: An empirical analysis of intersectional occupational biases in popular generative language models. In M. Ranzato, A. Beygelzimer, Y. Dauphin, P.S. Liang, and J. Wortman Vaughan, editors, Advances in Neural Information Processing Systems, volume 34, pages 2611â2624. Curran Associates, Inc., 2021.
[26] Nikita Kitaev, Steven Cao, and Dan Klein. Multilingual constituency parsing with self-attention and pre-training. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3499â3505, Florence, Italy, July 2019. Association for Computational Linguistics.
[27] Nikita Kitaev and Dan Klein. Constituency parsing with a self-attentive encoder. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 2676â2686, Melbourne, Australia, July 2018. Association for Computational Linguistics. | 2307.14430#57 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 58 | [28] Pang Wei Koh and Percy Liang. Understanding black-box predictions via inï¬uence functions, 2017.
[29] Po-Nien Kung, Sheng-Siang Yin, Yi-Cheng Chen, Tse-Hsuan Yang, and Yun-Nung Chen. Efï¬cient multi-task auxiliary learning: Selecting auxiliary data by feature similarity. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 416â428, Online and Punta Cana, Dominican Republic, November 2021. Association for Computational Linguistics.
[30] Michael Langberg and Leonard J. Schulman. Universal approximators for integrals, pages 598â607.
[31] Hugo Laurençon, Lucile Saulnier, et al. The bigscience roots corpus: A 1.6tb composite multilingual dataset, 2023.
[32] Katherine Lee, Daphne Ippolito, Andrew Nystrom, Chiyuan Zhang, Douglas Eck, Chris Callison-Burch, and Nicholas Carlini. Deduplicating training data makes language models better. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2022. | 2307.14430#58 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 59 | [33] David D Lewis. A sequential algorithm for training text classiï¬ers: Corrigendum and additional data. In Acm Sigir Forum, volume 29, pages 13â19. ACM New York, NY, USA, 1995.
[34] Yuchen Li, Yuanzhi Li, and Andrej Risteski. How do transformers learn topic structure: Towards a mechanistic understanding, 2023.
[35] Paul Pu Liang, Chiyu Wu, Louis-Philippe Morency, and Ruslan Salakhutdinov. Towards understanding and mitigating social biases in language models. In Marina Meila and Tong Zhang, editors, Proceedings of the 38th International Conference on Machine Learning, volume 139 of Proceedings of Machine Learning Research, pages 6565â6576. PMLR, 18â24 Jul 2021.
[36] Aman Madaan, Shuyan Zhou, Uri Alon, Yiming Yang, and Graham Neubig. Language models of code are few-shot commonsense learners, 2022.
[37] Eric J. Michaud, Ziming Liu, Uzay Girit, and Max Tegmark. The quantization model of neural scaling, 2023. | 2307.14430#59 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 60 | [37] Eric J. Michaud, Ziming Liu, Uzay Girit, and Max Tegmark. The quantization model of neural scaling, 2023.
[38] Sören Mindermann, Muhammed Razzak, Winnie Xu, Andreas Kirsch, Mrinank Sharma, Adrien Morisot, Aidan N. Gomez, Sebastian Farquhar, Jan Brauner, and Yarin Gal. Prioritized training on points that are learnable, worth learning, and not yet learned (workshop version), 2021.
[39] Baharan Mirzasoleiman, Jeff Bilmes, and Jure Leskovec. Coresets for data-efï¬cient training of machine learning models, 2019.
[40] Swaroop Mishra, Daniel Khashabi, Chitta Baral, and Hannaneh Hajishirzi. Cross-task generalization via natural language crowdsourcing instructions. In ACL, 2022.
13
[41] Robert C. Moore and William Lewis. Intelligent selection of language model training data. In Proceedings of the ACL 2010 Conference Short Papers, pages 220â224, Uppsala, Sweden, July 2010. Association for Computational Linguistics.
[42] MosaicML. Introducing mpt-7b: A new standard for open-source, commercially usable llms, 2023. | 2307.14430#60 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 61 | [42] MosaicML. Introducing mpt-7b: A new standard for open-source, commercially usable llms, 2023.
[43] Moin Nadeem, Anna Bethke, and Siva Reddy. Stereoset: Measuring stereotypical bias in pretrained language models. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), 2021.
[44] Neel Nanda, Lawrence Chan, Tom Lieberum, Jess Smith, and Jacob Steinhardt. Progress measures for grokking via mechanistic interpretability. In The Eleventh International Conference on Learning Representations, 2023.
[45] Arkadij SemenoviËc Nemirovskij and David Borisovich Yudin. Problem complexity and method efï¬ciency in optimiza- tion. 1983.
[46] Mansheej Paul, Surya Ganguli, and Gintare Karolina Dziugaite. Deep learning on a data diet: Finding important examples early in training, 2021.
[47] Jeff M. Phillips. Coresets and sketches, 2016.
[48] Ben Prystawski and Noah D. Goodman. Why think step-by-step? reasoning emerges from the locality of experience, 2023. | 2307.14430#61 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 62 | [48] Ben Prystawski and Noah D. Goodman. Why think step-by-step? reasoning emerges from the locality of experience, 2023.
[49] Siddharth Reddy, Igor Labutov, and Thorsten Joachims. Latent skill embedding for personalized lesson sequence recommendation, 2016.
[50] Sebastian Ruder, Parsa Ghaffari, and John G. Breslin. Data selection strategies for multi-domain sentiment analysis, 2017.
[51] Sebastian Ruder and Barbara Plank. Learning to select data for transfer learning with bayesian optimization. Proceed- ings of the 2017 Conference on Empirical Methods in Natural Language Processing, 2017.
[52] Shreyas Saxena, Oncel Tuzel, and Dennis DeCoste. Data parameters: A new family of parameters for learning a differentiable curriculum. In H. Wallach, H. Larochelle, A. Beygelzimer, F. d'Alché-Buc, E. Fox, and R. Garnett, editors, Advances in Neural Information Processing Systems, volume 32. Curran Associates, Inc., 2019.
[53] Ben Sorscher, Robert Geirhos, Shashank Shekhar, Surya Ganguli, and Ari S. Morcos. Beyond neural scaling laws: beating power law scaling via data pruning, 2022. | 2307.14430#62 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 63 | [54] Petru Soviany, Radu Tudor Ionescu, Paolo Rota, and Nicu Sebe. Curriculum learning: A survey. International Journal of Computer Vision, 130(6):1526â1565, Apr 2022.
[55] Claire Stevenson, Iris Smal, Matthijs Baas, Raoul Grasman, and Han van der Maas. Putting gpt-3âs creativity to the (alternative uses) test, 2022.
[56] Rohan Taori, Ishaan Gulrajani, Tianyi Zhang, Yann Dubois, Xuechen Li, Carlos Guestrin, Percy Liang, and Tatsunori B. Hashimoto. Stanford alpaca: An instruction-following llama model. https://github.com/tatsu-lab/ stanford_alpaca, 2023.
[57] Together. Redpajama-data: An open source recipe to reproduce llama training dataset, 2023.
[58] Mariya Toneva, Alessandro Sordoni, Remi Tachet des Combes, Adam Trischler, Yoshua Bengio, and Geoffrey J. Gordon. An empirical study of example forgetting during deep neural network learning, 2018. | 2307.14430#63 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 64 | [59] Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. Llama: Open and efï¬cient foundation language models, 2023.
[60] Neeraj Varshney, Swaroop Mishra, and Chitta Baral. Let the model decide its curriculum for multitask learning, 2022.
[61] Xiao Wang, Weikang Zhou, Qi Zhang, Jie Zhou, Songyang Gao, Junzhe Wang, Menghan Zhang, Xiang Gao, Yunwen Chen, and Tao Gui. Farewell to aimless large-scale pretraining: Inï¬uential subset selection for language model, 2023.
[62] Yizhong Wang, Yeganeh Kordi, Swaroop Mishra, Alisa Liu, Noah A. Smith, Daniel Khashabi, and Hannaneh Hajishirzi. Self-instruct: Aligning language model with self generated instructions, 2022.
14 | 2307.14430#64 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 65 | 14
[63] Yizhong Wang, Swaroop Mishra, et al. Super-naturalinstructions: Generalization via declarative instructions on 1600+ nlp tasks, 2022.
[64] Jason Wei, Maarten Bosma, Vincent Y. Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M. Dai, and Quoc V. Le. Finetuned language models are zero-shot learners, 2021.
[65] Richard T White. Research into learning hierarchies. Review of Educational Research, 43(3):361â375, 1973.
[66] Richard T. White and Robert M. Gagné. Past and future research on learning hierarchies. Educational Psychologist, 11(1):19â28, 1974.
[67] Xiaoxia Wu, Ethan Dyer, and Behnam Neyshabur. When do curricula work?, 2020.
[68] Sang Michael Xie, Hieu Pham, Xuanyi Dong, Nan Du, Hanxiao Liu, Yifeng Lu, Percy Liang, Quoc V. Le, Tengyu Ma, and Adams Wei Yu. Doremi: Optimizing data mixtures speeds up language model pretraining, 2023.
[69] Sang Michael Xie, Shibani Santurkar, Tengyu Ma, and Percy Liang. Data selection for language models via importance resampling, 2023. | 2307.14430#65 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 66 | [69] Sang Michael Xie, Shibani Santurkar, Tengyu Ma, and Percy Liang. Data selection for language models via importance resampling, 2023.
[70] Amir R. Zamir, Alexander Sax, William Shen, Leonidas Guibas, Jitendra Malik, and Silvio Savarese. Taskonomy: Disentangling task transfer learning. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Jun 2018.
[71] Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen, Christopher Dewan, Mona Diab, Xian Li, Xi Victoria Lin, Todor Mihaylov, Myle Ott, Sam Shleifer, Kurt Shuster, Daniel Simig, Punit Singh Koura, Anjali Sridhar, Tianlu Wang, and Luke Zettlemoyer. Opt: Open pre-trained transformer language models, 2022.
[72] Yi Zhang, Arturs Backurs, Sébastien Bubeck, Ronen Eldan, Suriya Gunasekar, and Tal Wagner. Unveiling transformers with lego: a synthetic reasoning task, 2022.
15
# A Broader Impacts and Limitations | 2307.14430#66 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 67 | 15
# A Broader Impacts and Limitations
Broader Impacts As more LMs are developed, a key criteria for their adoption and utility is if they exhibit a wide array of useful capabilities, such as generating harmless content, summarizing essays, and being conversational with the user. While improvements in other parts of the LM development pipeline such as training and architecture are important, many recent advances in building LMs with a wide array of useful capabilities have come from the data itself [9, 15, 17, 42, 56]. Our work is fundamental in investigating how LMs learn and how to select data to learn skills more efï¬ciently. However, we recognize that data selection methods can always be utilized to optimize for particular skills that may be considered malicious or negatively target or exclude speciï¬c groups [2]. Furthermore, pre-trained LMs have been found to have various biases [6, 25, 35, 43]. | 2307.14430#67 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 68 | Limitations The skills graph can either be provided (e.g., using a knowledge graph) or learned. Our work learns the skills graph using Algorithm 2 or Algorithm 3, which requires initial training runs on pairs of skills or each skill, respectively. This can be made more efï¬cient by performing these training runs on a smaller model and for fewer number of steps, but tradeoffs here have yet to be thoroughly investigated. SKILL-IT also assumes that the ordered skill set is provided; as discussed in sections 2.1 and 2.3, it is challenging to recover ordered skill sets simply via metadata attributes or embedding clustering. Otherwise, the best way to sample over collections of skills that form a complete or empty graph is random or stratiï¬ed sampling with no ordering to exploit. Our loss-based clustering approach presented in section 2.3 demonstrates that grouping by losses can provide an explanation for how skills are deï¬ned over data. An important direction for future work is to use such a clustering approach or other unsupervised algorithms in an end-to-end pipeline for skill discovery, skill graph learning, and data selection based on such skills.
# B Additional Algorithmic Details
# B.1 Derivation of SKILL-IT Update Rule | 2307.14430#68 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 69 | # B Additional Algorithmic Details
# B.1 Derivation of SKILL-IT Update Rule
First, we provide the derivation of our update rule from online mirror descent using the proximal point view [18]. We restate our optimization problem from (3):
m co. 1 . minimize â > Levai,j (fr) ®) j=l Piss, preAk-1 m st Levat,j (fe) = Levat,j(feâ-1) (1 â @AT pr-1) Vj ⬠[m],t=1,...,7 fi = O(fi-1, pr-1) Vt = 1,...T
Let Li(p) = 2 ST , Leva (fer) = 4 i; Levat,j(@( fi, p)); that is, p is the mixture we must choose at time ¢ and L, m Lai=j m Li=j is the average loss per skill of the model after it is trained on p at round t. A greedy approximation of (5) is minimize Li(p), peAk-1 given the model and mixtures at previous rounds. A linear approximation of L;(p) is
Li(p) © Li(pe-1) + (VLi-1 (pr-1), P â Di-1) (6)
Then, the problem of minimizing ¯Lt(p) can be approximated as | 2307.14430#69 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 70 | Then, the problem of minimizing ¯Lt(p) can be approximated as
# argmin, ¢qxâ1(7VLeâ1 (pr-1); P)
argmin, ¢qxâ1(7VLeâ1 (pr-1); P) (7)
after we drop terms from (6) that do not depend on p. Note that the 7 is a constant and does not impact the solution. The optimal solution to this problem is selecting the p that has the most weight on the slice with the largest gradient (e.g., a folow-the-leader sort of algorithm). To improve stability and prevent overfitting, we introduce regularization via a Bregman divergence Dp (p||pe_1) = h(p) â h(pe_1) â (Vh(pi-1), p â pe-1). After dropping terms that do not contain p, our problem is now
argmin,,< axâi (NV L1â1(pr-1); p) + h(p) â (VA(pe-1); P) (8)
Taking the gradient and setting it equal to 0 gives us
nV Li-1(pi-1) + VA(p) â VA(pr-1) = 0
nV Li-1(pi-1) + VA(p) â VA(pr-1) = 0 (9)
16 | 2307.14430#70 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 71 | nV Li-1(pi-1) + VA(p) â VA(pr-1) = 0 (9)
16
(7)
(9)
Algorithm 2: LEARNGRAPH (Brute-Force)
1: Input: Ordered skill set S = {s1, . . . , sk}. Number of training steps H, base model f . 2: for j â [k] do 3:
Train f on samples from Xsj for H steps and denote fH,j to be the model after training. Observe change in loss, δj
# j = Leval,j(f ) â Leval,j(fH,j).
4: 5: end for 6: for i, j â [k] do 7:
Train f on samples from Xsi ⪠Xsj for H steps and denote fH,i,j to be the model after training. Observe change in loss, δi,j if δij j then
# j = Leval,j(f ) â Leval,j(fH,i,j).
8:
j > δj Draw edge si â sj and set Aij > 0.
9: 10: end if 11: 12: end for 13: Return Adjacency matrix A â RkÃk
Algorithm 3: LEARNGRAPH (Approximate) | 2307.14430#71 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 72 | 9: 10: end if 11: 12: end for 13: Return Adjacency matrix A â RkÃk
Algorithm 3: LEARNGRAPH (Approximate)
1: Input: Ordered skill sets Strain and Seval. Number of training steps H, base model f . 2: for i â [k] do 3: 4: 5:
Train f on samples from Xstrain,i for H steps and denote fH,i to be the model after training. for j â [m] do
Observe change in loss, δi If δi end for
# j = Leval,j(f ) â Leval,j(fH,i).
j > 0, draw edge strain,i â seval,j and set Aij > 0.
6
7: 8: end for 9: Return Bipartite adjacency submatrix A â RkÃm
Similar to in standard multiplicative weights, we set h(p) = >>,
pi np; and Vh(p) = [In p; + 1];. Then,
# i
Similar to in standard multiplicative weights, we set h(p) = >>, pi np; and Vh(p) = [In p; + 1];. Then, | 2307.14430#72 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 73 | # i
Similar to in standard multiplicative weights, we set h(p) = >>, pi np; and Vh(p) = [In p; + 1];. Then,
Inp' = â nV iLy-1(Pr-1) => Visi = Ppexp(ânViLe(p2)) (10)
where V; is the ith element of the gradient. Now we wish to compute V;L:(pz) = 4 "4 Vi[Levai,j (fe41)] = 1ym Daj t VilLeval,j(@(ft, pe))]- Recall the dynamics model for Levat:
# 1 m
Leva (fi+1) = Levat,j(f:)(1 â Aljpe), qd)
The gradient of this model with respect to each training skill si is
âeeaus(a) = âAjjLevaij (ft) m =ViLi(p) = my âAjj Leva; (ft) (12)
Plugging this back into (10),
m Piai = Diexp (0 > Aijloa(F)) j=l (13)
where we can absorb the 1
# m into η.
# B.2 Graph Learning Method | 2307.14430#73 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 74 | where we can absorb the 1
# m into η.
# B.2 Graph Learning Method
We provide algorithms for learning the graph over an ordered skill set. In Algorithm 2, we discuss the brute-force approach for learning the adjacency matrix. This approach only works when Seval â Strain (e.g. pre-training and ï¬ne-tuning
17
cases), so we denote S = Strain in the algorithm box. In Algorithm 3, we discuss the linear approach for learning the adjacency matrix. This approach works even in the out-of-domain case when Seval and Strain are disjoint. In both approaches, the exact value of Aij can vary, but we can typically set it proportional to δi,j
# j â δj
j , the difference j, the change in loss itself, in the approximate case. The exact between the changes in loss, in the brute-force case or δi constructions and methods for learning each A in our experiments are in Appendix C.2.
# C Additional Experimental Details
# C.1 Datasets
We present details about each dataset used, including information on the skills and the validation dataset. A summary is presented in Table 2. | 2307.14430#74 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 75 | # C Additional Experimental Details
# C.1 Datasets
We present details about each dataset used, including information on the skills and the validation dataset. A summary is presented in Table 2.
Dataset Skill # skills Validation data Alpaca Pile of Law LEGO Addition NI (pre-training) NI (Spanish QG) NI (stance detection) NI (out-of-domain) RedPajama Instruction type Legal data source Reasoning chain depth Digit Task category Task category à language Task category Task category Data source 38 31 5 3 23 4 2 59, 12 7 50 samples per skill 645 samples per skill 100 samples per skill 100 samples per skill 50 samples per task 100 samples per task 50 samples per task 400 samples per task LM eval harness
Table 2: We list each dataset used as well as its corresponding skill. We include the number of skills in the training dataset, as well as details on how the validation dataset is constructed. | 2307.14430#75 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 76 | Table 2: We list each dataset used as well as its corresponding skill. We include the number of skills in the training dataset, as well as details on how the validation dataset is constructed.
⢠Alpaca dataset [56]: the Alpaca dataset consists of 52K instruction examples that were generated from text-davinci-003. We applied the Berkeley Neural Parser [26, 27] to each instruction, keeping 40777 samples it was able to parse successfully. If the sample began with a question, we annotated it with the skill âquestionâ, and otherwise we annotated it with the verb identiï¬ed from the parser. We grouped the data into a total of 38 skills, such as "list", "edit", "calculate", "describe" and "identify".
⢠Pile of Law [21]: the Pile of Law dataset consists of various sources of legal and administrative data, ranging from tax rulings to the worldâs constitutions. We evaluate on a subset of the Pile of Law validation dataset consisting of 13883 samples, where we selected max(645, source size) samples per source. We truncated each sample to be no more than 100K characters.
⢠LEGO [72]: for the LEGO synthetic, we set k = 5 and sample 192000 points across the skills. Our validation dataset consisted of 100 samples per skill. | 2307.14430#76 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 77 | ⢠LEGO [72]: for the LEGO synthetic, we set k = 5 and sample 192000 points across the skills. Our validation dataset consisted of 100 samples per skill.
⢠Addition: for the 3-digit addition synthetic, we set k = 3 and sample 192000 points across the skills. We use a validation dataset of 100 samples per skill.
⢠Natural Instructions [40, 63]: the Natural Instructions dataset is a large collection of tasks and their deï¬nitions in natural language. For the pre-training setting, we used a set of 23 task categories that had the largest degree (in-degree + out- degree) in the learned skills graph, for a total of 1, 232, 437 samples and 425 tasks to select from. We evaluated on 50 samples per task.
For the ï¬ne-tuning setting with Spanish question generation, we select data over 4 skills (Spanish question generation, Spanish question answering, English question generation, English question answering) for a total of 513210 samples and 212 tasks to select from. We evaluated on 100 samples per task.
For the ï¬ne-tuning setting with stance detection, we select data over 2 skills (stance detection, text matching) for a total of 50990 samples and 19 tasks to select from. We evaluated on 50 samples per task. | 2307.14430#77 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 78 | For the out-of-domain setting, we select data over all 59 task categories for a total of 2, 417, 867 samples and 753 tasks to select from. The test split consisted of 12 task categories and 119 tasks, and we evaluated on min(400, task size) samples per task.
⢠RedPajama [57]: the RedPajama dataset is a 1-trillion token dataset that aims to reproduce the LLaMA [59] training dataset. We select over the 7 data sources and evaluate using the LM evaluation harness [14].
18
Alpaca - 2.00 1.75 1.50 1.00 0.75 0.50 0.00 ite suggest summarize e] selec tal t translate Tew] write
Figure 8: Alpaca heatmap where i, jth entry is max(0, δi Diagonal entries are set to 0 for clearer visualization. j) (the change in loss on sj after training on si for 150 steps).
# C.2 Graph Learning Details
We describe how the skills graph was learned on each dataset. ⢠Alpaca (Figure 8): we use Algorithm 3 and train for K = 150 steps per skill. Each edge i â j has a weight of δi | 2307.14430#78 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 79 | j, the difference in loss on skill j before and after training on i. Next, we compare the average validation loss of skill-stratiï¬ed sampling versus random sampling when we train for K = 1000 steps. We ï¬nd that skill-stratiï¬ed sampling only does 0.007 better than random sampling, conï¬rming that Alpacaâs dense skills graph suggests that random sampling is the best we can do.
Pile of Law (Figure 9): we use Algorithm 3 and train for K = 150 steps. Each edge i â j has a weight of δi
difference in loss on skill j before and after training on i. j, the
⢠LEGO (Figure 10): we use both Algorithm 2 and Algorithm 3 and train for K = 6000 steps each. Each edge i â j has a weight of 0.5 if the amount of data associated with skill j that is needed to reach 0.01 validation loss is less when training on (i, j) than on j (edges are set to 0 if 0.01 validation loss is not reached, even if loss is decreasing). Each edge i â j is also set to 0.5 if training on i decreases loss directly on j. We set each diagonal entry of A to be 1. | 2307.14430#79 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 80 | ⢠Addition (Figure 11): we use Algorithm 2 and train for K = 6000 steps. Each edge i â j has a weight of 0.5 if the amount of data associated with skill j that is needed to reach 0.01 validation loss is less when training on (i, j) than on j (edges are set to 0 if 0.01 validation loss is not reached, even if loss is decreasing). We set each diagonal entry of A to be 1.
⢠Natural Instructions (Figure 12, 13, 14): we use Algorithm 3. For the pre-training setting, we train for K = 600 steps and assign each edge i â j a weight δi j equal to the change in loss on j in the ï¬rst 100 steps for all i, j â [k], including diagonal entries. For the ï¬ne-tuning setting, we train for K = 600 steps and assign each edge i â j a weight δi j equal to the change in loss before and after training. For the out-of-domain setting, we train for K = 600 steps and assign each edge i â j a weight δi j equal to the change in loss before and after training in the ï¬rst 100 steps. ⢠RedPajama (Figure 15): we use Algorithm 3 and train for 1 billion tokens per data source. We assign each edge i â j a | 2307.14430#80 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 82 | atticus_contracts bva_opinions canadian_decisions cfpb_creditcard contracts . congressional hearings courtlistener_docket_entry_documents ~ courtlistener_ opinions dol_ecab echr edgar eoir eurlex euro_parl federal_register founding docs hhs_alj_ opinions «os , __ icj-peij medicaid_policy qui dance nlrb decisions ~ oig olc_ memos r_legaladvice resource contracts scotus filings scotus oral arguments sec_administrative_ proceedings con tax_rulings un_debates _ _us_pills uspto_office_actions 2222 aooe SESE 2.08 00090 300 8 asd ga8e BBS eo ag So On Oo Q & Oo is S ressional_ hearing documen g) courtlistener_docket_ent Inlons ry | courtlist âdol_ecab ener op echr edgar eoir eurlex Pile of Law ) parl gister ding docs euro hhs alj_ opinions federal re foun ee dance ecisions medicaid _policy _g oig olc_ memos is S tracts scotus_filin scotus oral art | 2307.14430#82 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 84 | Figure 9: Pile of Law heatmap where i, jth entry is max(0, δi Diagonal entries are set to 0 for clearer visualization. j) (the change in loss on sj after training on si for 150 steps).
20
LEGO -1.0 4 - 0.8 N 0.6 sp) 0.4 st 0.2 Te) ~ 0.0 1 2 3 4 5
Figure 10: LEGO heatmap with k = 5 where i, jth entry is set to 0.5 if the number of steps needed to reach 0.01 loss on skill j when training on a balanced mixture of skills i and j is less than when training on skill j only.
Addition -1.0 77 0.8 0.6 N 0.4 ~* 0.2 0.0 1 2 3
Figure 11: Addition heatmap with k = 3 where i, jth entry is set to 0.5 if the number of steps needed to reach 0.01 loss on skill j when training on a balanced mixture of skills i and j is less than when training on skill j only.
21 | 2307.14430#84 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 85 | Natural Instructions answer _verification code_to_text discourse_connective_identification entity_generation entity_relation_classification information_extraction irony_detection preposition_prediction punctuation_error_detection question_answering question_generation question_understanding sentence_expansion sentiment_analysis stance_detection summarization text_categorization text_matching text_simplification text_to_code toxic_language_detection word_semantics wrong_candidate_generation 0.0 -e-e-e----------- | SSZLLSSSSIRSHOBIOSSHSVORS BS Isesoooog nts goushatogs SSSsSeseks ggg eSNNZe Sag BTS PSbeseessacoeeseesces Bos gnkoubog og eK Io POS Wo OG OGTR HLT 12 1G OG e iGo ak IAS POOAG I> lh 1 \8 ow Pesgaoo_ lOogsig eke sa MESS oEL RAS oD! B PU SSSRSEESeaes Sa e§s obese t aborts Beer 3 B Sess ateeakfen 3 R Fy 2 B°SBE os tose so $ 8 & 3 a o oH a9 Bon + & ¢ co go oBFTERs | 8 2 48 | 2307.14430#85 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 87 | Figure 12: Natural Instructions heatmap where i, jth entry is max(0, δi 100 steps). Diagonal entries are set to 0 for clearer visualization. j) (the change in loss on sj after training on si for
Stance detection Stance Detection - 1.0 0.5 Text Matching 0.0 Text Matching Stance Detection
Spanish question generation English QA Spanish QA English QG Spanish QG English QA Spanish QA English QG Spanish QG
Spanish question generation English QA Stance Detection - Spanish QA Text Matching English QG Spanish QG Text Matching English QA Spanish QA English QG Spanish QG Stance Detection
Figure 13: Spanish question generation and stance detection heatmaps where i, jth entry is max(0, δi on sj after training on si for 100 steps).
# (the change in loss
22 | 2307.14430#87 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 88 | Natural Instructions answer verification code _to_text coherence classification commonsense classification dialogue_generation dialogue state tracking discourse_connective identification discourse relation classification entity_generation entity_relation_ classification explanation fact_verification fill in the blank gender classification grammar error detection 0.12 information extraction intent_identification irony detection linguistic_probing mathematics misc. 0.10 named_entity_recognition negotiation_strategy detection number conversion paraphrasing poem_generation __, Pos_tagging 0.08 preposition_prediction program. execution punctuation_error detection question answering - question_decomposition question_generation question_understanding sentence composition sentence compression sentence_expansion sentence ordering sentence_perturbation sentiment_analysis 0.04 spam classification speaker identification spelling error detection stance detection stereotype_detection story_composition 0.02 style transfer summarization text_categorization text_completion text_matching text_quality_evaluation 0.00 text_simplification 2 text_to_code toxic_language detection translation word_relation classification word semantics wrong_candidate generation -0.14 0.06 (â2-8 -h- 8-8 -B | 2307.14430#88 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 90 | Figure 14: Natural Instructions heatmap for out-of-domain setting where rows are for the training skills and columns are for the evaluation skills. The i, jth entry is max(0, δi
23
RedPajama arxiv books 0.2 c4 common_crawl github 0.1 stackexchange wikipedia 0.0 ge arc_easy hellaswag arc_challen lambada_openai winogrande
Figure 15: RedPajama heatmap for out-of-domain setting where rows are for the training skills and columns are for the evaluation skills. The i, jth entry is max(0, δi
⢠LEGO: η = 0.5, T = 6, w = 3. We train for 6000 steps.
⢠Addition: η = 0.1, T = 5, w = 3. We train for 6000 steps.
⢠Natural Instructions (pre-training): η = 0.2, T = 1. We train for 5000 steps. | 2307.14430#90 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 91 | ⢠Natural Instructions (pre-training): η = 0.2, T = 1. We train for 5000 steps.
For the LEGO random baseline, when we selected points at random, we used an imbalanced training dataset with proportions 1:1:1:3:5. For the addition random baseline, we used an imbalanced dataset with randomly selected proportions: 13:14:18. For the curriculum learning baselines, the pacing function, g(i), denotes the size of the subset of the highest scoring samples that we uniformly select from in the ith epoch. We deï¬ne our pacing function as g(i) = iH M , where H is the number of steps and M is 5 epochs for LEGO and NI, and 3 for addition.
# SKILL-IT ï¬ne-tuning
⢠LEGO: η = 0.5, T = 10, w = 3. We train for 6000 steps.
⢠Addition: η = 0.1, T = 5, w = 3. We train for 6000 steps.
⢠Natural Instructions (Spanish QG): η = 0.8, T = 6, w = 3. We train for 600 steps.
⢠Natural Instructions (stance detection): η = 0.2, T = 6, w = 3. We train for 600 steps.
# SKILL-IT out-of-domain | 2307.14430#91 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 92 | ⢠Natural Instructions (stance detection): η = 0.2, T = 6, w = 3. We train for 600 steps.
# SKILL-IT out-of-domain
⢠Natural Instructions: η = 0.2, T = 10, w = 3. We train for 5000 steps.
⢠RedPajama: η = 100, T = 1. We train for 3 billion tokens.
All results are computed over 5 random seeds.
Batch sizes of 32 and 64 were used for the LEGO and addition synthetic on the 125M and 1.3B parameter model, respectively. Batch sizes of 4 and 16 were used for the Natural Instructions experiments on the 125M and 1.3B parameter model.
For the out-of-domain Natural Instructions experiment and Alpaca graph learning experiments, a learning rate of 5e-6 with linear scheduler and 50 warmup steps was used. For the Natural Instructions continual pre-training experiment on the 1.3B parameter model, a learning rate of 1e-6 was used. All other experiments used a learning rate of 5e-5. All experiments used AdamW with betas = 0.9, 0.999, eps = 1e-8, and weight decay = 0.01. A context window of 512 was used for all experiments except LEGO and addition, which used a window of 128. | 2307.14430#92 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 93 | Experiments with the Addition dataset were run using an Nvidia RTX A6000. Other experiments using the GPT-Neo 125M parameter model were run on an Nvidia Tesla P100. Experiments using the GPT-Neo 1.3B parameter model were run on an Nvidia Tesla A100.
24
Model performance on LEGO skill 4 Model performance on LEGO skill 0.9 ââ Trained on skill 4 ââ Trained on skill 4 =ââ Trained on skills 2, 4 ââ Trained on skills 3, 4 ° Ny nv v Validation Loss ° | f=) id a oo 0 2000 4000 6000 0 2000 4000 6000 Steps Steps
4
Figure 16: Performance on LEGO skill 4 when training on skill 4, skills 2 and 4, and skills 3 and 4. Even though skill 3 and skill 4 share an edge in the LEGO syntheticâs underlying reasoning chain (i.e. a model predicting correct for the fourth variable is one extra step beyond predicting correct for the third variable), we ï¬nd that training on skills 2 and 4 helps improve performance on skill 4 more.
Model performance on LEGO skill 3 (tree) Model performance on LEGO skill 2 (tree) | 2307.14430#93 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 94 | Model performance on LEGO skill 3 (tree) Model performance on LEGO skill 2 (tree)
â Trained on skill 3 â Trained on skill 3 0.6 ââ Trained on skill 2 a 0.6 ââ Trained on skill 2 6 4 3 ic] 3 0.2 0.2 s 0.0 0.0 0 1000 2000 3000 0 1000 2000 3000 Steps Steps
a 6 4
3 ic] 3 3 s
Figure 17: Performance on LEGO skill 2 and 3 when training on skills 2 and 3. The reasoning pattern is a tree rather than a chain over k = 4 variables. Skills 2 and 3 are at the same âdepthâ in the graph and both depend on skill 1, so there is positive inï¬uence between the skills despite there being no edge between 2 and 3 in the LEGO reasoning graph.
# D Additional Experimental Results
# D.1 Additional examples of LEGO ordered skill sets | 2307.14430#94 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 95 | # D Additional Experimental Results
# D.1 Additional examples of LEGO ordered skill sets
For the LEGO synthetic, it may appear obvious that the skills graph is equivalent to the reasoning chain over the variables. However, in Figure 16 we see that this is not the case. Training on skills 2 and 4 together results in lower loss on skill 4 than when trained on skill 4 alone. However, training on skills 3 and 4 together results in roughly the same loss on skill 4 as when training on skill 4 alone, even though skill 3 and skill 4 share an edge in the LEGO syntheticâs underlying reasoning chain. This suggests that our intuition for how skills inï¬uence each other does not always match how the model learns skills.
Next, we consider a slightly more complex reasoning pattern on the LEGO synthetic. Instead of a chain, we construct a tree, where two variables in the LEGO synthetic are both deï¬ned in terms of the same parent variable. For example,
Input: c = val 1, y = not w, v = val c, w = not c. Output: y = 1. | 2307.14430#95 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 96 | Input: c = val 1, y = not w, v = val c, w = not c. Output: y = 1.
In this example, k = 4 and both v and w are written in terms of c, and the reasoning graph has edges 1 â 2, 1 â 3, 2 â 4. In this case, we see that training on skill 2 or skill 3 both improve losses on skills 2 and 3 (Figure 17). However, unlike the previous ï¬gures, training on skills 2 and 4 or skills 3 and 4 do not signiï¬cantly help reduce loss on skill 4 (Figure 18). Again, these measurements demonstrate that the reasoning graph does not necessarily equal the skills graph.
# D.2 Unsupervised skill recovery
We explore several clustering techniques for recovering the skills in the LEGO synthetic on the validation dataset. Our results are shown in Table 3.
We ï¬rst cluster based on the pre-trained model embeddings of the last token and the average token. We also report
25 | 2307.14430#96 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 97 | We ï¬rst cluster based on the pre-trained model embeddings of the last token and the average token. We also report
25
Model performance on LEGO skill 4 (tree) Model performance on LEGO skill 4 â Trained on skill 4 0.8 â Trained on skill 4 ââ Trained on skills 2, 4 a ââ Trained on skills 3, 4 $0.6 Bo6 y eoâ 4 B04 E \Y £0. £04 ic] zg 0.2 $0.2 0.0 0.0 0 1000 2000 3000 0 1000 2000 3000 Steps Steps
# a
° 4
# ic] zg
# (tree)
Figure 18: Performance on LEGO skill 4 when training on skills 2, 4 and skills 3, 4. We ï¬nd that in both cases, the beneï¬t from training on additional skills is minor. For instance, training on 2 and 4 reaches 0.01 loss in 2700 steps, while training on 4 only reaches it in 2100 steps.
Cluster method Pretrained embedding of last token Pretrained embedding of average token Trained model embedding of last token Sentence-BERT embedding Losses over multiple runs Accuracy 24.8 ± 0.5 25.2 ± 1.1 38.4 ± 0.8 23.9 ± 0.7 61.0 ± 1.6 | 2307.14430#97 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 98 | Table 3: Clustering-based skill recovery methods on the LEGO dataset. The validation dataset we cluster consists of 500 points with k = 5, and results are reported over 10 runs of k-means.
accuracies of clustering based on the trained model embeddingâs last token, where we train the model using random sampling for 6000 steps, and clustering based on Sentence-BERT embeddings. Among these four methods, using the trained model embeddings has the highest accuracy of 38.4 points.
Next, we cluster points based on losses. In particular, we do 10 runs, each for 6000 steps and with a randomly sampled mixture of skills. For each run, we evaluate the model on the validation dataset at 120 checkpoints. Then, each sample in the validation dataset has 1200 losses associated with it, comprising a feature vector for that sample. We perform k-means clustering on these features, which has an accuracy of 61.0 points, signiï¬cantly higher than the second best accuracy of 38.4.
# D.3 Full results for Section 4
# D.3.1 Per-skill performance
In this section, we provide tables containing the per skill break-down of our results from Section 4. | 2307.14430#98 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 99 | # D.3 Full results for Section 4
# D.3.1 Per-skill performance
In this section, we provide tables containing the per skill break-down of our results from Section 4.
Continual Pre-training In the continual pre-training setting, we report two additional baselines that combine curriculum learning with skills. Curriculum learning has been proposed for multitask learning [60], in which groups of data are ranked by their average score and then trained in order of this ranking (with mixing of previously seen groups to avoid forgetting). We construct two baselines, Skill-curriculum and Skill-anticurriculum, using Algorithm 1 from [60]. In contrast to the random baseline which has imbalanced skills, this approach has knowledge of skills and thus uses a skill-stratiï¬ed training dataset to sample from. We set the fraction of the previous group to be frac = 0.4, as we found that setting frac = 0.0 resulted in forgetting. | 2307.14430#99 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 100 | We report loss per skill for the LEGO synthetic in Table 4, which corresponds to the results in Figure 4. We report accuracy per skill in Table 5 and Figure 19. We report the loss per skill for the Addition synthetic in Table 6, which also correspond to to the results in Figure 4. Finally, we report validation loss per task category for the Natural Instructions continual pre-training experiment in Table 7, where we ï¬nd that SKILL-IT outperforms random sampling by 3.2% on average across skills.
26 | 2307.14430#100 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 101 | Skill 1 Skill 2 Skill 3 Skill 4 Skill 5 Average Random Curriculum Anticurriculum Skill-stratiï¬ed Skill-curriculum 0±0.000 0±0.000 0±0.000 0±0.000 0±0.000 Skill-anticurriculum 0.001±0.001 SKILL-IT 0±0.000 0.675±0.041 0.645±0.052 0.690±0.003 0.045±0.036 0.484±0.200 0.174±0.118 0.002±0.002 0.688±0.008 0.686±0.018 0.695±0.004 0.056±0.029 0.698±0.027 0.245±0.091 0.024±0.031 0.673±0.049 0.674±0.042 0.693±0.003 0.079±0.044 0.697±0.010 0.443±0.125 0.013±0.010 0.667±0.056 0.671±0.0459 0.689±0.004 0.050±0.025 0.689±0.007 0.566±0.118 0.022±0.021 0.541±0.031 0.535±0.029 | 2307.14430#101 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 104 | Skill 1 Skill 2 Skill 3 Skill 4 Skill 5 Average 100.0±0.0 100.0±0.0 100.0±0.0 100.0±0.0 100.0±0.0 Skill-anticurriculum 100.0±0.0 100.0±0.0 Random Curriculum Anticurriculum Skill-stratiï¬ed Skill-curriculum SKILL-IT 54.2±5.9 60.0±10.6 53.4±2.3 98.2±1.8 75.2±30.1 90.2±8.1 99.2±0.8 58.0±3.1 55.2±5.8 49.0±4.8 98.2±1.3 52.2±3.7 88.2±8.3 99.0±1.0 48.0±6.3 51.2±6.3 48.2±6.4 97.8±1.6 51.0±4.6 73.2±12.2 99.4±0.5 54.4±7.3 51.8±6.1 56.0±5.7 98.2±1.3 54.4±3.1 62.4±9.4 99.6±0.5 62.9±3.5 | 2307.14430#104 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 106 | Table 5: Results on accuracy per skill (binary classiï¬cation) for LEGO pre-training experiment, averaged over 5 random seeds.
LEGO Skill 1 LEGO Skill 2 LEGO Skill 3 1.0 . 0.8 p> is £ =] 8 0.6 < 0.4 0 2000 4000 6000 i} 2000 4000 6000 i} 2000 4000 6000 LEGO Skill 4 LEGO Skill 5 Average per skill 1.0 1.0 1.0 0.9 0.9 0.8 P08 0.8 5 0.6 5 0.7 0.7 : ââ Random & ââ Curriculum 0.6 â Anticurriculum 0.6 0.4 â Skill-stratified 0.5 â Skill-curriculum 0.5 â Skill-anticurriculum 04 0.2 â skit 0 2000 4000 6000 0 2000 4000 6000 0 2000 4000 6000 Steps Steps Steps
Figure 19: Accuracy of SKILL-IT on each skill (binary classiï¬cation) on the LEGO synthetic in the continual pre-training setting. SKILL-IT attains higher accuracy more quickly than baselines that both do and do not utilize the notion of skills. | 2307.14430#106 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 107 | Skill 1 Skill 2 Skill 3 Average 0.008±0.007 0.009±0.011 0.007±0.010 0.012±0.011 0.016±0.013 Skill-anticurriculum 0.005±0.008 0.004±0.003 Random Curriculum Anticurriculum Skill-stratiï¬ed Skill-curriculum SKILL-IT 0.020±0.019 0.010±0.008 0.012±0.013 0.015±0.015 0.019±0.013 0.037±0.028 0.009±0.007 0.005±0.005 0.008±0.010 0.008±0.017 0.010±0.020 0.010±0.003 1.141±1.126 0.013±0.017 0.011±0.014 0.009±0.010 0.009±0.014 0.012±0.016 0.015±0.010 0.395±0.371 0.009±0.011
Table 6: Results on validation loss per skill for Addition pre-training experiment, averaged over 5 random seeds.
27
2 8 | 2307.14430#107 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 108 | Skill Random Curriculum Anticurriculum Skill-stratiï¬ed Skill-curriculum Skill-anticurriculum Answer Veriï¬cation Code to Text Discourse Connective Identiï¬cation Entity Generation Entity Relation Classiï¬cation Information Extraction Irony Detection Preposition Prediction Punctuation Error Detection Question Answering Question Generation Question Understanding Sentence Expansion Sentiment Analysis Stance Detection Summarization Text Categorization Text Matching Text Simpliï¬cation Text to Code Toxic Language Detection Word Semantics Wrong Candidate Generation 2.297±0.058 0.246±0.021 2.927±0.069 2.033±0.421 1.020±0.147 2.154±0.040 3.024±0.154 0.979±0.124 2.950±0.065 2.277±0.005 2.617±0.005 1.965±0.051 2.501±0.095 3.203±0.012 1.810±0.100 2.961±0.015 2.488±0.023 2.177±0.059 2.155±0.023 0.560±0.037 3.106±0.027 2.092±0.027 | 2307.14430#108 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 109 | 2.177±0.059 2.155±0.023 0.560±0.037 3.106±0.027 2.092±0.027 2.438±0.021 2.368±0.055 0.203±0.019 3.084±0.067 2.012±0.437 1.014±0.140 2.247±0.037 3.798±0.095 0.887±0.147 3.120±0.052 2.367±0.006 2.777±0.015 2.199±0.059 2.598±0.097 3.415±0.016 1.775±0.120 3.149±0.023 2.692±0.029 2.232±0.055 2.193±0.039 0.495±0.036 3.496±0.017 2.334±0.034 2.606±0.039 2.391±0.061 1.099±0.115 2.932±0.058 2.363±0.234 1.533±0.138 2.352±0.042 2.942±0.158 1.488±0.213 2.961±0.064 2.398±0.006 2.695±0.008 | 2307.14430#109 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 110 | 2.942±0.158 1.488±0.213 2.961±0.064 2.398±0.006 2.695±0.008 2.060±0.033 2.583±0.074 3.209±0.010 2.231±0.128 3.041±0.014 2.553±0.006 2.316±0.048 2.325±0.033 1.215±0.052 3.058±0.029 2.156±0.064 2.519±0.027 2.180±0.059 0.178±0.016 2.805±0.071 1.803±0.384 0.859±0.131 2.140±0.037 2.680±0.146 0.845±0.152 3.264±0.061 2.542±0.004 2.783±0.021 1.958±0.051 2.225±0.095 3.278±0.014 1.385±0.070 2.960±0.019 2.570±0.015 2.152±0.061 1.926±0.026 0.490±0.029 3.199±0.024 1.916±0.043 2.506±0.026 | 2307.14430#110 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 111 | 1.926±0.026 0.490±0.029 3.199±0.024 1.916±0.043 2.506±0.026 2.249±0.116 0.126±0.009 2.891±0.001 1.853±0.483 0.825±0.022 2.286±0.022 3.889±0.066 0.941±0.019 3.019±0.010 2.689±0.001 3.062±0.006 2.385±0.022 2.311±0.076 3.607±0.012 1.361±0.114 3.323±0.028 3.001±0.007 2.324±0.004 2.037±0.005 0.433±0.014 3.758±0.025 1.784±0.048 2.849±0.029 2.325±0.085 1.232±0.070 2.925±0.011 2.068±0.719 0.959±0.009 2.338±0.025 2.099±0.152 1.044±0.029 3.360±0.013 2.707±0.016 2.876±0.032 2.100±0.054 | 2307.14430#111 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 112 | 1.044±0.029 3.360±0.013 2.707±0.016 2.876±0.032 2.100±0.054 2.408±0.074 3.308±0.015 1.823±0.189 3.021±0.013 2.635±0.014 2.304±0.035 2.156±0.011 1.455±0.086 3.155±0.050 2.424±0.038 2.574±0.018 Average 2.173±0.028 2.307±0.025 2.366±0.026 2.115±0.027 2.304±0.031 2.317±0.052 SKILL-IT 2.158±0.059 0.223±0.017 2.784±0.068 1.863±0.418 0.908±0.146 2.073±0.042 2.797±0.155 0.876±0.173 3.216±0.055 2.448±0.008 2.666±0.012 1.895±0.043 2.236±0.083 3.213±0.012 1.556±0.125 2.907±0.012 2.448±0.017 | 2307.14430#112 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 115 | Skill Random Skill-stratiï¬ed SKILL-IT Answerability Classiï¬cation Cause Effect Classiï¬cation Coreference Resolution Data to Text Dialogue Act Recognition Grammar Error Correction Keyword Tagging Overlap Extraction Question Rewriting Textual Entailment Title Generation Word Analogy 3.048±0.003 2.068±0.004 3.101±0.003 2.363±0.004 2.329±0.009 2.399±0.008 2.744±0.005 2.749±0.011 2.591±0.009 2.472±0.002 3.027±0.002 1.665±0.016 3.076±0.002 2.101±0.005 3.142±0.004 2.388±0.005 2.364±0.010 2.418±0.009 2.760±0.007 2.763±0.012 2.628±0.011 2.503±0.003 3.037±0.002 1.682±0.015 3.043±0.003 2.067±0.006 3.099±0.004 2.359±0.005 2.320±0.009 2.389±0.007 2.733±0.006 | 2307.14430#115 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 117 | Table 8: Validation loss per skill for data selection in out-of-domain setting over Natural Instructions train task split and test task split.
In Table 9 we provide a breakdown of the RedPajama experimentâs accuracy per evaluation skill, corresponding to the results in Figure 7.
1 Billion Tokens 2 Billion Tokens 3 Billion Tokens Uniform SKILL-IT Uniform SKILL-IT Uniform SKILL-IT ARC Challenge (acc norm) ARC Easy (acc norm) BoolQ COPA HellaSwag (acc norm) LAMBADA OpenAI PIQA (acc norm) Winogrande 35.4 62.2 68.9 81.0 63.9 64.4 74.8 62.8 34.6 61.2 68.2 82.0 63.7 67.0 75.0 63.9 35.3 62.4 67.7 80.0 63.8 65.9 75.5 63.9 34.9 61.7 68.6 81.0 63.9 66.7 75.2 63.2 34.6 62.5 67.2 81.0 64.0 66.8 75.0 63.4 34.8 62.0 68.7 81.0 63.9 66.0 75.7 63.1 Average accuracy 64.2 64.4 64.3 64.4 64.3 64.4 | 2307.14430#117 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 118 | Table 9: Performance of model trained on RedPajama with uniform sampling and SKILL-IT on LM evaluation harness. Unless otherwise noted, accuracy is reported for each task.
# D.3.2 Weight trajectories
We provide SKILL-ITâs weight trajectories for each result. The weight per skill across training steps for the LEGO pre- training experiment corresponding to Figure 4 (left) is shown in Figure 20. We see that SKILL-IT initially allocates more weight to skill 2 and less to 1, 3, 4, 5. Since skill 1 is learned quickly, the weight on skill 1 immediately drops to below 0.1 at 1000 steps. The weight on skills 3, 4, and 5 increase from around 0 to 3000 steps, during which their respective validation losses are higher than those of skills 1 and 2. Near the end of training, all losses are converging to 0, and so the weight per skill is roughly uniform.
The weight per skill across training steps for the addition pre-training experiment corresponding to Figure 4 (right) is shown in Figure 21. SKILL-IT allocates more weight to skill 2, which has an edge to skill 1 as shown in Figure 11. It also allocates very little weight to skill 3, which is learned faster than the other two skills. Eventually, it puts more weight on skill 1, the hardest skill, and then converges to uniform sampling as all validation losses approach 0. | 2307.14430#118 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 119 | The weight per skill across training steps for the LEGO ï¬ne-tuning experiment and the Spanish question generation and stance detection experiments corresponding to Figure 5 is shown in Figure 22. Since there is only one target skill in these experiments, the mixture of weights approaches uniform as the loss on the target skill approaches 0. It is interesting
29
Skill-It LEGO weights
0.35 â skill1 â Skill 2 0.30 â skill3 â skill 4 0.25 â skill5 ; | Weight per skill o N Oo 0.10 0.05 ââ y y + + + \ 0 1000 2000 3000 4000 5000 6000 Steps
Figure 20: Weight per skill for LEGO pre-training experiment. SKILL-IT initially allocates more weight to skill 2, but eventually puts more weight on harder skills (3, 4, 5) before converging to uniform sampling when all losses converge roughly to 0.
to explore how to reduce edge weights and regularization so that the mixture approaches the target skill instead, although preliminary experiments where we decayed the edge weight and the strength of the Bregman divergence term did not appear better. We hypothesize that since training on a uniform mixture (as in Figure 3) did strictly better than training on the target skill and their loss curves did not intersect during the training run, it is better to allocate non-negligible weight on all skills throughout the training run. | 2307.14430#119 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 120 | The weight per skill across training steps for the Natural Instructions out-of-domain experiment corresponding to Figure 6 is shown in Figure 23, where the legend is provided for the top 10 task categories with the largest weights. While the initial weights based on the skills graph roughly establishes the order of weight magnitude, the differences among the losses on the evaluation skills increases the range of weights as training continues. As validation losses saturate, the weights also converge to ï¬xed values.
# D.4 Experiments on 1.3B parameter model
We demonstrate that the skills graph learned on the 125M parameter model can be used for data selection with the GPT-Neo- 1.3B model. We present results in the continual pre-training setting on the LEGO synthetic and Natural Instructions.
All results are reported over 3 random seeds. For the LEGO experiment, we train for 1500 steps with η = 0.5, T = 30, w = 3. For the NI experiment, we train for 5000 steps with η = 0.2, and T = 1. The skill graphs were learned using the 125M parameter model as described in section C.2. | 2307.14430#120 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 121 | In Figure 24, we train the 1.3B model using SKILL-IT for the LEGO synthetic and ï¬nd that it still outperforms random and skill-stratiï¬ed sampling on average. In particular, while performance across sampling methods is similar for early skills, the discrepancy is larger for skill 5, for which SKILL-IT allocates more weight to dynamically. In Figure 25, we provide the weight trajectories of SKILL-IT. We observe that the weight trajectories are similar to that on the 125M parameter model, where initial weight is allocated towards skill 2. Later on, more weight is allocated towards skills 4 and 5, whose losses are higher, and eventually the weight mixture converges to uniform as all losses converge to near 0.
In Table 10, we report performance of SKILL-IT with the 1.3B model on the Natural Instructions pre-training experiment and ï¬nd that the trends from the smaller model holdâSKILL-IT outperforms random and skill-stratiï¬ed sampling on average.
# D.5 Ablations
We report ablations on the skills graph and the online component of SKILL-IT. Instead of using A in Algorithm 1, we study the performance when the identity matrix is used instead; intuitively, this corresponds to a misspeciï¬ed skills graph where
30
Skill-It addition weights | 2307.14430#121 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 122 | 30
Skill-It addition weights
0.55 0.50 0.45 0.40 0.35 0.30 Weight per skill 0.25 0.20 0.15 weights 0 1000 Steps 2000 3000 4000 5000 6000
Figure 21: Weight per skill for addition pre-training experiment. SKILL-IT initially allocates more weight to skill 2, which has an edge to skill 1, while allocating little weight to skill 3 which is learned quickly. Eventually, SKILL-IT puts more weight on the harder skill 1 before converging to uniform sampling when all losses roughly approach 0.
Skill-It LEGO weights for learning skill 3 Skill-It weights for Spanish QG Weight per skill 0.35 Zane 0 â English QA â Spanish QA â English QG â Spanish QG Weight per skill Ss 0.450 0.425 Skill-It weights for Stance Detection a â Stance Detection â Text Matching Nao 0 1000 2000 3000 4000 5000 6000 Steps 0 100 200 300 400 500 600 Steps 0 100 200 300 400 500 600 Steps
Figure 22: Weight per skill for ï¬ne-tuning experiments. Left: LEGO; Center: Spanish question generation; Right: stance detection.
Skill-It weights for Natural Instructions | 2307.14430#122 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 123 | Figure 22: Weight per skill for ï¬ne-tuning experiments. Left: LEGO; Center: Spanish question generation; Right: stance detection.
Skill-It weights for Natural Instructions
Skill-It weights for Natural Instructions S is S N Weight per skill ooo 9 6 oo 8 PB & es 2 ° N 2 ° 6 0 1000 2000 3000 Steps 4000 5000 question_generation question_answering text_categorization sentiment_analysis wrong_candidate_generation text_matching summarization information extraction question_understanding toxic_language_detection
Figure 23: Weight per skill for Natural Instructions out-of-domain experiment. The legend shows the top 10 skills with the largest weight. While the relative order of weight magnitude does not change signiï¬cantly across training, the incorporation of loss dramatically increases the range of the weights, showing the importance of an online algorithm.
31 | 2307.14430#123 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 124 | 31
LEGO Skill 1 LEGO Skill 2 LEGO Skill 3 10° 10° R107 4 3 10 10-1 g 10-2 4 10- $1073 10-2 3 to a 10-4 g 10-3 10-5 10-4 0 500 1000 1500 0 500 1000 1500 0 500 1000 1500 10° LEGO Skill 4 10° LEGO Skill 5 10° Average per skill B fo} -1 c107 10-1 10 a 4 g 10-2 10-2 10-2 3 â Random g -3| ââ Skill-stratified > 10-3 - 10-3 10-8 â Skill-It 0 500 1000 1500 0 500 1000 1500 0 500 1000 1500 Steps Steps Steps
Figure 24: Performance of SKILL-IT for LEGO pre-training setting when skills graph is learned on a 125M parameter model and used for data selection with a 1.3B model. SKILL-IT on average still outperforms random and skill-stratiï¬ed sampling, suggesting that ï¬ndings on ordered skill sets can transfer from small models to large models.
Skill-It LEGO weights for 1.3B param model
Weight per skill o fg 9S bo ow B® ° > 0.0 0 250 500 750 1000 1250 1500 Steps | 2307.14430#124 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 125 | Skill-It LEGO weights for 1.3B param model
Weight per skill o fg 9S bo ow B® ° > 0.0 0 250 500 750 1000 1250 1500 Steps
Figure 25: Weight per skill for LEGO pre-training experiment on 1.3B parameter model. The trajectories are similar to those of the 125M parameter model in Figure 20. SKILL-IT initially allocates more weight to skill 2, but eventually puts more weight on skills 4 and 5 before converging to uniform sampling when all losses converge to near 0.
32 | 2307.14430#125 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 126 | Skill Random Skill-stratiï¬ed SKILL-IT Answer Veriï¬cation Code to Text Discourse Connective Identiï¬cation Entity Generation Entity Relation Classiï¬cation Information Extraction Irony Detection Preposition Prediction Punctuation Error Detection Question Answering Question Generation Question Understanding Sentence Expansion Sentiment Analysis Stance Detection Summarization Text Categorization Text Matching Text Simpliï¬cation Text to Code Toxic Language Detection Word Semantics Wrong Candidate Generation 2.005±0.059 0.302±0.032 2.529±0.046 2.108±0.328 1.130±0.048 2.032±0.013 2.802±0.125 1.095±0.040 2.633±0.027 1.947±0.003 2.214±0.007 1.928±0.020 2.054±0.018 2.771±0.009 1.814±0.151 2.531±0.009 2.289±0.016 1.967±0.008 1.861±0.003 0.614±0.030 2.853±0.020 1.999±0.023 2.187±0.028 1.903±0.069 | 2307.14430#126 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 127 | 0.614±0.030 2.853±0.020 1.999±0.023 2.187±0.028 1.903±0.069 0.204±0.022 2.372±0.054 1.788±0.429 0.836±0.006 1.992±0.006 2.528±0.146 0.686±0.041 3.188±0.055 2.119±0.003 2.345±0.008 1.837±0.031 1.828±0.060 2.818±0.006 1.500±0.117 2.472±0.012 2.341±0.021 1.913±0.005 1.692±0.023 0.518±0.030 2.911±0.019 1.870±0.039 2.192±0.023 1.890±0.072 0.269±0.032 2.393±0.056 1.885±0.461 0.841±0.010 1.933±0.013 2.585±0.149 0.774±0.029 2.726±0.025 2.073±0.001 2.263±0.010 1.700±0.042 | 2307.14430#127 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 129 | Table 10: Results when skills graph for Natural Instructions learned on 125M parameter model is used for data selection with a 1.3B model. We see that SKILL-IT on average still outperforms random and skill-stratiï¬ed sampling, even though the edges used by SKILL-IT are not derived from the larger model.
no skill inï¬uences another skill. We refer this approach as âNo graphâ. Note that the opposite case of a complete graph recovers skill-stratiï¬ed sampling, which we already have as a baseline. | 2307.14430#129 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 130 | Second, instead of sampling over multiple rounds and weighting according to the loss of each skill, we study the effect of setting T = 1, which only uses a softmax on A to yield static weights on the skills. We refer to this approach as âStaticâ. We omit results on Natural Instructions continual pre-training, since SKILL-IT uses T = 1 and using no graph with T = 1 recovers skill-stratiï¬ed sampling. Intuitively, we expect the static version of SKILL-IT to perform somewhat well unless there is signiï¬cant discrepancy among the losses (e.g. in synthetics where the loss on one skill can be close to 0 while the other is not, versus in Natural Instructions where all losses decrease consistently). For both ablations, we sweep over values of η = [0.1, 0.2, 0.5, 0.8]. | 2307.14430#130 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 131 | Figure 26 shows the comparison between SKILL-IT and no graph on the continual pre-training LEGO experiment, and Figure 27 shows the comparison between SKILL-IT and a static approach. We see that both the graph and the online dynamics of SKILL-IT are important for its performance. In particular, using no graph results in allocating signiï¬cant weight to harder skills early on, even though many of them have easier prerequisite skills (such as skill 3 having edges to skills 1 and 2). Using a static graph results in consistent allocation of signiï¬cant weight to prerequisite skills even after their validation losses converge to near 0, and thus the harder skills that have higher loss are not learned quickly afterwards.
We perform the same ablation on the Addition datasetâthe results for this are shown in Figures 28 and Figure 29. We ï¬nd that these simple baselines, including using a static graph and no graph perform similarly to SKILL-IT on average across all skillsâwhile SKILL-IT performs the best on skill 2 compared to vanilla multiplicative weights, and SKILL-IT performs the best on skill 1 compared to a static graph. This suggests that Addition is somewhat easier than the other datasets that we consider, as SKILL-IT still outperforms other baselines, as shown in Figure 4. | 2307.14430#131 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 132 | Figure 30 compares SKILL-IT, no graph, and static data selection for the LEGO ï¬ne-tuning experiment. No graph can be interpreted as allocating equal weight to all training skills not equal to the target skill, and varying this weight versus the weight on the target skill. While SKILL-IT and setting T = 1 behave similarly, we see that SKILL-IT is slightly better than using no graph. For instance, SKILL-IT obtains a validation loss of 0.05 in 2000 steps, compared to 2050-2200 steps when using no graph.
Figure 31 and 32 compare SKILL-IT, no graph, and static data selection for the Natural Instructions ï¬ne-tuning experiments. For both Spanish QG and stance detection, SKILL-IT attains lower loss than using no graph or using T = 1 round.
Figure 33 compares SKILL-IT and static data selection for the Natural Instructions out-of-domain experiment. SKILL-IT
33 | 2307.14430#132 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 133 | Figure 33 compares SKILL-IT and static data selection for the Natural Instructions out-of-domain experiment. SKILL-IT
33
LEGO Skill 1 LEGO Skill 2 LEGO Skill 3 10° 10° Bi04 a 2 10-1 $10 a 1 § 10-3 an 3 10-2 s 10+ 0 2000 4000 6000 0 2000 4000 6000 0 2000 4000 6000 LEGO Skill 4 LEGO Skill 5 Average per skill B Ej 3 2 a 4 1071 1071 g 10-1 No graph (0.1) 3 â No graph (0.2) s â No graph (0.5) 3 â No graph (0.8) â Skill-It 102 10-2 0 2000 4000 6000 0 2000 4000 6000 0 2000 4000 6000 Steps Steps Steps
Figure 26: Comparison of SKILL-IT versus using the identity adjacency matrix (no skills graph) with η = 0.1, 0.2, 0.5, 0.8 on the LEGO continual pre-training experiment. The latter does not capture the relationship between skills, and we ï¬nd that SKILL-IT attains lower loss on all skills. | 2307.14430#133 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 134 | LEGO Skill 1 LEGO Skill 2 LEGO Skill 3 10° SB 8 gd 2 a 8 a g 10-1 3 a > 0 2000. 4000 ~â«6000 0 2000. 4000~â«6000 0 2000 4000 ~â-6000 10° LEGO Skill 4 LEGO Skill 5 Average per skill B 8 gd 2 a 4 10-1 A 2 10-1 10"! Static (0.1) 3 â Static (0.2) Z â Static (0.5) 3 â Static (0.8) â skillt 102 10-2 0 2000. 4000 ~â«6000 0 2000. -4000~â«6000 0 2000 4000 ~â- 6000 Steps Steps Steps
Figure 27: Comparison of SKILL-IT versus using static data selection (T = 1) with η = 0.1, 0.2, 0.5, 0.8 on the LEGO continual pre-training experiment. While SKILL-IT eventually allocates more weights to skills 3, 4, 5, which have higher loss, the static approach is not able to do this. We ï¬nd that SKILL-IT attains lower loss on all skills.
34 | 2307.14430#134 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 135 | 34
Addition Skill 1 Addition Skill 2 Validation Loss (Log) 10° 10-1 10-2 0 2000 4000 Addition Skill 3 6000 0 2000 4000 6000 Average per skill Validation Loss (Log) 10° 107 10-2 â No graph (0.1) â No graph (0.2) â No graph (0.5) â No graph (0.8) â skirt 0 2000 Steps 4000 6000 0 2000 4000 Steps 6000
Figure 28: Comparison of SKILL-IT versus using the identity adjacency matrix (no skills graph) with η = 0.1, 0.2, 0.5, 0.8 on the Addition continual pre-training experiment. The latter does not capture the relationship between skills, and we ï¬nd that SKILL-IT attains lower loss on skill 2, but attains similar performance to methods that do not use the skills graph.
Addition Skill 1 Addition Skill 2 Validation Loss (Log) 10° 107 10-2 2000 4000 Addition Skill 3 6000 0 2000 4000 6000 Average per skill Validation Loss (Log) 10° 107 10-2 â Static (0.1) â Static (0.2) â Static (0.5) â Static (0.8) â skillt 2000 Steps 4000 6000 0 2000 4000 Steps 6000 | 2307.14430#135 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 136 | Figure 29: Comparison of SKILL-IT versus using static data selection (T = 1) with η = 0.1, 0.2, 0.5, 0.8 on the Addition continual pre-training experiment. We ï¬nd that SKILL-IT attains lower loss on skill 1, but attains similar performance to the static methods.
# Validation Loss
° ° ° ° re) is a & 2 ° Performance on Skill 3 Performance on Skill 3 Skill-It No graph (0.1) No graph No graph No graph (0.8) 0.2) ( ( (0.5) ( 0 Steps 1000 2000 3000 4000 5000 6000 â Static (0.1) 0.8 â Static (0.2) o â Static (0.5) 5 0.6 â Static (0.8) a â skill s 3 0.4 g 0.2 0.0 0 1000 2000 3000 4000 5000 6000 Steps
Figure 30: Comparison of SKILL-IT versus using no graph (left) and static data selection (right) with η = 0.1, 0.2, 0.5, 0.8 on the LEGO ï¬ne-tuning experiment. All approaches have roughly the same loss trajectories, but SKILL-IT is slightly lower than using no graph.
35
# Validation Loss | 2307.14430#136 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
2307.14430 | 137 | 35
# Validation Loss
NON ON pe uo uw a o a N ES i} Performance on Spanish QG Performance on Spanish QG 2.60 ââ No graph (0.1) ââ Static (0.1) ââ No graph (0.2) 2.55 ââ Static (0.2) ââ No graph (0.5) g ââ Static (0.5) ââ No graph (0.8) | 2.50 ââ Static (0.8) ââ Skill-It -] ââ Skill-It 2.45 cI zg g 2.40 2.35 200 300 400 500 600 Steps 200 300 400 500 600 Steps
Figure 31: Comparison of SKILL-IT versus using no graph (left) and static data selection (right) with η = 0.1, 0.2, 0.5, 0.8 on the Natural Instructions Spanish QG ï¬ne-tuning experiment. SKILL-IT attains lower validation loss than both no graph and static data selection. | 2307.14430#137 | Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models | The quality of training data impacts the performance of pre-trained large
language models (LMs). Given a fixed budget of tokens, we study how to best
select data that leads to good downstream model performance across tasks. We
develop a new framework based on a simple hypothesis: just as humans acquire
interdependent skills in a deliberate order, language models also follow a
natural order when learning a set of skills from their training data. If such
an order exists, it can be utilized for improved understanding of LMs and for
data-efficient training. Using this intuition, our framework formalizes the
notion of a skill and of an ordered set of skills in terms of the associated
data. First, using both synthetic and real data, we demonstrate that these
ordered skill sets exist, and that their existence enables more advanced skills
to be learned with less data when we train on their prerequisite skills.
Second, using our proposed framework, we introduce an online data sampling
algorithm, Skill-It, over mixtures of skills for both continual pre-training
and fine-tuning regimes, where the objective is to efficiently learn multiple
skills in the former and an individual skill in the latter. On the LEGO
synthetic in the continual pre-training setting, Skill-It obtains 36.5 points
higher accuracy than random sampling. On the Natural Instructions dataset in
the fine-tuning setting, Skill-It reduces the validation loss on the target
skill by 13.6% versus training on data associated with the target skill itself.
We apply our skills framework on the recent RedPajama dataset to continually
pre-train a 3B-parameter LM, achieving higher accuracy on the LM Evaluation
Harness with 1B tokens than the baseline approach of sampling uniformly over
data sources with 3B tokens. | http://arxiv.org/pdf/2307.14430 | Mayee F. Chen, Nicholas Roberts, Kush Bhatia, Jue Wang, Ce Zhang, Frederic Sala, Christopher Ré | cs.CL, cs.LG | null | null | cs.CL | 20230726 | 20230726 | [
{
"id": "2101.00027"
},
{
"id": "2005.14165"
}
] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.