arxiv_id
stringlengths 9
12
| abstract
stringlengths 431
13.8k
|
|---|---|
2403.13043 | Scaling up the size of vision models has been thede factostandard to obtain more powerful visual representations. In this work, we discuss the point beyond which larger vision models arenotnecessary. First, we demonstrate the power ofScaling onScales (S2) , whereby a pre-trained and frozen smaller vision model (e.g., ViT-B or ViT-L) , run over multiple image scales, can outperform larger models (e.g., ViT-H or ViT-G) on classification, segmentation, depth estimation, Multimodal LLM (MLLM) benchmarks, and robotic manipulation. Notably, S2achieves state-of-the-art performance in detailed understanding of MLLM on the V∗benchmark, surpassing models such as GPT-4V. We examine the conditions under which S2is a preferred scaling approach compared to scaling on model size. While larger models have the advantage of better generalization on hard examples, we show that features of larger vision models can be well approximated by those of multi-scale smaller models. This suggests most, if not all, of the representations learned by current large pre-trained models can also be obtained from multi-scale smaller models. Our results show that a multi-scale smaller model has comparable learning capacity to a larger model, and pre-training smaller models with S2can match or even exceed the advantage of larger models. We release a Python package that can apply S2on any vision model withone line of code:https://github.com/bfshi/scaling_on_scales. |
2212.06727 | Vision transformers (ViTs) are quickly becoming the de-facto architecture for computer vision, yet we understand very little about why they work and what they learn. While existing studies visually analyze the mechanisms of convolutional neural networks, an analogous exploration of ViTs remains challenging. In this paper, we first address the obstacles to performing visualizations on ViTs. Assisted by these solutions, we observe that neurons in ViTs trained with language model supervision (e.g., CLIP) are activated by semantic concepts rather than visual features. We also explore the underlying differences between ViTs and CNNs, and we find that transformers detect image background features, just like their convolutional counterparts, but their predictions depend far less on high-frequency information. On the other hand, both architecture types behave similarly in the way features progress from abstract patterns in early layers to concrete objects in late layers. In addition, we show that ViTs maintain spatial information in all layers except the final layer. In contrast to previous works, we show that the last layer most likely discards the spatial information and behaves as a learned global pooling operation. Finally, we conduct large-scale visualizations on a wide range of ViT variants, including DeiT, CoaT, ConViT, PiT, Swin, and Twin, to validate the effectiveness of our method. |
2403.17887 | We
empirically study
a simple layer-pruning strategy
for
popular
families
of open-weight pretrained LLMs,
finding minimal degradation of performance on
different
question-answering benchmarks until
after
a large fraction (up to half) of the layers
are removed.
To prune these models,
we identify the optimal block of layers to prune
by considering
similarity
across
layers;
then,
to “heal” the damage, we
perform a small amount of finetuning.
In particular, we use
parameter-efficient finetuning (PEFT) methods,
specifically quantization and Low Rank Adapters (QLoRA) ,
such that each of our experiments can be performed on a single A100 GPU.
From a practical perspective, these results suggest that layer pruning methods can complement other PEFT strategies to further reduce computational resources of finetuning on the one hand,
and can improve the
memory and latency of inference on the other hand.
From a scientific perspective,
the robustness of these LLMs to the
deletion of
layers
implies either that current pretraining methods
are
not properly
leveraging
the
parameters in the deeper layers of the network
or that the shallow layers play a critical role in storing knowledge. |
2403.19154 | When prompting language models to complete a task, users often leave important aspects unsaid. While asking questions could resolve this ambiguity, models often struggle to ask good questions. We explore a language model’s ability to self-improveby rewarding the model for generating useful questions—a simple method we dub STaR-GATE. We generate a synthetic dataset of 25,500 unique persona-task prompts to simulate conversations between a pretrained language model—theQuestioner—and aRoleplayerwhose preferences are unknown to theQuestioner. By asking questions, theQuestionerelicits preferences from theRoleplayer. TheQuestioneris iteratively finetuned on questions that increase the probability of high-quality responses to the task, which are generated by anOraclewith access to theRoleplayer’s latent preferences. After two iterations of self-improvement, theQuestionerasks better questions, allowing it to generate responses that are preferred over responses from the initial model on72%of tasks. Our results indicate that teaching a language model to ask better questions leads to better personalized responses. |
2210.13468 | Despite the outstanding performance of deep neural networks in different applications, they remain computationally expensive and require a great amount of memory. This motivates more research on reducing the resources needed for implementing such networks. An efficient approach addressed for this purpose is matrix factorization, which has been shown to be effective on different networks. In this paper, we design a training algorithm which utilizes binary matrix factorization and show its efficiency in reducing the required resources in deep neural networks. In effect, this technique leads to the fast and practical implementation of such networks. |
2401.02415 | Humans generally acquire new skills without compromising the old; however, the opposite holds for Large Language Models (LLMs) ,e.g., from LLaMA to CodeLLaMA.
To this end, we propose a new post-pretraining method for LLMs with an expansion of Transformer blocks.
We tune the expanded blocks using only new corpus, efficiently and effectively improving the model’s knowledge without catastrophic forgetting.
In this paper, we experiment on the corpus of code and math, yieldingLLaMA Pro-8.3B, a versatile foundation model initialized from LLaMA2-7B, excelling in general tasks, programming, and mathematics.LLaMA Proand its instruction-following counterpart (LLaMA Pro - Instruct) achieve advanced performance among various benchmarks,
demonstrating superiority over existing open models in the LLaMA family and the immense potential of reasoning and addressing diverse tasks as an intelligent agent.
Our findings provide valuable insights into integrating natural and programming languages, laying a solid foundation for developing advanced language agents that operate effectively in various environments. |
2402.15449 | Recent approaches to improving the extraction of text embeddings from autoregressive large language models (LLMs) have largely focused on improvements to data, backbone pretrained language models, or improving task-differentiation via instructions. In this work, we address an architectural limitation of autoregressive models: token embeddings cannot contain information from tokens that appear later in the input. To address this limitation, we propose a simple approach, “echo embeddings,” in which we repeat the input twice in context and extract embeddings from the second occurrence. We show that echo embeddings of early tokens can encode information about later tokens, allowing us to maximally leverage high-quality LLMs for embeddings. On the MTEB leaderboard, echo embeddings improve over classical embeddings by overzero-shot and by aroundwhen fine-tuned. Echo embeddings with a Mistral-7B model achieve state-of-the-art compared to prior open source models that do not leverage synthetic fine-tuning data.111Our code and pre-trained models are released athttps://github.com/jakespringer/echo-embeddings. |
2402.14848 | This paper explores the impact of extending input lengths on the capabilities of Large Language Models (LLMs) .
Despite LLMs advancements in recent times, their performance consistency across different input lengths is not well understood.
We investigate this aspect by introducing a novel QA reasoning framework, specifically designed to assess the impact of input length.
We isolate the effect of input length using multiple versions of the same sample, each being extended with padding of different lengths, types and locations.
Our findings show a notable degradation in LLMs’ reasoning performance at much shorter input lengths than their technical maximum.
We show that the degradation trend appears in every version of our dataset, although at different intensities.
Additionally, our study reveals that traditional perplexity metrics do not correlate with performance of LLMs’ in long input reasoning tasks.
We analyse our results and identify failure modes that can serve as useful guides for future research, potentially informing strategies to address the limitations observed in LLMs. |
2403.17919 | The machine learning community has witnessed impressive advancements since the first appearance of large language models (LLMs) , yet their huge memory consumption has become a major roadblock to large-scale training. Parameter Efficient Fine-Tuning techniques such as Low-Rank Adaptation (LoRA) have been proposed to alleviate this problem, but their performance still fails to match full parameter training in most large-scale fine-tuning settings. Attempting to complement this deficiency, we investigate layerwise properties of LoRA on fine-tuning tasks and observe an uncommon skewness of weight norms across different layers. Utilizing this key observation, a surprisingly simple training strategy is discovered, which outperforms both LoRA and full parameter training in a wide range of settings with memory costs as low as LoRA. We name itLayerwiseImportanceSampledAdamW (LISA) , a promising alternative for LoRA, which applies the idea of importance sampling to different layers in LLMs and randomly freeze most middle layers during optimization. Experimental results show that with similar or less GPU memory consumption, LISA surpasses LoRA or even full parameter tuning in downstream fine-tuning tasks, where LISA consistently outperforms LoRA by over-in terms of MT-Bench scores. On large models, specifically LLaMA-2-70B, LISA achieves on-par or better performance than LoRA on MT-Bench, GSM8K, and PubMedQA, demonstrating its effectiveness across different domains. |
2311.18828 | Diffusion models generate high-quality images but require dozens of forward passes.
We introduce Distribution Matching Distillation (DMD) , a procedure to transform a diffusion model into a one-step image generator with minimal impact on image quality.
We enforce the one-step image generator match the diffusion model at distribution level, by minimizing an approximate KL divergence whose gradient can be expressed as the difference between 2 score functions, one of the target distribution and the other of the synthetic distribution being produced by our one-step generator. The score functions are parameterized as two diffusion models trained separately on each distribution.
Combined with a simple regression loss matching the large-scale structure of the multi-step diffusion outputs, our method outperforms all published few-step diffusion approaches, reaching 2.62 FID on ImageNet 6464 and 11.49 FID on zero-shot COCO-30k, comparable to Stable Diffusion but orders of magnitude faster.
Utilizing FP16 inference, our model can generate images at 20 FPS on modern hardware. |
2403.10131 | Pretraining Large Language Models (LLMs) on large corpora of textual data is now a standard paradigm.
When using these LLMs for many downstream applications, it is common to additionally bake in new knowledge (e.g., time-critical news, or private domain knowledge) into the pretrained model either through RAG-based-prompting, or finetuning.
However, the optimal methodology for the model to gain such new knowledge remains an open question.
In this paper, we present Retrieval Augmented Fine Tuning (RAFT) , a training recipe that improves the model’s ability to answer questions in an "open-book" in-domain setting. In RAFT, given a question, and a set of retrieved documents, we train the model to ignore those documents that don’t help in answering the question, which we call, distractor documents. RAFT accomplishes this by citing verbatim the right sequence from the relevant document that would help answer the question. This coupled with RAFT’s chain-of-thought-style response helps improve the model’s ability to reason. In domain specific RAG, RAFT consistently improves the model’s performance across PubMed, HotpotQA, and Gorilla datasets, presenting a post-training recipe to improve pre-trained LLMs to in-domain RAG. RAFT’s code and demo are open-sourced athttps://github.com/ShishirPatil/gorilla |
2110.15943 | We introduce MetaICL (Meta-training forIn-ContextLearning) , a new meta-training framework for few-shot learning where a pretrained language model is tuned to do in-context learning on a large set of training tasks.
This meta-training enables the model to more effectively learn a new task in context at test time, by simply conditioning on a few training examples with no parameter updates or task-specific templates.
We experiment on a large, diverse collection of tasks consisting of 142 NLP datasets including classification, question answering, natural language inference, paraphrase detection and more, across seven different meta-training/target splits.
MetaICL outperforms a range of baselines including in-context learning without meta-training and multi-task learning followed by zero-shot transfer.
We find that the gains are particularly significant for target tasks that have domain shifts from the meta-training tasks, and that using a diverse set of the meta-training tasks is key to improvements.
We also show that MetaICL approaches (and sometimes beats) the performance of models fully finetuned on the target task, and outperforms much bigger models with nearly 8x parameters.
Finally, we show that MetaICL is complementary to human-written instructions, and the best performance can be achieved by combining both approaches. |
2402.19364 | We propose a novel approach to iterated sparse matrix dense matrix multiplication, a fundamental computational kernel in scientific computing and graph neural network training. In cases where matrix sizes exceed the memory of a single compute node, data transfer becomes a bottleneck. An approach based on dense matrix multiplication algorithms leads to suboptimal scalability and fails to exploit the sparsity in the problem. To address these challenges, we propose decomposing the sparse matrix into a small number of highly structured matrices calledarrowmatrices, which are connected by permutations. Our approach enables communication-avoiding multiplications, achieving a polynomial reduction in communication volume per iteration for matrices corresponding to planar graphs and other minor-excluded families of graphs. Our evaluation demonstrates that our approach outperforms a state-of-the-art method for sparse matrix multiplication on matrices with hundreds of millions of rows, offering near-linear strong and weak scaling. |
2402.02622 | The transformer architecture fromis now ubiquitous across application domains, from natural language processing to speech processing and image understanding. We propose DenseFormer, a simple modification to the standard architecture that improves the perplexity of the model without increasing its size—adding a few thousand parameters for large-scale models in the 100B parameters range. Our approach relies on an additional averaging step after each transformer block, which computes a weighted average of current and past representations—we refer to this operation as Depth-Weighted-Average (DWA) . The learned DWA weights exhibit coherent patterns of information flow, revealing the strong and structured reuse of activations from distant layers. Experiments demonstrate that DenseFormer is more data efficient, reaching the same perplexity of much deeper transformer models, and that for the same perplexity, these new models outperform transformer baselines in terms of memory efficiency and inference time. |
2006.10901 | Scientific workloads have traditionally exploited high levels of sparsity to accelerate computation and reduce memory requirements. While deep neural networks can be made sparse, achieving practical speedups on GPUs is difficult because these applications have relatively moderate levels of sparsity that are not sufficient for existing sparse kernels to outperform their dense counterparts. In this work, we study sparse matrices from deep learning applications and identify favorable properties that can be exploited to accelerate computation. Based on these insights, we develop high-performance GPU kernels for two sparse matrix operations widely applicable in neural networks: sparse matrix–dense matrix multiplication and sampled dense–dense matrix multiplication. Our kernels reach 27% of single-precision peak on Nvidia V100 GPUs. Using our kernels, we demonstrate sparse Transformer and MobileNet models that achieve 1.2–2.1speedups and up to 12.8memory savings without sacrificing accuracy. |
2301.03598 | We introduceStream-K, a work-centric parallelization of matrix
multiplication (GEMM) and related computations in dense linear algebra.
Whereas contemporary decompositions are primarily tile-based, our method
operates by partitioning an even share of the aggregate inner loop
iterations among physical processing elements. This provides a
near-perfect utilization of computing resources, regardless of how
efficiently the output tiling for any given problem quantizes across the
underlying processing elements. On GPU processors, ourStream-Kparallelization of GEMM produces
a peak speedup of up to 14and 6.7, and an average performance response that is both higher and more consistent
across 32,824 GEMM problem geometries than state-of-the-art
math libraries such as CUTLASS and cuBLAS. Furthermore, we achieve this performance
from asingletile size configuration per floating-point precision, whereas today’s math libraries
employ complex kernel-selection heuristics to select from a large
ensemble of kernel variants. |
2306.11975 | Deep learning hardware achieves high throughput and low power consumption by reducing computing precision and specializing in matrix multiplication.
For machine learning inference, fixed-point value computation is commonplace, where the input and output values and the model parameters are quantized.
Thus, many processors are now equipped with fast integer matrix multiplication units (IMMU) .
It is of significant interest to find a way to harness these IMMUs to improve the performance of HPC applications while maintaining accuracy.
We focus on the Ozaki scheme, which computes a high-precision matrix multiplication by using lower-precision computing units, and show the advantages and disadvantages of using IMMU.
The experiment using integer Tensor Cores shows that we can compute double-precision matrix multiplication faster than cuBLAS and an existing Ozaki scheme implementation on FP16 Tensor Cores on NVIDIA consumer GPUs.
Furthermore, we demonstrate accelerating a quantum circuit simulation by up to 4.33 while maintaining the FP64 accuracy. |
2311.03099 | In this paper, we unveil that Language Models (LMs) can acquire new capabilities by assimilating parameters from homologous models without retraining or GPUs. We first introduce DARE to set most delta parameters (i.e., the disparity between fine-tuned and pre-trained parameters) to zeros without affecting the abilities of Supervised Fine-Tuning (SFT) LMs, which randomlyDrops delta parameters with a ratioAndREscales the remaining ones byto approximate the original embeddings. Then, we use DARE as a versatile plug-and-play technique to sparsify delta parameters of multiple SFT homologous models for mitigating parameter interference and merge them into a single model by parameter fusing. We experiment with encoder- and decoder-based LMs, showing that: (1) SFT delta parameter value ranges are typically small (within 0.005) with extreme redundancy, and DARE can effortlessly eliminate 90% or even 99% of them. (2) DARE can merge multiple task-specific LMs into one LM with diverse capabilities. For instance, the amalgamation of WizardLM and WizardMath significantly enhances the GSM8K zero-shot accuracy of WizardLM from 2.2 to 66.3, retaining the instruction-following proficiency while surpassing WizardMath’s 64.2 performance. Our merged LM also ranks first among models with 7 billion parameters on the Open LLM Leaderboard. |
2203.14680 | Transformer-based language models (LMs) are at the core of modern NLP, but their internal prediction construction process is opaque and largely not understood. In this work, we make a substantial step towards unveiling this underlying prediction process, by reverse-engineering the operation of the feed-forward network (FFN) layers, one of the building blocks of transformer models.
We view the token representation as a changing distribution over the vocabulary, and the output from each FFN layer as an additive update to that distribution. Then, we analyze the FFN updates in the vocabulary space, showing that each update can be decomposed to sub-updates corresponding to single FFN parameter vectors, each promoting concepts that are often human-interpretable.
We then leverage these findings for controlling LM predictions, where we reduce the toxicity ofGPT2by almost 50%, and for improving computation efficiency with a simple early exit rule, saving 20% of computation on average.111Our codebase is available athttps://github.com/aviclu/ffn-values. |
2212.04089 | Changing how pre-trained models behave—e.g., improving their performance on a downstream task or mitigating biases learned during pre-training—is a common practice when developing machine learning systems.
In this work, we propose a new paradigm for steering the behavior of neural networks, centered aroundtask vectors.
A task vector specifies a direction in the weight space of a pre-trained model, such that movement in that direction improves performance on the task.
We build task vectors by subtracting the weights of a pre-trained model from the weights of the same model after fine-tuning on a task.
We show that these task vectors can be modified and combined together through arithmetic operations such as negation and addition, and the behavior of the resulting model is steered accordingly.
Negating a task vector decreases performance on the target task, with little change in model behavior on control tasks.
Moreover, adding task vectors together can improve performance on multiple tasks at once.
Finally, when tasks are linked by an analogy relationship of the form “is toasis to”, combining task vectors from three of the tasks can improve performance on the fourth, even when no data from the fourth task is used for training.
Overall, our experiments with several models, modalities and tasks show that task arithmetic is a simple, efficient and effective way of editing models. |
2306.01708 | Transfer learning – i.e., further fine-tuning a pre-trained model on a downstream task – can confer significant advantages, including improved downstream performance, faster convergence, and better sample efficiency. These advantages have led to a proliferation of task-specific fine-tuned models, which typically can only perform a single task and do not benefit from one another. Recently, model merging techniques have emerged as a solution to combine multiple task-specific models into a single multitask model without performing additional training. However, existing merging methods often ignore the interference between parameters of different models, resulting in large performance drops when merging multiple models. In this paper, we demonstrate that prior merging techniques inadvertently lose valuable information due to two major sources of interference: (a) interference due to redundant parameter values and (b) disagreement on the sign of a given parameter’s values across models. To address this, we propose our method,TrIm, Elect Sign & Merge(Ties-Merging) , which introduces three novel steps when merging models: (1) resetting parameters that only changed a small amount during fine-tuning, (2) resolving sign conflicts, and (3) merging only the parameters that are in alignment with the final agreed-upon sign.
We find thatTies-Mergingoutperforms several existing methods in diverse settings covering a range of modalities, domains, number of tasks, model sizes, architectures, and fine-tuning settings. We further analyze the impact of different types of interference on model parameters, and highlight the importance of resolving sign interference.111Our code is available athttps://github.com/prateeky2806/ties-merging |
2403.13187 | We present a novel application of evolutionary algorithms to automate the creation of powerful foundation models. While model merging has emerged as a promising approach for LLM development due to its cost-effectiveness, it currently relies on human intuition and domain knowledge, limiting its potential. Here, we propose an evolutionary approach that overcomes this limitation by automatically discovering effective combinations of diverse open-source models, harnessing their collective intelligence without requiring extensive additional training data or compute. Our approach operates in both parameter space and data flow space, allowing for optimization beyond just the weights of the individual models. This approach even facilitates cross-domain merging, generating models like a Japanese LLM with Math reasoning capabilities. Surprisingly, our Japanese Math LLM achieved state-of-the-art performance on a variety of established Japanese LLM benchmarks, even surpassing models with significantly more parameters, despite not being explicitly trained for such tasks. Furthermore, a culturally-aware Japanese VLM generated through our approach demonstrates its effectiveness in describing Japanese culture-specific content, outperforming previous Japanese VLMs. This work not only contributes new state-of-the-art models back to the open-source community, but also introduces a new paradigm for automated model composition, paving the way for exploring alternative, efficient approaches to foundation model development.111EvoLLM-JP,EvoVLM-JPrelease:https://github.com/SakanaAI/evolutionary-model-merge |
2402.16153 | While Large Language Models (LLMs) demonstrate impressive capabilities in text generation, we find that their ability has yet to be generalized to music, humanity’s creative language.
We introduceChatMusician111See Contributions and Acknowledgments section for full author list., an open-source LLM that integrates intrinsic musical abilities.
It is based on continual pre-training and finetuning LLaMA2 on a text-compatible music representation, ABC notation, and the music is treated as a second language.
ChatMusician can understand and generate music with a pure text tokenizer without any external multi-modal neural structures or tokenizers.
Interestingly, endowing musical abilities does not harm language abilities, even achieving a slightly higher MMLU score.
Our model is capable of composing well-structured, full-length music, conditioned on texts, chords, melodies, motifs, musical forms, etc, surpassing GPT-4 baseline.
On our meticulously curated college-level music understanding benchmark,MusicTheoryBench, ChatMusician surpasses LLaMA2 and GPT-3.5 on zero-shot setting by a noticeable margin.
Our work reveals that LLMs can be an excellent compressor for music, but there remains significant territory to be conquered.
We release our 4B token music-language corporaMusicPile, the collected MusicTheoryBench, code, model and demo inGitHub. |
2203.14465v2 | Generating step-by-step "chain-of-thought" rationales improves language model performance on complex reasoning tasks like mathematics or commonsense question-answering. However, inducing language model rationale generation currently requires either constructing massive rationale datasets or sacrificing accuracy by using only few-shot inference.
We propose a technique to iteratively leverage a small number of rationale examples and a large dataset without rationales, to bootstrap the ability to perform successively more complex reasoning.
This technique, the "Self-Taught Reasoner" (STaR) , relies on a simple loop: generate rationales to answer many questions, prompted with a few rationale examples; if the generated answers are wrong,
try again to generate a rationale given the correct answer; fine-tune on all the rationales that ultimately yielded correct answers; repeat.
We show that STaR significantly improves performance on multiple datasets compared to a model fine-tuned to directly predict final answers, and performs comparably to fine-tuning a 30larger state-of-the-art language model on CommensenseQA. Thus, STaR lets a model improve itself by learning from its own generated reasoning. |
2402.10171 | We study the continual pretraining recipe for scaling language models’ context lengths to 128K, with a focus on data engineering.
We hypothesize that long context modeling, in particularthe ability to utilize information at arbitrary input locations, is a capability that is
mostly already acquired through large-scale pretraining, and that this capability can be readily extended to contexts substantially longer than seen during training (e.g., 4K to 128K) through lightweight continual pretraining on appropriate data mixture.
We investigate thequantityandqualityof the data for continual pretraining:
(1) for quantity, we show that 500 million to 5 billion tokens are enough to enable the model to retrieve information anywhere within the 128K context;
(2) for quality, our results equally emphasizedomain balanceandlength upsampling.
Concretely, we find that naïvely upsampling longer data on certain domains like books, a common practice of existing work, gives suboptimal performance, and that a balanced domain mixture is important.
We demonstrate that continual pretraining of the full model on 1B-5B tokens of such data is an effective and affordable strategy for scaling the context length of language models to 128K.
Our recipe outperforms strong open-source long-context models
and closes the gap to frontier models like GPT-4 128K. |
2403.08763 | Large language models (LLMs) are routinely pre-trained on billions of tokens, only to start the process over again once new data becomes available. A much more efficient solution is to continually pre-train these models—saving significant compute compared to re-training. However, the distribution shift induced by new data typically results in degraded performance on previous data or poor adaptation to the new data. In this work, we show that a simple and scalable combination of learning rate (LR) re-warming, LR re-decaying, and replay of previous data is sufficient to match the performance of fully re-training from scratch on all available data, as measured by the final loss and the average score on several language model (LM) evaluation benchmarks. Specifically, we show this for a weak but realistic distribution shift between two commonly used LLM pre-training datasets (English→English) and a stronger distribution shift (English→German) at the 405M parameter model scale with large dataset sizes (hundreds of billions of tokens). Selecting the weak but realistic shift for larger-scale experiments, we also find that our continual learning strategies match the re-training baseline for a 10B parameter LLM. Our results demonstrate that autoregressive transformer-based LLMs can be successfully updated via simple and scalable continual learning strategies, matching the re-training baseline using only a fraction of the compute. Finally, inspired by previous work, we propose alternatives to the cosine learning rate schedule that help circumvent forgetting induced by LR re-warming and that are not bound to a fixed token budget. |
2403.07691 | While recent preference alignment algorithms for language models have demonstrated promising results, supervised fine-tuning (SFT) remains imperative for achieving successful convergence. In this paper, we study the crucial role of SFT within the context of preference alignment, emphasizing that a minor penalty for the disfavored generation style is sufficient for preference-aligned SFT. Building on this foundation, we introduce a straightforward and innovative reference model-free monolithic odds ratio preference optimization algorithm, ORPO, eliminating the necessity for an additional preference alignment phase. We demonstrate, both empirically and theoretically, that the odds ratio is a sensible choice for contrasting favored and disfavored styles during SFT across the diverse sizes from 125M to 7B. Specifically, fine-tuning Phi-2 (2.7B) , Llama-2 (7B) , and Mistral (7B) with ORPO on the UltraFeedback alone surpasses the performance of state-of-the-art language models with more than 7B and 13B parameters: achieving up to 12.20% on(Figure1) , 66.19% on IFEval (instruction-level loose, Table6) , and 7.32 in MT-Bench (Figure12) . We release code111https://github.com/xfactlab/orpoand model checkpoints for Mistral-ORPO-(7B) 222https://huggingface.co/kaist-ai/mistral-orpo-alphaand Mistral-ORPO-(7B) .333https://huggingface.co/kaist-ai/mistral-orpo-beta |
2403.09629 | When writing and talking, people sometimes pause to think. Although reasoning-focused works have often framed reasoning as a method of answering questions or completing agentic tasks, reasoning is implicit in almost all written text. For example, this applies to the steps not stated between the lines of a proof or to the theory of mind underlying a conversation. In the Self-Taught Reasoner (STaR, Zelikman et al. 2022) , useful thinking is learned by inferring rationales from few-shot examples in question-answering and learning from those that lead to a correct answer. This is a highly constrained setting – ideally, a language model could instead learn to infer unstated rationales in arbitrary text. We presentQuiet-STaR, a generalization of STaR in which LMs learn to generate rationales at each token to explain future text, improving their predictions. We address key challenges, including 1) the computational cost of generating continuations, 2) the fact that the LM does not initially know how to generate or use internal thoughts, and 3) the need to predict beyond individual next tokens. To resolve these, we propose a tokenwise parallel sampling algorithm, using learnable tokens indicating a thought’s start and end, and an extended teacher-forcing technique. Encouragingly, generated rationales disproportionately help model difficult-to-predict tokens and improve the LM’s ability to directly answer difficult questions. In particular, after continued pretraining of an LM on a corpus of internet text with Quiet-STaR, we find zero-shot improvements on GSM8K (5.9%10.9%) and CommonsenseQA (36.3%47.2%) and observe a perplexity improvement of difficult tokens in natural text. Crucially, these improvements require no fine-tuning on these tasks. Quiet-STaR marks a step towards LMs that can learn to reason in a more general and scalable way. |
2305.17888 | Several post-training quantization methods have been applied to large language models (LLMs) , and have been shown to perform well down to 8-bits. We find that these methods break down at lower bit precision, and investigate quantization aware training for LLMs (LLM-QAT) to push quantization levels even further. We propose a data-free distillation method that leverages generations produced by the pre-trained model, which better preserves the original output distribution and allows quantizing any generative model independent of its training data, similar to post-training quantization methods. In addition to quantizing weights and activations, we also quantize the KV cache, which is critical for increasing throughput and support long sequence dependencies at current model sizes. We experiment with LLaMA models of sizes 7B, 13B, and 30B, at quantization levels down to 4-bits. We observe large improvements over training-free methods, especially in the low-bit settings. |
2403.08715 | Humans learn social skills through both imitation and social interaction. This social learning process is largely understudied by existing research on building language agents. Motivated by this gap, we propose an interactive learning method,\trjnfamilySOTOPIA-, improving the social intelligence of language agents. This method leverages behavior cloning and self-reinforcement training on filtered social interaction data according to large language model (LLM) ratings.
We show that our training method allows a 7B LLM to reach the social goal completion ability of an expert model (GPT-4-based agent) , while improving the safety of language agents and maintaining general QA ability on the MMLU benchmark.
We also find that this training paradigm uncovers some difficulties in LLM-based evaluation of social intelligence: LLM-based evaluators overestimate the abilities of the language agents trained specifically for social interaction. |
2403.08245 | We present ScatterMoE, an implementation of Sparse Mixture-of-Experts (SMoE) on GPUs.
ScatterMoE builds upon existing implementations, and overcoming some of the limitations to improve inference and training speed, and memory footprint.
This implementation achieves this by avoiding padding and making excessive copies of the input. We introduceParallelLinear, the main component we use to build our implementation and the various kernels used to speed up the operation.
We benchmark our implementation against Megablocks, and show that it enables a higher throughput and lower memory footprint.
We also show howParallelLinearenables extension of the Mixture-of-Experts concept by demonstrating with an implementation of Mixture of Attention. https://github.com/shawntan/scattermoe |
2403.03864 | This paper introduces the novel task of multimodal puzzle solving, framed within the context of visual question-answering. We present a new dataset,AlgoPuzzleVQAdesigned to challenge and evaluate the capabilities of multimodal language models in solving algorithmic puzzles that necessitate both visual understanding, language understanding, and complex algorithmic reasoning. We create the puzzles to encompass a diverse array of mathematical and algorithmic topics such as boolean logic, combinatorics, graph theory, optimization, search, etc., aiming to evaluate the gap between visual data interpretation and algorithmic problem-solving skills. The dataset is generated automatically from code authored by humans. All our puzzles have exact solutions that can be found from the algorithm without tedious human calculations. It ensures that our dataset can be scaled up arbitrarily in terms of reasoning complexity and dataset size. Our investigation reveals that large language models (LLMs) such as GPT4V and Gemini exhibit limited performance in puzzle-solving tasks. We find that their performance is near random in a multi-choice question-answering setup for a significant number of puzzles. The findings emphasize the challenges of integrating visual, language, and algorithmic knowledge for solving complex reasoning problems. |
2402.13991 | Most language model pre-training frameworks concatenate multiple documents into fixed-length sequences and usecausal maskingto compute the likelihood of each token given its context; this strategy is widely adopted due to its simplicity and efficiency.
However, to this day, the influence of the pre-training sequence composition strategy on the generalisation properties of the model remains under-explored.
In this work, we find that applying causal masking can lead to the inclusion of distracting information from previous documents during pre-training, which negatively impacts the performance of the models on language modelling and downstream tasks.
Inintra-document causal masking, the likelihood of each token is only conditioned on the previous tokens in the same document, eliminating potential distracting information from previous documents and significantly improving performance.
Furthermore, we find that concatenating related documents can reduce some potential distractions during pre-training, and our proposed efficient retrieval-based sequence construction method,Bm25Chunk, can improve in-context learning (+11.6%) , knowledge memorisation (+9.8%) , and context utilisation (+7.2%) abilities of language models without sacrificing efficiency.00footnotetext:https://github.com/yuzhaouoe/pretraining-data-packing |
2303.06349 | Recurrent Neural Networks (RNNs) offer fast inference on long sequences but are hard to optimize and slow to train. Deep state-space models (SSMs) have recently been shown to perform remarkably well on long sequence modeling tasks, and have the added benefits of fast parallelizable training and RNN-like fast inference. However, while SSMs are superficially similar to RNNs, there are important differences that make it unclear where their performance boost over RNNs comes from.
In this paper, we show that careful design of deep RNNs using standard signal propagation arguments can recover the impressive performance of deep SSMs on long-range reasoning tasks, while also matching their training speed.
To achieve this, we analyze and ablate a series of changes to standard RNNs including linearizing and diagonalizing the recurrence, using better parameterizations and initializations, and ensuring proper normalization of the forward pass. Our results provide new insights on the origins of the impressive performance of deep SSMs, while also introducing an RNN block called the Linear Recurrent Unit that matches both their performance on theLong Range Arenabenchmark and their computational efficiency. |
2402.18679 | Large Language Model (LLM)-based agents have demonstrated remarkable effectiveness. However, their performance can be compromised in data science scenarios that require real-time data adjustment, expertise in optimization due to complex dependencies among various tasks, and the ability to identify logical errors for precise reasoning. In this study, we introduce the Data Interpreter, a solution designed to solve with code that emphasizes three pivotal techniques to augment problem-solving in data science: 1) dynamic planning with hierarchical graph structures for real-time data adaptability; 2) tool integration dynamically to enhance code proficiency during execution, enriching the requisite expertise; 3) logical inconsistency identification in feedback, and efficiency enhancement through experience recording. We evaluate the Data Interpreter on various data science and real-world tasks. Compared to open-source baselines, it demonstrated superior performance, exhibiting significant improvements in machine learning tasks, increasing from 0.86 to 0.95. Additionally, it showed a 26% increase in the MATH dataset and a remarkable 112% improvement in open-ended tasks. The solution will be released at https://github.com/geekan/MetaGPT. |
2106.08295 | While neural networks have advanced the frontiers in many applications, they often come at a high computational cost. Reducing the power and latency of neural network inference is key if we want to integrate modern networks into edge devices with strict power and compute requirements. Neural network quantization is one of the most effective ways of achieving these savings but the additional noise it induces can lead to accuracy degradation. In this white paper, we introduce state-of-the-art algorithms for mitigating the impact of quantization noise on the network’s performance while maintaining low-bit weights and activations. We start with a hardware motivated introduction to quantization and then consider two main classes of algorithms: Post-Training Quantization (PTQ) and Quantization-Aware-Training (QAT) . PTQ requires no re-training or labelled data and is thus a lightweight push-button approach to quantization. In most cases, PTQ is sufficient for achieving 8-bit quantization with close to floating-point accuracy. QAT requires fine-tuning and access to labeled training data but enables lower bit quantization with competitive results. For both solutions, we provide tested pipelines based on existing literature and extensive experimentation that lead to state-of-the-art performance for common deep learning models and tasks. |
2403.07816 | We investigate efficient methods for training Large Language Models (LLMs) to possess capabilities in multiple specialized domains, such as coding, math reasoning and world knowledge. Our method, named Branch-Train-MiX (BTX) , starts from a seed model, which is branched to train experts in embarrassingly parallel fashion with high throughput and reduced communication cost. After individual experts are asynchronously trained, BTX brings together their feedforward parameters as experts in Mixture-of-Expert (MoE) layers and averages the remaining parameters, followed by an MoE-finetuning stage to learn token-level routing. BTX generalizes two special cases, the Branch-Train-Merge method, which does not have the MoE finetuning stage to learn routing, and sparse upcycling, which omits the stage of training experts asynchronously. Compared to alternative approaches, BTX achieves the best accuracy-efficiency tradeoff. |
2402.12374 | As the usage of large language models (LLMs) grows, performing efficient inference with these models becomes increasingly important.
While speculative decoding has recently emerged as a promising direction for speeding up inference, existing methods are limited in their ability to scale to larger speculation budgets, and adapt to different hyperparameters and hardware.
This paper introducesSequoia, a scalable, robust, and hardware-aware algorithm for speculative decoding.
To attain better scalability,Sequoiaintroduces a dynamic programming algorithm to find the optimal tree structure for the speculated tokens.
To achieve robust speculative performance,Sequoiauses a novel sampling and verification method that outperforms prior work across different decoding temperatures.
Finally,Sequoiaintroduces a hardware-aware tree optimizer that maximizes speculative performance by automatically selecting the token tree size and depth for a given hardware platform.Sequoiaimproves the decoding speed of Llama2-7B, Llama2-13B, and Vicuna-33B on an A100 GPU by up to,, and. For offloading setting on L40,Sequoiaachieves as low as 0.56 s/token for exact Llama2-70B inference latency, which ison our optimized offloading system (5.6 s/token) ,than DeepSpeed-Zero-Inference,than Huggingface Accelerate. The code is available athttps://github.com/Infini-AI-Lab/Sequoia. |
2403.06504 | Recent advances in large language models have brought immense value to the world, with their superior capabilities stemming from the massive number of parameters they utilize. However, even the GPUs with the highest memory capacities, currently peaking at 80GB, are far from sufficient to accommodate these vast parameters and their associated optimizer states when conducting stochastic gradient descent-based optimization.
One approach to hosting such huge models is to aggregate device memory from many GPUs. In particular, it takes 32 NVIDIA A100 GPUs to fit a model with 100 billion parameters for fine-tuning. However, this approach introduces prohibitive costs for most academic researchers, who always have a limited budget for many high-end GPU servers. In this paper, we focus on huge model fine-tuning on a single, even low-end, GPU in a commodity server, which is accessible to most AI researchers.
In such a scenario, the state-of-the-art work ZeRO-Infinity suffers from two severe issues when running in a commodity server: 1) low GPU utilization due to inefficient swapping, and 2) limited trainable model size due to CPU memory capacity. The underlying reason is that ZeRO-Infinity is optimized for running on high-end GPU servers. To this end, we presentFuyou, a low-cost training framework that enables efficient 100B huge model fine-tuning on a low-end server with a low-end GPU and limited CPU memory capacity. The key idea is to add the SSD-CPU communication as an optimization dimension and thus carefully co-optimize computation and data swapping from a systematic approach to maximize GPU utilization. To do so, Fuyou consists of three innovations. First, we propose a synchronous out-of-core CPU optimizer that overlaps with backward propagation to maximize the GPU utilization. Second, we propose a GPU-CPU-SSD fully-pipelined activation swapping mechanism to allow for a significantly larger model fine-tuning. Third, we present an automatic activation swapping management to automatically determine the optimal amount of swapping activations so as to minimize the epoch time.
The experimental results show that 1) Fuyou is able to fine-tune 175B GPT-3 on a consumer GPU RTX 4090 with high GPU utilization, while ZeRO-Infinity fails to fine-tune; and 2) when training a small GPT-3 13B model, Fuyou achieves 156 TFLOPS on an RTX 4090 GPU while ZeRO-Infinity only achieves 45 TFLOPS. |
2403.04652 | We introduce the Yi model family, a series of language and multimodal models that demonstrate strong multi-dimensional capabilities.
The Yi model family is based on 6B and 34B pretrained language models, then we extend them to chat models, 200K long context models, depth-upscaled models, and vision-language models.
Our base models achieve strong performance on a wide range of benchmarks like MMLU,
and our finetuned chat models deliver strong human preference rate on major evaluation platforms like AlpacaEval and Chatbot Arena.
Building upon our scalable super-computing infrastructure and the classical transformer architecture,
we attribute the performance of Yi models primarily to its data quality resulting from our data-engineering efforts.
For pretraining, we construct 3.1 trillion tokens of English and Chinese corpora using a cascaded data deduplication and quality filtering pipeline.
For finetuning, we polish a small scale (less than 10K) instruction dataset over multiple iterations such that every single instance has been verified directly by our machine learning engineers.
For vision-language, we combine the chat language model with a vision transformer encoder and train the model to align visual representations to the semantic space of the language model.
We further extend the context length to 200K through lightweight continual pretraining and demonstrate strong needle-in-a-haystack retrieval performance.
We show that extending the depth of the pretrained checkpoint through continual pretraining further improves performance.
We believe that given our current results, continuing to scale up model parameters using thoroughly optimized data will lead to even stronger frontier models. |
2310.16795 | Mixture-of-Experts (MoE) architectures offer a general solution to the high inference costs of large language models (LLMs) via sparse routing, bringing faster and more accurate models, at the cost of massive parameter counts. For example, the SwitchTransformer-c2048 model
has 1.6 trillion parameters, requiring 3.2TB of accelerator memory to run efficiently, which makes practical deployment challenging and expensive. In this paper, we present a solution to this memory problem, in form of a new compression and execution framework called QMoE. Specifically, QMoE consists of a scalable algorithm which accurately compresses trillion-parameter MoEs to less than 1 bit per parameter, in a custom format co-designed with bespoke GPU decoding kernels to facilitate efficient end-to-end compressed inference, with minor runtime overheads relative to uncompressed execution. Concretely, QMoE can compress the 1.6 trillion parameter SwitchTransformer-c2048 model to less than 160GB (20x compression, 0.8 bits per parameter) at only minor accuracy loss, in less than a day on a single GPU. This enables, for the first time, the execution of a trillion-parameter model on affordable commodity hardware, like a single server with 4x NVIDIA A6000 or 8x NVIDIA 3090 GPUs, at less than 5% runtime overhead relative to ideal uncompressed inference. The source code and compressed models are available atgithub.com/IST-DASLab/qmoe. |
2305.18411 | We study the effect of width on the dynamics of feature-learning neural networks across a variety of architectures and datasets. Early in training, wide neural networks trained on online data have not only identical loss curves but also agree in their point-wise test predictions throughout training.
For simple tasks such as CIFAR-5m this holds throughout training for networks of realistic widths. We also show that structural properties of the models, including internal representations, preactivation distributions, edge of stability phenomena, and large learning rate effects are consistent across large widths. This motivates the hypothesis that phenomena seen in realistic models can be captured by infinite-width, feature-learning limits. For
harder tasks (such as ImageNet and language modeling) , and later training times, finite-width deviations grow systematically. Two distinct effects cause these deviations across widths. First, the network output has an initialization-dependent variance scaling inversely with width, which can be removed by ensembling networks. We observe, however,
that ensembles of narrower networks perform worse than a single wide network. We call this thebiasof narrower width. We conclude with a spectral perspective on the origin of this finite-width bias. |
2403.03507 | Training Large Language Models (LLMs) presents significant memory challenges, predominantly due to the growing size of weights and optimizer states.
Common memory-reduction approaches, such as low-rank adaptation (LoRA) , add a trainable low-rank matrix to the frozen pre-trained weight in each layer, reducing trainable parameters and optimizer states. However, such approaches typically underperform training with full-rank weights in both pre-training and fine-tuning stages since they limit the parameter search to a low-rank subspace and alter the training dynamics, and further, may require full-rank warm start.
In this work, we propose Gradient Low-Rank Projection (GaLore) , a training strategy that allowsfull-parameterlearning but is morememory-efficientthan common low-rank adaptation methods such as LoRA.
Our approach reduces memory usage by up to 65.5% in optimizer states while maintaining both efficiency and performance for pre-training on LLaMA 1B and 7B architectures with C4 dataset with up to 19.7B tokens, and on fine-tuning RoBERTa on GLUE tasks.
Our 8-bit GaLore further reduces optimizer memory by up to 82.5% and total training memory by 63.3%, compared to a BF16 baseline.
Notably, we demonstrate, for the first time, the feasibility of pre-training a 7B model on consumer GPUs with 24GB memory (e.g., NVIDIA RTX 4090) without model parallel, checkpointing, or offloading strategies. |
2204.07143 | We presentNeighborhood Attention (NA) , the first efficient and scalable sliding window attention mechanism for vision.
NA is a pixel-wise operation, localizing self attention (SA) to the nearest neighboring pixels, and therefore enjoys a linear time and space complexity compared to the quadratic complexity of SA.
The sliding window pattern allows NA’s receptive field to grow without needing extra pixel shifts, and preserves translational equivariance, unlike Swin Transformer’s Window Self Attention (WSA) .
We develop(Neighborhood Attention Extension) , a Python package with efficient C++ and CUDA kernels, which allows NA to run up to 40% faster than Swin’s WSA while using up to 25% less memory.
We further presentNeighborhood Attention Transformer (NAT) , a new hierarchical transformer design based on NA that boosts image classification and downstream vision performance.
Experimental results on NAT are competitive; NAT-Tiny reaches 83.2% top-1 accuracy on ImageNet, 51.4% mAP on MS-COCO and 48.4% mIoU on ADE20K, which is 1.9% ImageNet accuracy, 1.0% COCO mAP, and 2.6% ADE20K mIoU improvement over a Swin model with similar size.
To support more research based on sliding window attention, we open source our project and release our checkpoints. |
2402.14992 | The versatility of large language models (LLMs) led to the creation of diverse benchmarks that thoroughly test a variety of language models’ abilities. These benchmarks consist of tens of thousands of examples making evaluation of LLMs very expensive. In this paper, we investigate strategies to reduce the number of evaluations needed to assess the performance of an LLM on several key benchmarks. For example, we show that to accurately estimate the performance of an LLM on MMLU, a popular multiple-choice QA benchmark consisting of 14K examples, it is sufficient to evaluate this LLM on 100 curated examples. We release evaluation tools and tiny versions of popular benchmarks: Open LLM Leaderboard, MMLU, HELM, and AlpacaEval 2.0.
Our empirical analysis demonstrates that these tools and tiny benchmarks are sufficient to reliably and efficiently reproduce the original evaluation results111To use our methods for efficient LLM evaluation, please checkhttps://github.com/felipemaiapolo/tinyBenchmarks. This repository includes a Python package for model evaluation and tutorials. Additionally, we have uploaded tiny datasets onhuggingface.co/tinyBenchmarksand developed aGoogle Colab demoin which you can easily use our tools to estimate LLM performances on MMLU. To reproduce the results in this paper, please check thisGitHub repository.. |
2402.19472 | Standardized benchmarks drive progress in machine learning.
However, with repeated testing, the risk of overfitting grows as algorithms over-exploit benchmark idiosyncrasies. In our work, we seek to mitigate this challenge by compilingever-expandinglarge-scale benchmarks calledLifelong Benchmarks. As exemplars of our approach, we createLifelong-CIFAR10andLifelong-ImageNet, containing (for now) 1.69M and 1.98M test samples, respectively. While reducing overfitting, lifelong benchmarks introduce a key challenge: the high cost of evaluating a growing number of models across an ever-expanding sample set. To address this challenge, we also introduce an efficient evaluation framework:Sort & Search (S&S) , which reuses previously evaluated models by leveraging dynamic programming algorithms to selectively rank and sub-select test samples, enabling cost-effective lifelong benchmarking. Extensive empirical evaluations across31,000 models demonstrate thatS&Sachieves highly-efficient approximate accuracy measurement, reducing compute cost from 180 GPU days to 5 GPU hours (1000x reduction) on a single A100 GPU, with low approximation error. As such, lifelong benchmarks offer a robust, practical solution to the “benchmark exhaustion” problem. |
2201.12086 | Vision-Language Pre-training (VLP) has advanced the performance for many vision-language tasks.
However,
most existing pre-trained models only excel in either understanding-based tasks or generation-based tasks.
Furthermore,
performance improvement has been largely achieved by scaling up the dataset with noisy image-text pairs collected from the web,
which is a suboptimal source of supervision.
In this paper,
we propose BLIP,
a new VLP framework which transfers flexibly to both vision-language understanding and generation tasks.
BLIP effectively utilizes the noisy web data by bootstrapping the captions, where a captioner generates synthetic captions and a filter removes the noisy ones.
We achieve state-of-the-art results on a wide range of vision-language tasks,
such as image-text retrieval (+2.7% in average recall@1) , image captioning (+2.8% in CIDEr) ,
and VQA (+1.6% in VQA score) .
BLIP also demonstrates strong generalization ability when directly transferred to video-language tasks in a zero-shot manner.
Code, models, and datasets are released. |
2403.02325 | Highlighting particularly relevant regions of an image can improve the performance of vision-language models (VLMs) on various vision-language (VL) tasks by guiding the model to attend more closely to these regions of interest.
For example, VLMs can be given a “visual prompt”, where visual markers such as bounding boxes delineate key image regions; this approach has become popular due to the improvement it provides in tasks requiring region-level information.
However, current VLMs that can incorporate visual guidance are either proprietary and expensive or require costly training on curated data that includes visual prompts.
We introduceContrastive Region Guidance(CRG) ,
a training-free guidance method that enables open-source VLMs to respond to visual prompts.CRGcontrasts model outputs produced with and without visual prompts,
factoring out biases revealed by the model when answeringwithoutthe information required to produce a correct answer (i.e., the model’s prior) .CRGachieves substantial improvements in a wide variety of VL tasks: When region annotations are provided,CRGincreases absolute accuracy by up toon ViP-Bench, a collection of six diverse region-based tasks such as recognition, math, and object relationship reasoning.
We also showCRG’s applicability to spatial reasoning, where we obtain up toimprovement on the hardest setting of What’sUp, as well as to compositional generalization – improving accuracy byandon two challenging splits from SugarCrepe – and to image-text alignment for generated images, where we improve by up toAUROC andF1 points on SeeTRUE.
For cases that do not have reference regions for the prompt, we also show thatCRGallows us to re-rank regions proposed by an object detection model in referring expression comprehension and phrase grounding benchmarks like RefCOCO/RefCOCO+/RefCOCOg and Flickr30K Entities, with an average improvement ofin accuracy when multiple proposals are available. In our analysis, we explore alternative masking strategies forCRG,
demonstrate howCRGimpacts the model’s probability over relevant text phrases, and evaluate the role of the region guidance strength, empirically validatingCRG’s design choices. |
2301.12597 | The cost of vision-and-language pre-training has become increasingly prohibitive due to end-to-end training of large-scale models.
This paper proposes BLIP-2, a generic and efficient pre-training strategy that bootstraps vision-language pre-training from off-the-shelf frozen pre-trained image encoders and frozen large language models.
BLIP-2 bridges the modality gap with a lightweight Querying Transformer, which is pre-trained in two stages.
The first stage bootstraps vision-language representation learning from a frozen image encoder.
The second stage bootstraps vision-to-language generative learning from a frozen language model.
BLIP-2 achieves state-of-the-art performance on various vision-language tasks,
despite having significantly fewer trainable parameters than existing methods.
For example,
our model outperforms Flamingo80B by 8.7% on
zero-shot VQAv2 with 54x fewer trainable parameters.
We also demonstrate the model’s emerging capabilities of zero-shot image-to-text generation that can follow natural language instructions. |
2402.01032 | Transformers are the dominant architecture for sequence modeling, but there is growing interest in models that use a fixed-size latent state that does not depend on the sequence length, which we refer to as “generalized state space models” (GSSMs) .
In this paper we show that while GSSMs are promising in terms of inference-time efficiency, they are limited compared to transformer models on tasks that require copying from the input context.
We start with a theoretical analysis of the simple task of string copying and prove that a two layer transformer can copy strings of exponential length while GSSMs are fundamentally limited by their fixed-size latent state.
Empirically, we find that transformers outperform GSSMs in terms of efficiency and generalization on synthetic tasks that require copying the context.
Finally, we evaluate pretrained large language models and find that transformer models dramatically outperform state space models at copying and retrieving information from context.
Taken together, these results suggest a fundamental gap between transformers and GSSMs on tasks of practical interest. |
2212.08045v2 | Multimodal models are becoming increasingly effective, in part due to unified components, such as the Transformer architecture. However, multimodal models still often consist of many task- and modality-specific pieces and training procedures. For example, CLIP (Radford et al., 2021) trains independent text and image towers via a contrastive loss. We explore an additional unification: the use of a pure pixel-based model to perform image, text, and multimodal tasks. Our model is trained with contrastive loss alone, so we call it CLIP-Pixels Only (CLIPPO) . CLIPPO uses a single encoder that processes both regular images and text rendered as images. CLIPPO performs image-based tasks such as retrieval and zero-shot image classification almost as well as CLIP-style models, with half the number of parameters and no text-specific tower or embedding. When trained jointly via image-text contrastive learning and next-sentence contrastive learning, CLIPPO can perform well on natural language understanding tasks, without any word-level loss (language modelling or masked language modelling) , outperforming pixel-based prior work. Surprisingly, CLIPPO can obtain good accuracy in visual question answering, simply by rendering the question and image together. Finally, we exploit the fact that CLIPPO does not require a tokenizer to show that it can achieve strong performance on multilingual multimodal retrieval without modifications.Code and pretrained models are available as part ofbig_visionhttps://github.com/google-research/big_vision. |
2401.03321v2 | Recent works showed the possibility of building open-vocabulary large language models (LLMs) that directly operate on pixel representations and are implemented as encoder-decoder models that reconstruct masked image patches of rendered text.
However, these pixel-based LLMs are limited to autoencoding tasks and cannot generate new text as images.
As such, they cannot be used for open-answer or generative language tasks.
In this work, we overcome this limitation and introducePixar, the first pixel-based autoregressive LLM that does not rely on a pre-defined vocabulary for both input and output text.
Consisting of only a decoder,Pixarcan answer free-form generative tasks
while keeping the text representation learning performance on par with previous encoder-decoder models.
Furthermore, we highlight the challenges to autoregressively generate non-blurred text as images and link this to the usual maximum likelihood objective.
We propose a simple adversarial pretraining that significantly improves the readability and performance ofPixarmaking it comparable to GPT2 on short text generation tasks.
This paves the way to building open-vocabulary LLMs that are usable for free-form generative tasks and questions the necessity of the usual symbolic input representation – text as tokens – for these challenging tasks. |
2402.17861 | Audits are critical mechanisms for identifying the risks and limitations of deployed artificial intelligence (AI) systems.
However, the effective execution of AI audits remains incredibly difficult. As a result,
practitioners make use of various tools to support their efforts.
Drawing on interviews with 35 AI audit practitioners and a landscape analysis of 390 tools, we map the current ecosystem of available AI audit tools.
While there are many tools designed to assist practitioners with setting standards and evaluating AI systems, these tools often fell short of supporting the accountability goals of AI auditing in practice.
We thus highlight areas for future tool development beyond evaluation—from harms discovery to advocacy—and outline challenges practitioners faced in their efforts to use AI audit tools.
We conclude that resources are lacking to adequately support the full scope of needs for many AI audit practitioners and recommend that the field move beyond tools for just evaluation, towards more comprehensive infrastructure for AI accountability. |
2210.03347 | Visually-situated language is ubiquitous—sources range from textbooks with diagrams to web pages with images and tables, to mobile apps with buttons and forms. Perhaps due to this diversity, previous work has typically relied on domain-specific recipes with limited sharing of the underlying data, model architectures, and objectives. We presentPix2Struct, a pretrained image-to-text model for purely visual language understanding, which can be finetuned on tasks containing visually-situated language.Pix2Structis pretrained by learning to parse masked screenshots of web pages into simplified HTML. The web, with its richness of visual elements cleanly reflected in the HTML structure, provides a large source of pretraining data well suited to the diversity of downstream tasks. Intuitively, this objective subsumes common pretraining signals such as OCR, language modeling, and image captioning. In addition to the novel pretraining strategy, we introduce a variable-resolution input representation and a more flexible integration of language and vision inputs, where language prompts such as questions are rendered directly on top of the input image. For the first time, we show that a single pretrained model can achieve state-of-the-art results in six out of nine tasks across four domains: documents, illustrations, user interfaces, and natural images. |
2311.12022 | We presentGPQA, a challenging dataset of 448 multiple-choice questions written by domain experts in biology, physics, and chemistry.
We ensure that the questions are high-quality and extremely difficult: experts who have or are pursuing PhDs in the corresponding domains reach 65% accuracy (74% when discounting clear mistakes the experts identified in retrospect) , while highly skilled non-expert validators only reach 34% accuracy, despite spending on average over 30 minutes with unrestricted access to the web (i.e., the questions are “Google-proof”) .
The questions are also difficult for state-of-the-art AI systems, with our strongest GPT-4–based baseline achieving 39% accuracy.
If we are to use future AI systems to help us answer very hard questions—for example, when developing new scientific knowledge—we need to developscalable oversightmethods that enable humans to supervise their outputs, which may be difficult even if the supervisors are themselves skilled and knowledgeable.
The difficulty ofGPQAboth for skilled non-experts and frontier AI systems should enable realistic scalable oversight experiments, which we hope can help devise ways for human experts to reliably
get truthful information from
AI systems that surpass human capabilities. |
2111.15664 | Understanding document images (e.g., invoices) is a core but challenging task since it requires complex functions such asreading textand aholistic understanding of the document.
Current Visual Document Understanding (VDU) methods outsource the task of reading text to off-the-shelf Optical Character Recognition (OCR) engines and focus on the understanding task with the OCR outputs.
Although such OCR-based approaches have shown promising performance, they suffer from 1) high computational costs for using OCR; 2) inflexibility of OCR models on languages or types of documents; 3) OCR error propagation to the subsequent process.
To address these issues, in this paper, we introduce a novel OCR-free VDU model namedDonut, which stands forDocumentunderstandingtransformer.
As the first step in OCR-free VDU research, we propose a simple architecture (i.e., Transformer) with a pre-training objective (i.e.,cross-entropy loss) .
Donut is conceptually simple yet effective. Through extensive experiments and analyses, we show a simple OCR-free VDU model,Donut, achieves state-of-the-art performances on various VDU tasks in terms of both speed and accuracy.
In addition, we offer a synthetic data generator that helps the model pre-training to be flexible in various languages and domains.
The code, trained model, and synthetic data are available athttps://github.com/clovaai/donut. |
2307.07262 | Tokenization is a critical part of modern NLP pipelines. However, contemporary tokenizers for Large Language Models are based on statistical analysis of text corpora, without much consideration to the linguistic features. I propose a linguistically motivated tokenization scheme, MorphPiece, which is based partly on morphological segmentation of the underlying text. A GPT-style causal language model trained on this tokenizer (called MorphGPT) shows comparable or superior performance on a variety of supervised and unsupervised NLP tasks, compared to the OpenAI GPT-2 model. Specifically I evaluated MorphGPT on language modeling tasks, zero-shot performance on GLUE Benchmark with various prompt templates, massive text embedding benchmark (MTEB) for supervised and unsupervised performance, and lastly with another morphological tokenization scheme (FLOTA) and find that the model trained on MorphPiece outperforms GPT-2 on most evaluations, at times with considerable margin, despite being trained for about half the training iterations. |
1606.01305 | We propose zoneout, a novel method for regularizing RNNs.
At each timestep, zoneout stochastically forces some hidden units to maintain their previous values.
Like dropout, zoneout uses random noise to train a pseudo-ensemble, improving generalization.
But by preserving instead of dropping hidden units, gradient information and state information are more readily propagated through time, as in feedforward stochastic depth networks.
We perform an empirical investigation of various RNN regularizers, and find that zoneout gives significant performance improvements across tasks. We achieve competitive results with relatively simple models in character- and word-level language modelling on the Penn Treebank and Text8 datasets, and combining with recurrent batch normalizationyields state-of-the-art results on permuted sequential MNIST. |
1901.08817 | Recurrent neural networks are a widely used class of neural architectures. They have, however, two shortcomings. First, it is difficult to understand what exactly they learn. Second, they tend to work poorly on sequences requiring long-term memorization, despite having this capacity in principle. We aim to address both shortcomings with a class of recurrent networks that use a stochastic state transition mechanism between cell applications. This mechanism, which we term state-regularization, makes RNNs transition between a finite set of learnable states. We evaluate state-regularized RNNs on (1) regular languages for the purpose of automata extraction; (2) nonregular languages such as balanced parentheses, palindromes, and the copy task where external memory is required; and (3) real-word sequence learning tasks for sentiment analysis, visual object recognition, and language modeling. We show that state-regularization (a) simplifies the extraction of finite state automata modeling an RNN’s state transition dynamics; (b) forces RNNs to operate more like automata with external memory and less like finite state machines; (c) makes RNNs have better interpretability and explainability. |
1804.11188 | Successful recurrent models such as long short-term memories (LSTMs) and gated recurrent units (GRUs) use *ad hoc* gating mechanisms. Empirically these models have been found to improve the learning of medium to long term temporal dependencies and to help with vanishing gradient issues. We prove that learnable gates in a recurrent model formally provide *quasiinvariance to general time transformations* in the input data. We recover part of the LSTM architecture from a simple axiomatic approach. This result leads to a new way of initializing gate biases in LSTMs and GRUs. Experimentally, this new *chrono initialization* is shown to greatly improve learning of long term dependencies, with minimal implementation effort. Recurrent neural networks (e.g. (Jaeger, 2002)) are a standard machine learning tool to model and represent temporal data; mathematically they amount to learning the parameters of a parameterized dynamical system so that its behavior optimizes some criterion, such as the prediction of the next data in a sequence. Handling long term dependencies in temporal data has been a classical issue in the learning of recurrent networks. Indeed, stability of a dynamical system comes at the price of exponential decay of the gradient signals used for learning, a dilemma known as the *vanishing gradient* problem (Pascanu et al., 2012; Hochreiter, 1991; Bengio et al., 1994). This has led to the introduction of recurrent models specifically engineered to help with such phenomena. Use of feedback connections (Hochreiter & Schmidhuber, 1997) and control of feedback weights through gating mechanisms (Gers et al., 1999) partly alleviate the vanishing gradient problem. The resulting architectures, namely long short-term memories (LSTMs (Hochreiter & Schmidhuber, 1997; Gers et al., 1999)) and gated recurrent units (GRUs (Chung et al., 2014)) have become a standard for treating sequential data. Using orthogonal weight matrices is another proposed solution to the vanishing gradient problem, thoroughly studied in (Saxe et al., 2013; Le et al., 2015; Arjovsky et al., 2016; Wisdom et al., 2016; Henaff et al., 2016). This comes with either computational overhead, or limitation in representational power. Furthermore, restricting the weight matrices to the set of orthogonal matrices makes forgetting of useless information difficult. The contribution of this paper is threefold: ∙ We show that postulating invariance to time transformations in the data (taking invariance to time warping as an axiom) necessarily leads to a gate-like mechanism in recurrent models (Section 1). This provides a clean derivation of part of the popular LSTM and GRU architectures from first principles. In this framework, gate values appear as time contraction or time dilation coefficients, similar in spirit to the notion of time constant introduced in (Mozer, 1992). ∙ From these insights, we provide precise prescriptions on how to initialize gate biases (Section 2) depending on the range of time dependencies to be captured. It has previously been advocated that setting the bias of the forget gate of LSTMs to 1 or 2 provides overall good performance (Gers & Schmidhuber, 2000; Jozefowicz et al., 2015). The viewpoint here 1 explains why this is reasonable in most cases, when facing medium term dependencies, but fails when facing long to very long term dependencies. ∙ We test the empirical benefits of the new initialization on both synthetic and real world data (Section 3). We observe substantial improvement with long-term dependencies, and slight gains or no change when short-term dependencies dominate. |
1810.09536 | Natural language is hierarchically structured: smaller units (e.g., phrases) are nested within larger units (e.g., clauses) . When a larger constituent ends, all of the smaller constituents that are nested within it must also be closed. While the standard LSTM architecture allows different neurons to track information at different time scales, it does not have an explicit bias towards modeling a hierarchy of constituents. This paper proposes to add such an inductive bias byorderingthe neurons; a vector of master input and forget gates ensures that when a given neuron is updated, all the neurons that follow it in the ordering are also updated. Our novel recurrent architecture,ordered neuronsLSTM (ON-LSTM) , achieves good performance on four different tasks: language modeling, unsupervised parsing, targeted syntactic evaluation, and logical inference111The code can be found athttps://github.com/yikangshen/Ordered-Neurons.. |
2311.04823 | Transformers have surpassed RNNs in popularity due to their superior abilities in parallel training and long-term dependency modeling.
Recently, there has been a renewed interest in using linear RNNs for efficient sequence modeling.
These linear RNNs often employ gating mechanisms in the output of the linear recurrence layer while ignoring the significance of using forget gates within the recurrence. In this paper, we propose a gated linear RNN model dubbed Hierarchically Gated Recurrent Neural Network (HGRN) , which includes forget gates that are lower bounded by a learnable value. The lower bound increases monotonically when moving up layers. This allows the upper layers to model long-term dependencies and the lower layers to model more local, short-term dependencies. Experiments on language modeling, image classification, and long-range arena benchmarks showcase the efficiency and effectiveness of our proposed model. The source code is available athttps://github.com/OpenNLPLab/HGRN††⋆Equal contribution.Indicates corresponding author (Email address:[email protected]) .. |
2402.18334 | We introduce Bonito, an open-source model forconditional task generation: the task of converting unannotated text into task-specific training datasets for instruction tuning.
Our goal is to enable zero-shot task adaptation of large language models on users’ specialized, private data.
We train Bonito on a new large-scale dataset with 1.65M examples created by remixing existing instruction tuning datasets intometa-templates.
The meta-templates for a dataset produce training examples where the input is the unannotated text and the task attribute and the output consists of the instruction and the response.
We use Bonito to generate synthetic tasks for seven datasets from specialized domains across three task types—yes-no question answering, extractive question answering, and natural language inference—and adapt language models.
We show that Bonito significantly improves the average performance of pretrained and instruction tuned models over the de facto self supervised baseline.
For example, adapting Mistral-Instruct-v2 and instruction tuned variants of Mistral and Llama2 with Bonito improves the strong zero-shot performance by 22.1 F1 points whereas the next word prediction objective undoes some of the benefits of instruction tuning and reduces the average performance by 0.8 F1 points.
We conduct additional experiments with Bonito to understand the effects of the domain, the size of the training set, and the choice of alternative synthetic task generators.
Overall, we show that learning with synthetic instruction tuning datasets is an effective way to adapt language models to new domains.
The model, dataset, and code are available athttps://github.com/BatsResearch/bonito. |
2402.19427 | Recurrent neural networks (RNNs) have fast inference and scale efficiently on long sequences, but they are difficult to train and hard to scale. We propose Hawk, an RNN with gated linear recurrences, and Griffin, a hybrid model that mixes gated linear recurrences with local attention. Hawk exceeds the reported performance of Mamba on downstream tasks, while Griffin matches the performance of Llama-2 despite being trained on over 6 times fewer tokens. We also show that Griffin can extrapolate on sequences significantly longer than those seen during training. Our models match the hardware efficiency of Transformers during training, and during inference they have lower latency and significantly higher throughput. We scale Griffin up to 14B parameters, and explain how to shard our models for efficient distributed training. |
2401.04658 | Linear attention is an efficient attention mechanism that has recently emerged as a promising alternative to conventional softmax attention. With its ability to process tokens in linear computational complexities, linear attention, in theory, can handle sequences of unlimited length without sacrificing speed,i.e.,maintaining a constant training speed for various sequence lengths with a fixed memory consumption.
However, due to the issue with cumulative summation (cumsum) , current Linear Attention algorithms cannot demonstrate their theoretical advantage in a casual setting. In this paper, we present Lightning Attention-2, the first linear attention implementation that enables linear attention to realize its theoretical computational benefits. To achieve this, we leverage the thought of tiling, separately handling the intra-block and inter-block components in linear attention calculation. Specifically, we utilize the conventional attention computation mechanism for the intra-blocks and apply linear attention kernel tricks for the inter-blocks. A tiling technique is adopted through both forward and backward procedures to take full advantage of the GPU hardware. We implement our algorithm in Triton to make it IO-aware and hardware-friendly. Various experiments are conducted on different model sizes and sequence lengths. Lightning Attention-2 retains consistent training and inference speed regardless of input sequence length and is significantly faster than other attention mechanisms. The source code is available atLightning Attention-2. |
1606.01981 | Recent results show that deep neural networks achieve excellent performance even when, during training, weights are quantized and projected to a binary representation.
Here, we show that this is just the tip of the iceberg: these same networks, during testing, also exhibit a remarkable robustness to distortions beyond quantization, including additive and multiplicative noise, and a class of non-linear projections where binarization is just a special case.
To quantify this robustness, we show that one such network achievestest error on CIFAR-10 even witheffective bits per weight.
Furthermore, we find that a common training heuristic—namely, projecting quantized weights during backpropagation—can be altered (or even removed) and networks still achieve a base level of robustness during testing.
Specifically, training with weight projections other than quantization also works, as does simply clipping the weights, both of which have never been reported before.
We confirm our results for CIFAR-10 and ImageNet datasets.
Finally, drawing from these ideas, we propose a stochastic projection rule that leads to a new state of the art network withtest error on CIFAR-10 using no data augmentation. |
2402.17764 | Recent research, such as BitNet, is paving the way for a new era of 1-bit Large Language Models (LLMs) . In this work, we introduce a 1-bit LLM variant, namelyBitNet b1.58, in which every single parameter (or weight) of the LLM is ternary {-1, 0, 1}. It matches the full-precision (i.e., FP16 or BF16) Transformer LLM with the same model size and training tokens in terms of both perplexity and end-task performance, while being significantly more cost-effective in terms of latency, memory, throughput, and energy consumption.
More profoundly, the 1.58-bit LLM defines a new scaling law and recipe for training new generations of LLMs that are both high-performance and cost-effective. Furthermore, it enables a new computation paradigm and opens the door for designing specific hardware optimized for 1-bit LLMs. |
2402.17762 | We observe an empirical phenomenon in Large Language Models (LLMs) —very few activations exhibit significantly larger values than others (e.g., 100,000 times larger) . We call themmassive activations. First, we demonstrate the widespread existence of massive activations across various LLMs and characterize their locations. Second, we find their values largely stay constant regardless of the input, and they function as indispensable bias terms in LLMs. Third, these massive activations lead to the concentration of attention probabilities to their corresponding tokens, and further, implicit bias terms in the self-attention output. Last, we also study massive activations in Vision Transformers.111Code is available athttps://github.com/locuslab/massive-activations. |
2402.08017 | We introduceLumos, the first end-to-end multimodal question-answering system with text understanding capabilities. At the core ofLumosis a Scene Text Recognition (STR) component that extracts text from first person point-of-view images, the output of which is used to augment input to a Multimodal Large Language Model (MM-LLM) . While buildingLumos, we encountered numerous challenges related to STR quality, overall latency, and model inference. In this paper, we delve into those challenges, and discuss the system architecture, design choices, and modeling techniques employed to overcome these obstacles. We also provide a comprehensive evaluation for each component, showcasing high quality and efficiency. |
2402.16828 | The scalability of deep learning models is fundamentally limited by computing resources, memory, and communication. Although methods like low-rank adaptation (LoRA) have reduced the cost of model finetuning, its application in model pre-training remains largely unexplored. This paper explores extending LoRA to model pre-training, identifying the inherent constraints and limitations of standard LoRA in this context. We introduceLoRA-the-Explorer(LTE) , a novel bi-level optimization algorithm designed to enable parallel training of multiple low-rank heads across computing nodes, thereby reducing the need for frequent synchronization. Our approach includes extensive experimentation on vision transformers using various vision datasets, demonstrating that LTE is competitive with standard pre-training.minyoungg.github.io/LTE |
2305.09515 | Diffusion models have gained significant attention in the realm of image generation due to their exceptional performance. Their success has been recently expanded to text generation via generating all tokens within a sequence concurrently.
However, natural language exhibits a far more pronounced sequential dependency in comparison to images, and the majority of existing language models are trained with a left-to-right auto-regressive approach.
To account for the inherent sequential characteristic of natural language, we introduce Auto-Regressive Diffusion (AR-Diffusion) .AR-Diffusionensures that the generation of tokens on the right depends on the generated ones on the left, a mechanism achieved through employing a dynamic number of denoising steps that vary based on token position. This results in tokens on the left undergoing fewer denoising steps than those on the right, thereby enabling them to generate earlier and subsequently influence the generation of tokens on the right.
In a series of experiments on various text generation tasks, including text summarization, machine translation, and common sense generation,AR-Diffusionclearly demonstrated its superiority over existing diffusion language models and that it can befaster when achieving comparable results. Our code is available athttps://github.com/microsoft/ProphetNet/tree/master/AR-diffusion. |
2401.05268 | Language agents have achieved considerable performance on various complex question-answering tasks. Despite the incessant exploration in this field, existing language agent systems still struggle with costly, non-reproducible data reliance and face the challenge of compelling a single model for multiple functions. To this end, we introduceAutoAct, an automatic agent learning framework that does not rely on large-scale annotated data and synthetic trajectories from closed-source models (e.g., GPT-4) . Given limited data with a tool library,AutoActfirst automatically synthesizes planning trajectories without any assistance from humans or strong closed-source models. Then,AutoActleverages adivision-of-laborstrategy to automatically differentiate based on the target task information and synthesized trajectories, producing a sub-agent group to complete the task. We conduct comprehensive experiments with different LLMs, which demonstrates thatAutoActyields better or parallel performance compared to various strong baselines. Further analysis demonstrates the effectiveness of thedivision-of-laborstrategy, with the trajectory quality generated byAutoActsignificantly outperforming that of others111Code will be available athttps://github.com/zjunlp/AutoAct.. |
2209.12951 | A proper parametrization of state transition matrices of linear state-space models (SSMs) followed by standard nonlinearities enables them to efficiently learn representations from sequential data, establishing the state-of-the-art on a large series of long-range sequence modeling benchmarks. In this paper, we show that we can improve further when the structural SSM such as S4 is given by a linear liquid time-constant (LTC) state-space model. LTC neural networks are causal continuous-time neural networks with an input-dependent state transition module, which makes them learn to adapt to incoming inputs at inference. We show that by using a diagonal plus low-rank decomposition of the state transition matrix introduced in S4, and a few simplifications, the LTC-based structural state-space model, dubbed Liquid-S4, achieves the new state-of-the-art generalization across sequence modeling tasks with long-term dependencies such as image, text, audio, and medical time-series, with an average performance of 87.32% on the Long-Range Arena benchmark. On the full raw Speech Command recognition dataset Liquid-S4 achieves 96.78% accuracy with 30% reduction in parameter counts compared to S4. The additional gain in performance is the direct result of the Liquid-S4’s kernel structure that takes into account the similarities of the input sequence samples during training and inference. |
2402.10949 | Large Language Models (LLMs) have demonstrated remarkable problem-solving and basic mathematics abilities. However, their efficacy is highly contingent on the formulation of the prompt. This study endeavors to quantify the influence of incorporating “positive thinking” into the system message of the prompt, then compare that to systematic prompt optimization. We assess the performance of 60 combinations of system message snippets, tested with and without Chain of Thought prompting, across three models with parameters ranging from 7 to 70 billion on the GSM8K dataset. Our findings reveal that results do not universally generalize across models. In most instances, the inclusion of “positive thinking” prompts positively affected model performance. Notably, however, Llama2-70B exhibited an exception when not utilizing Chain of Thought, as the optimal system message was found to be none at all. Given the combinatorial complexity, and thus computation time, of experimenting with hand-tuning prompts for large black-box models, we then compared the performance of the best “positive thinking” prompt against the output of systematic prompt optimization. We show that employing an automated prompt optimizer emerges as the most effective method for enhancing performance, even when working with smaller open-source models. Additionally, our findings reveal that the highest-scoring, automatically-optimized prompt exhibits a degree of peculiarity far beyond expectations. |
2312.14890 | Complex reasoning ability is one of the most important features of current Large Language Models (LLMs) , which has also been leveraged to play an integral role in complex decision-making tasks. Therefore, the investigation into the reasoning capabilities of LLMs is critical: numerous benchmarks have been established to assess the reasoning abilities of LLMs. However, current benchmarks are inadequate in offering a rigorous evaluation of the full extent of reasoning abilities that LLMs are capable of achieving. They are also prone to the risk of overfitting, as these benchmarks, being publicly accessible and static, allow models to potentially tailor their responses to specific benchmark metrics, thereby inflating their performance.
Addressing these limitations, our research introduces a new benchmark, namedNPHardEval. This benchmark is designed to evaluate the reasoning abilities of LLMs across a broad spectrum of 900 algorithmic questions, extending up to the NP-Hard complexity class. These questions are meticulously chosen to represent a wide range of complexity class below the NP-hard complexity class, offering a rigorous measure of the reasoning ability of LLMs.
Through this study, we shed light on the current state of reasoning in LLMs, providing an objective and rigorous perspective through the comparison of LLMs’ performance across complex classes.
Our findings contribute significantly to understanding the current capabilities of LLMs in reasoning tasks and lay the groundwork for future advancements in enhancing the reasoning abilities of these models.
Moreover, this benchmark is designed with adynamic update mechanism, where the datapoints are refreshed on a monthly basis. Such regular updates play a crucial role in mitigating the risk of LLMs overfitting to the benchmark, promoting a more accurate and reliable assessment of their reasoning capabilities. The benchmark dataset and code ofNPHardEvalare available athttps://github.com/casmlab/NPHardEval. |
2312.15234 | In the rapidly evolving landscape of artificial intelligence (AI) , generative large language models (LLMs) stand at the forefront, revolutionizing how we interact with our data.
However, the computational intensity and memory consumption of deploying these models present substantial challenges in terms of serving efficiency, particularly in scenarios demanding low latency and high throughput.
This survey addresses the imperative need for efficient LLM serving methodologies from a machine learning system (MLSys) research perspective, standing at the crux of advanced AI innovations and practical system optimizations.
We provide in-depth analysis, covering a spectrum of solutions, ranging from cutting-edge algorithmic modifications to groundbreaking changes in system designs.
The survey aims to provide a comprehensive understanding of the current state and future directions in efficient LLM serving, offering valuable insights for researchers and practitioners in overcoming the barriers of effective LLM deployment, thereby reshaping the future of AI. |
2402.14672 | The applications of large language models (LLMs) have expanded well beyond the confines of text processing, signaling a new era where LLMs are envisioned as generalist language agents capable of operating within complex real-world environments.
These environments are often highly expansive, making it impossible for the LLM to process them within its short-term memory.
Motivated by recent research on extending the capabilities of LLMs with tools, this paper investigates the intriguing potential of tools to augment LLMs in handling such complexity.
To this end, we design customized tools to aid in the proactive exploration within these massive environments.
Such tools can serve as amiddlewarelayer shielding the LLM from environmental complexity.
In two representative complex environments—knowledge bases (KBs) and databases—we demonstrate the significant potential of augmenting language agents with tools in complex environments.
Notably, equipped with these tools, GPT-4 achieves2.8the performance of the best baseline in tasks requiring access to database content and2.2in KB tasks.
Our findings illuminate the path for advancing language agents in complex real-world applications.111Our code and data will be released atOSU_NLP/Fuxi. |
2309.02772 | Recently, Large Language Models (LLMs) have shown impressive abilities in code generation. However, existing LLMs’ decoding strategies are designed for Natural Language (NL) generation, overlooking the differences between NL and programming languages (PL) . Due to this oversight, a better decoding strategy for code generation remains an open question. In this paper, we conduct the first systematic study to explore a decoding strategy specialized in code generation. With an analysis of loss distributions of code tokens, we find that code tokens can be divided into two categories: challenging tokens that are difficult to predict and confident tokens that can be easily inferred. Among them, the challenging tokens mainly appear at the beginning of a code block. Inspired by the above findings, we propose a simple yet effective method: Adaptive Temperature (AdapT) sampling, which dynamically adjusts the temperature coefficient when decoding different tokens. We apply a larger temperature when sampling for challenging tokens, allowing LLMs to explore diverse choices. We employ a smaller temperature for confident tokens avoiding the influence of tail randomness noises. We apply AdapT sampling to LLMs with different sizes and conduct evaluations on two popular datasets. Results show that AdapT sampling significantly outperforms state-of-the-art decoding strategy. |
2402.14658 | The introduction of large language models has significantly advanced code generation. However, open-source models often lack the execution capabilities and iterative refinement of advanced systems like the GPT-4 Code Interpreter. To address this, we introduceOpenCodeInterpreter, a family of open-source code systems designed for generating, executing, and iteratively refining code. Supported by Code-Feedback, a dataset featuring 68K multi-turn interactions,OpenCodeInterpreterintegrates execution and human feedback for dynamic code refinement. Our comprehensive evaluation ofOpenCodeInterpreteracross key benchmarks such as HumanEval, MBPP, and their enhanced versions from EvalPlus reveals its exceptional performance. Notably,OpenCodeInterpreter-33B achieves an accuracy of 83.2 (76.4) on the average (and plus versions) of HumanEval and MBPP, closely rivaling GPT-4’s 84.2 (76.2) and further elevates to 91.6 (84.6) with synthesized human feedback from GPT-4.OpenCodeInterpreterbrings the gap between open-source code generation models and proprietary systems like GPT-4 Code Interpreter. |
2312.03732 | As large language models (LLMs) have become increasingly compute and memory intensive, parameter-efficient fine-tuning (PEFT) methods are now a common strategy to fine-tune LLMs. A popular PEFT method is Low-Rank Adapters (LoRA) , which adds trainable low-rank “adapters” to selected layers. Each adapter consists of a low-rank matrix product, multiplicatively scaled by a rank-dependent factor. This scaling factor, which divides adapters by a factor of the rank, results in slowed learning and stunted performance for LoRA with higher-rank adapters. Consequently, the use of LoRA in practice has generally been limited to very low ranks. In this work, we study the impact of the scaling factor on the learning process and prove that LoRA adapters should be divided by a factor of the square root of the rank. Modifying LoRA with the appropriate scaling factor, which we call the rank-stabilized LoRA (rsLoRA) method, easily provides for a fine-tuning compute/performance trade-off, where larger ranks can be used to trade off increased computational resources during training for better fine-tuning performance, with no change in inference computing cost. |
2402.13178 | While large language models (LLMs) have achieved state-of-the-art performance on a wide range of medical question answering (QA) tasks, they still face challenges with hallucinations and outdated knowledge. Retrieval-augmented generation (RAG) is a promising solution and has been widely adopted. However, a RAG system can involve multiple flexible components, and there is a lack of best practices regarding the optimal RAG setting for various medical purposes. To systematically evaluate such systems, we propose the Medical Information Retrieval-Augmented Generation Evaluation (Mirage) , a first-of-its-kind benchmark including 7,663 questions from five medical QA datasets. UsingMirage, we conducted large-scale experiments with over 1.8 trillion prompt tokens on 41 combinations of different corpora, retrievers, and backbone LLMs through theMedRagtoolkit introduced in this work. Overall,MedRagimproves the accuracy of six different LLMs by up to 18% over chain-of-thought prompting, elevating the performance of GPT-3.5 and Mixtral to GPT-4-level. Our results show that the combination of various medical corpora and retrievers achieves the best performance. In addition, we discovered a log-linear scaling property and the “lost-in-the-middle” effects in medical RAG. We believe our comprehensive evaluations can serve as practical guidelines for implementing RAG systems for medicine. |
2401.14887 | Retrieval-Augmented Generation (RAG) systems represent a significant advancement over traditional Large Language Models (LLMs) .
RAG systems enhance their generation ability by incorporating external data retrieved through an Information Retrieval (IR) phase, overcoming the limitations of standard LLMs, which are restricted to their pre-trained knowledge and limited context window.
Most research in this area has predominantly concentrated on the generative aspect of LLMs within RAG systems.
Our study fills this gap by thoroughly and critically analyzing the influence of IR components on RAG systems.
This paper analyzes which characteristics a retriever should possess for an effective RAG’s prompt formulation, focusing on the type of documents that should be retrieved.
We evaluate various elements, such as the relevance of the documents to the prompt, their position, and the number included in the context.
Our findings reveal, among other insights, that including irrelevant documents can unexpectedly enhance performance by more thanin accuracy, contradicting our initial assumption of diminished quality.
These results underscore the need for developing specialized strategies to integrate retrieval with language generation models, thereby laying the groundwork for future research in this field.111The code and data are available atgithub.com/florin-git/The-Power-of-Noise |
2402.14740 | AI alignment in the shape of Reinforcement Learning from Human Feedback (RLHF) is increasingly treated as a crucial ingredient for high performance large language models.Proximal Policy Optimization(PPO) has been positioned by recent literature as the canonical method for the RL part of RLHF. However, it involves both high computational cost and sensitive hyperparameter tuning. We posit that most of the motivational principles that led to the development of PPO are less of a practical concern in RLHF and advocate for a less computationally expensive method that preserves and even increases performance. We revisit theformulationof alignment from human preferences in the context of RL. Keeping simplicity as a guiding principle, we show that many components of PPO are unnecessary in an RLHF context and that far simpler REINFORCE-style optimization variants outperform both PPO and newly proposed “RL-free” methods such as DPO and RAFT. Our work suggests that careful adaptation to LLMs alignment characteristics enables benefiting from online RL optimization at low cost. |
2402.13753 | Large context window is a desirable feature in large language models (LLMs) .
However, due to high fine-tuning costs, scarcity of long texts, and catastrophic values introduced by new token positions, current extended context windows are limited to around 128k tokens. This paper introduces LongRoPE that, for the first time, extends the context window of pre-trained LLMs to an impressive2048ktokens, with up to only 1k fine-tuning steps at within 256k training lengths, while maintaining performance at the original short context window. This is achieved
by three key innovations: (i) we identify and exploit two forms of non-uniformities in positional interpolation through an efficient search, providing a better initialization for fine-tuning and enabling an 8extension in non-fine-tuning scenarios; (ii) we introduce a progressive extension strategy that first fine-tunes a 256k length LLM and then conducts a second positional interpolation on the fine-tuned extended LLM to achieve a 2048k context window; (iii) we readjust LongRoPE on 8k length to recover the short context window performance.
Extensive experiments on LLaMA2 and Mistral across various tasks demonstrate the effectiveness of our method. Models extended via LongRoPE retain the original architecture with minor modifications to the positional embedding, and can reuse most pre-existing optimizations. Code will be available athttps://github.com/microsoft/LongRoPE |
2202.11233 | We introduce Retrieval Augmented Classification (RAC) , a generic approach to augmenting standard image classification pipelines with an explicit retrieval module. RAC consists of a standard base image encoder fused with a parallel retrieval branch that queries a non-parametric external memory of pre-encoded images and associated text snippets. We apply RAC to the problem of long-tail classification and demonstrate a significant improvement over previous state-of-the-art on Places365-LT and iNaturalist-2018 (andrespectively) , despite using only the training datasets themselves as the external information source. We demonstrate that RAC’s retrieval module, without prompting, learns a high level of accuracy on tail classes.
This, in turn, frees the base encoder to focus on common classes, and improve its performance thereon. RAC represents an alternative approach to utilizing large, pretrained models without requiring fine-tuning, as well as a first step towards more effectively making use of external memory within common computer vision architectures. |
2402.12366 | Reinforcement learning with AI feedback (RLAIF) is a popular paradigm for improving the instruction-following abilities of powerful pre-trained language models. RLAIF first performs supervised fine-tuning (SFT) using demonstrations from a teacher model and then further fine-tunes the model with reinforcement learning (RL) , using feedback from a critic model. While recent popular open-source models have demonstrated substantial improvements in performance from the RL step, in this paper we question whether the complexity of this RL step is truly warranted for AI feedback. We show that the improvements of the RL step are virtually entirely due to the widespread practice of using a weaker teacher model (e.g. GPT-3.5) for SFT data collection than the critic (e.g., GPT-4) used for AI feedback generation. Specifically, we show that simple supervised fine-tuning with GPT-4 as the teacher outperforms existing RLAIF pipelines. More generally, we find that the gains from RLAIF vary substantially across base model families, test-time evaluation protocols, and critic models. Finally, we provide a mechanistic explanation for when SFT may outperform the full two-step RLAIF pipeline as well as suggestions for making RLAIF maximally useful in practice. Code is available at:https://github.com/architsharma97/dpo-rlaif. |
2212.09741 | We introduceInstructOR, a new method for computing text embeddings given task instructions: every text input
is embedded together with instructions explaining the use case (e.g., task and domain descriptions) .
Unlike encoders from prior work that are more specialized,InstructORis a single embedder that can generate text embeddings tailored to different downstream tasks and domains,withoutany further training.
We first annotate instructions for 330 diverse tasks and trainInstructORon this multitask mixture with a contrastive loss.
We evaluateInstructORon 70 embedding evaluation tasks (66 of which areunseenduring training) , ranging from classification and information retrieval to semantic textual similarity and text generation evaluation.InstructOR, while having an order of magnitude fewer parameters than the previous best model, achieves state-of-the-art performance, with an average improvement of 3.4% compared to the previous best results on the 70 diverse datasets.
Our analysis suggests thatInstructORis robust to changes in instructions, and that instruction finetuning mitigates the challenge of training a single model on diverse datasets.
Our model, code, and data are available athttps://instructor-embedding.github.io. |
2305.06972 | Recent progress in artificial intelligence (AI), particularly in the domain of large language models (LLMs), has resulted in powerful and versatile dual-use systems. This intelligence can be put towards a wide variety of beneficial tasks, yet it can also be used to cause harm. This study explores one such harm by examining how LLMs can be used for spear phishing, a form of cybercrime that involves manipulating targets into divulging sensitive information. I first explore LLMs' ability to assist with the reconnaissance and message generation stages of a spear phishing attack, where I find that LLMs are capable of assisting with the email generation phase of a spear phishing attack. To explore how LLMs could potentially be harnessed to scale spear phishing campaigns, I then create unique spear phishing messages for over 600 British Members of Parliament using OpenAI's GPT-3.5 and GPT-4 models. My findings provide some evidence that these messages are not only realistic but also cost-effective, with each email costing only a fraction of a cent to generate. Next, I demonstrate how basic prompt engineering can circumvent safeguards installed in LLMs, highlighting the need for further research into robust interventions that can help prevent models from being misused. To further address these evolving risks, I explore two potential solutions: structured access schemes, such as application programming interfaces, and LLM-based defensive systems. Keywords Artificial Intelligence · AI Policy · Cybersecurity · Spear Phishing · Large Language Models |
2307.10169 | Large Language Models (LLMs) went from non-existent to ubiquitous in the machine learning discourse within a few years. Due to the fast pace of the field, it is difficult to identify the remaining challenges and already fruitful application areas. In this paper, we aim to establish a systematic set of open problems and application successes so that ML researchers can comprehend the field’s current state more quickly and become productive.11footnotetext:Equal contribution.22footnotetext:{jean.kaddour,robert.mchardy}[email protected], [email protected] |
1602.07868 | We presentweight normalization: a reparameterization of the weight vectors in a neural network that decouples the length of those weight vectors from their direction. By reparameterizing the weights in this way we improve the conditioning of the optimization problem and we speed up convergence of stochastic gradient descent. Our reparameterization is inspired by batch normalization but does not introduce any dependencies between the examples in a minibatch. This means that our method can also be applied successfully to recurrent models such as LSTMs and to noise-sensitive applications such as deep reinforcement learning or generative models, for which batch normalization is less well suited. Although our method is much simpler, it still provides much of the speed-up of full batch normalization. In addition, the computational overhead of our method is lower, permitting more optimization steps to be taken in the same amount of time. We demonstrate the usefulness of our method on applications in supervised image recognition, generative modelling, and deep reinforcement learning. |
2402.12348 | As Large Language Models (LLMs) are integrated into critical real-world applications, their strategic and logical reasoning abilities are increasingly crucial.
This paper evaluates LLMs’ reasoning abilities in competitive environments through game-theoretic tasks, e.g., board and card games that require pure logic and strategic reasoning to compete with opponents.
We first proposeGTBench, a language-driven environment composing 10 widely-recognized tasks, across a comprehensive game taxonomy: complete versus incomplete information, dynamic versus static, and probabilistic versus deterministic scenarios.
Then, we investigate two key problems: ➊ Characterizing game-theoretic reasoning of LLMs; ➋ LLM-vs-LLM competitions as reasoning evaluation.
We observe that ➊ LLMs have distinct behaviors regarding various gaming scenarios; for example, LLMs fail in complete and deterministic games yet they are competitive in probabilistic gaming scenarios; ➋ Open-source LLMs, e.g., CodeLlama-34b-Instruct, are less competitive than commercial LLMs, e.g., GPT-4, in complex games.
In addition, code-pretraining greatly benefits strategic reasoning, while advanced reasoning methods such as Chain-of-Thought (CoT) and Tree-of-Thought (ToT) do not always help.
Detailed error profiles are also provided for a better understanding of LLMs’ behavior. |
2312.01552 | Alignment tuning has become the de facto standard practice for enabling base large language models (LLMs) to serve as open-domain AI assistants.
The alignment tuning process typically involves instruction learning through supervised fine-tuning (SFT) and preference tuning via reinforcement learning from human feedback (RLHF) .
A recent study, LIMA, shows that using merely 1K examples for SFT can achieve significant alignment performance as well, suggesting that the effect of alignment tuning might be “superficial. ”
This raises questions abouthow exactly the alignment tuning transforms a base LLM. We analyze the effect of alignment tuning by examining the token distribution shift between base LLMs and their aligned counterpart (e.g., Llama-2 and Llama-2-chat) .
Our findings reveal that base LLMs and their alignment-tuned versions perform nearly identically in decoding on the majority of token positions (i.e., they share the top-ranked tokens) .
Most distribution shifts occur withstylistictokens (e.g., discourse markers, safety disclaimers) . These direct evidence strongly supports the hypothesis that alignment tuning primarily learns to adopt the language style of AI assistants, and that the knowledge required for answering user queries predominantly comes from the base LLMs themselves. Based on these findings, we rethink the alignment of LLMs by posing the research question:how effectively can we align base LLMs without SFT or RLHF?To address this,
we introduce a simple, tuning-free alignment method,Urial(Untuned LLMs withRestyledIn-contextALignment) .Urialachieves effective alignment purely through in-context learning (ICL) with base LLMs, requiring as few as three constant stylistic examples and a system prompt.
We conduct a fine-grained and interpretable evaluation on a diverse set of examples, namedjust-eval-instruct.
Results demonstrate that base LLMs withUrialcan match or even surpass the performance of LLMs aligned with SFT (Mistral-7b-Instruct) or SFT+RLHF (Llama-2-70b-chat) .
We show that the gap between tuning-free and tuning-based alignment methods can be significantly reduced through strategic prompting and ICL.
Our findings on superficial nature of alignment tuning and results withUrialsuggest that deeper analysis and theoretical understanding of alignment is crucial to future LLM research. |
2312.14135 | When we look around and perform complex tasks, how we see and selectively process what we see is crucial. However, the lack of this visual search mechanism in current multimodal LLMs (MLLMs) hinders their ability to focus on important visual details, especially when handling high-resolution and visually crowded images. To address this, we introduce V∗, an LLM-guided visual search mechanism that employs the world knowledge in LLMs for efficient visual querying. When combined with an MLLM, this mechanism enhances collaborative reasoning, contextual understanding, and precise targeting of specific visual elements. This integration results in a new MLLM meta-architecture, namedShow, sEArch, and TelL(SEAL) . We further create V∗Bench, a benchmark specifically designed to evaluate MLLMs in their ability to process high-resolution images and focus on visual details. Our study highlights the necessity of incorporating visual search capabilities into multimodal systems. The code is availablehere. |
2402.10176 | Recent work has shown the immense potential of synthetically generated datasets for training large language models (LLMs) , especially for acquiring targeted skills.
Current large-scale math instruction tuning datasets such as MetaMathQAand MAmmoTHare constructed using outputs from closed-source LLMs with commercially restrictive licenses.
A key reason limiting the use of open-source LLMs in these data generation pipelines has been the wide gap between the mathematical skills of the best closed-source LLMs, such as GPT-4, and the best open-source LLMs.
Building on the recent progress in open-source LLMs, our proposed prompting novelty, and some brute-force scaling, we construct OpenMathInstruct-1, a math instruction tuning dataset with 1.8M problem-solution pairs.
The dataset is constructed by synthesizingcode-interpretersolutions for GSM8K and MATH, two popular math reasoning benchmarks, using the recently released and permissively licensed Mixtral model.
Our best model, OpenMath-CodeLlama-70B, trained on a subset of OpenMathInstruct-1, achieves a score of 84.6% on GSM8K and 50.7% on MATH, which is competitive with the bestgpt-distilledmodels.
We release our code, models, and the OpenMathInstruct-1 dataset under a commercially permissive license.111Data and models are available athttps://huggingface.co/collections/nvidia/openmath-65c5619de2ba059be0775014Code is available athttps://github.com/Kipok/NeMo-Skills |
2311.15566 | The high computational and memory requirements of generative large language models (LLMs) make it challenging to serve them cheaply.
This paper aims to reduce the monetary cost for serving LLMs by leveraging preemptible GPU instances on modern clouds, which offer accesses to spare GPU resources at a much cheaper price than regular instances but may be preempted by the cloud provider at any time.
Serving LLMs on preemptible instances requires addressing challenges induced by frequent instance preemptions and the necessity of migrating instances to handle these preemptions. This paper presents SpotServe, the first distributed LLM serving system on preemptible instances.
Several key techniques in SpotServe realize fast and reliable serving of generative LLMs on cheap preemptible instances.
First, SpotServe dynamically adapts the LLM parallelization configuration for dynamic instance availability and fluctuating workload, while balancing the trade-off among the overall throughput, inference latency and monetary costs.
Second, to minimize the cost of migrating instances for dynamic reparallelization, the task of migrating instances is formulated as a bipartite graph matching problem in SpotServe, which uses the Kuhn-Munkres algorithm to identify an optimal migration plan that minimizes communication cost.
Finally, to take advantage of the grace period offered by modern cloud platforms, we introduce stateful inference recovery, a new inference mechanism that commits inference progress at a much finer granularity and allows SpotServe to cheaply resume inference upon preemption.
We evaluate SpotServe on real spot instance preemption traces and various popular LLMs and show that SpotServe can reduce the P99 tail latency by 2.4 - 9.1compared with the best existing LLM serving systems.
We also show that SpotServe can leverage the price advantage of preemptive instances, savingmonetary cost compared with only using on-demand instances. The code is open-sourced at:https://github.com/Hsword/SpotServe. |
1903.07435 | Recent work has shown that LSTMs trained on a generic language
modeling objective capture syntax-sensitive generalizations such as
long-distance number agreement. We have however no mechanistic
understanding of how they accomplish this remarkable feat. Some have
conjectured it depends on heuristics that do not truly take
hierarchical structure into account. We present here a detailed
study of the inner mechanics of number tracking in LSTMs at the
single neuron level. We discover that long-distance number
information is largely managed by two “number units”.
Importantly, the behaviour of these units is partially controlled by
other units independently shown to track syntactic structure. We
conclude that LSTMs are, to some extent, implementing genuinely
syntactic processing mechanisms, paving the way to a more general
understanding of grammatical encoding in LSTMs. |
2402.04315 | While recent Large Language Models (LLMs) have proven useful in answering user queries, they are prone to hallucination, and their responses often lack credibility due to missing references to reliable sources. An intuitive solution to these issues would be to include in-text citations referring to external documents as evidence. While previous works have directly prompted LLMs to generate in-text citations, their performances are far from satisfactory, especially when it comes to smaller LLMs. In this work, we propose an effective training framework usingfine-grained rewardsto teach LLMs to generate highly supportive and relevant citations, while ensuring the correctness of their responses. We also conduct a systematic analysis of applying these fine-grained rewards to common LLM training strategies, demonstrating its advantage over conventional practices.
We conduct extensive experiments on Question Answering (QA) datasets taken from the ALCE benchmark and validate the model’s generalizability using EXPERTQA. On LLaMA-2-7B, the incorporation of fine-grained rewards achieves the best performance among the baselines, even surpassing that of GPT-3.5-turbo.111Our code will be released soon. |
2402.09668 | The training of large language models (LLMs) is expensive. In this paper, we study data-efficient approaches for pre-training LLMs,i.e., techniques that aim to optimize the Pareto frontier of model quality and training resource/data consumption. We seek to understand the tradeoffs associated with data selection routines based on (i) expensive-to-computedata-qualityestimates, and (ii) maximization ofcoverageand diversity-based measures in the feature space. Our first technique,Ask-LLM, leverages the zero-shot reasoning capabilities of instruction-tuned LLMs to directly assess the quality of a training example. To target coverage, we proposeDensitysampling, which models the data distribution to select a diverse sample. In our comparison ofsamplers, involving hundreds of evaluation tasks and pre-training runs, we find thatAsk-LLMandDensityare the best methods in their respective categories. Coverage sampling canrecoverthe performance of the full data, while models trained onAsk-LLMdata consistentlyoutperformfull-data training—even when we reject% of the original dataset, while converging up to% faster. |
2402.09906 | All text-based language problems can be reduced to either generation or embedding. Current models only perform well at one or the other. We introduce generative representational instruction tuning (GRIT) whereby a large language model is trained to handle both generative and embedding tasks by distinguishing between them through instructions. Compared to other open models, our resultingGritLM 7Bsets a new state of the art on the Massive Text Embedding Benchmark (MTEB) and outperforms all models up to its size on a range of generative tasks. By scaling up further,GritLM 8x7Boutperforms all open generative language models that we tried while still being among the best embedding models. Notably, we find thatGRITmatches training on only generative or embedding data, thus we can unify both at no performance loss. Among other benefits, the unification viaGRITspeeds up Retrieval-Augmented Generation (RAG) by > 60% for long documents, by no longer requiring separate retrieval and generation models. Models, code, etc. are freely available athttps://github.com/ContextualAI/gritlm. |
2305.18425 | In this paper, we present an efficient method for storing fine-tuned models by leveraging the low-rank properties of weight residuals. Our key observation is that weight residuals in large overparameterized models exhibit even stronger low-rank characteristics. Based on this insight, we propose Efficient Residual Encoding (ERE) , a novel approach that achieves efficient storage of fine-tuned model weights by approximating the low-rank weight residuals. Furthermore, we analyze the robustness of weight residuals and push the limit of storage efficiency by utilizing additional quantization and layer-wise rank allocation.
Our experimental results demonstrate that our method significantly reduces memory footprint while preserving performance in various tasks and modalities. We release our code.111Our code is available at:https://github.com/cloneofsimo/ere |
Subsets and Splits