Update README.md

README.md

@@ -1,80 +1,93 @@
 ---
-language:
-- en
 library_name: transformers
-license: cc-by-4.0
 pipeline_tag: question-answering
+license: mit
+base_model: meta-llama/Llama-3.1-8B-Instruct
+tags: []
 ---
 
-# Model Card for Llama-3.1-8B-Instruct LoRA for Knowledge Incorporation
+# Model Card for How Much Knowledge Can You Pack into a LoRA Adapter without Harming LLM?
 
-This model is a Low-Rank Adaptation (LoRA) of Llama-3.1-8B-Instruct, designed to enhance its question-answering capabilities by incorporating new knowledge, as described in the paper [How Much Knowledge Can You Pack into a LoRA Adapter without Harming LLM?](https://arxiv.org/abs/2502.14502).
+This model card describes a LoRA model presented in [How Much Knowledge Can You Pack into a LoRA Adapter without Harming LLM?](https://arxiv.org/abs/2502.14502).
 
 ## Model Details
 
-- **Developed by:** Sergey Pletenev et al.
-- **Model type:** `LlamaForCausalLM` with LoRA
+### Model Description
+
+The performance of Large Language Models (LLMs) on many tasks is greatly limited by the knowledge learned during pre-training and stored in the model's parameters. Low-rank adaptation (LoRA) is a popular and efficient training technique for updating or domain-specific adaptation of LLMs. In this study, we investigate how new facts can be incorporated into the LLM using LoRA without compromising the previously learned knowledge. We fine-tuned Llama-3.1-8B-instruct using LoRA with varying amounts of new knowledge. Our experiments have shown that the best results are obtained when the training data contains a mixture of known and new facts. However, this approach is still potentially harmful because the model's performance on external question-answering benchmarks declines after such fine-tuning. When the training data is biased towards certain entities, the model tends to regress to few overrepresented answers. In addition, we found that the model becomes more confident and refuses to provide an answer in only few cases. These findings highlight the potential pitfalls of LoRA-based LLM updates and underscore the importance of training data composition and tuning parameters to balance new knowledge integration and general model capabilities.
+
+
+- **Developed by:** Sergey Pletenev, Maria Marina, Daniil Moskovskiy, Vasily Konovalov, Pavel Braslavski, Alexander Panchenko, Mikhail Salnikov
+- **Model type:** LLM
 - **Language(s) (NLP):** English
-- **License:** CC-BY-4.0
+- **License:** mit
 - **Finetuned from model:** meta-llama/Meta-Llama-3.1-8B-Instruct
 
 ### Model Sources
 
-- **Repository:** [https://github.com/memyprokotow/knowledge_lora](https://github.com/memyprokotow/knowledge_lora)
+- **Repository:** [https://github.com/memyprokotow/knowledge-lora-adapt](https://github.com/memyprokotow/knowledge-lora-adapt)
 - **Paper:** [How Much Knowledge Can You Pack into a LoRA Adapter without Harming LLM?](https://arxiv.org/abs/2502.14502)
-- **Datasets:**
-    - [Dbpedia dump](https://databus.dbpedia.org/dbpedia/mappings/mappingbased-objects)
-    - [Precollected triples and questions](https://drive.google.com/file/d/1pCtfRlvBW769384AgmfNBpIU8OmftfKd/view?usp=sharing)
-    - [Questions with labelled knowledge categories](https://drive.google.com/file/d/1-NDeTa8TMRNY9UIsIqtI-Iw4vq-rda35/view?usp=sharing)
 
 
 ## Uses
 
 ### Direct Use
 
-This model can be used for question-answering tasks, particularly those involving the new knowledge incorporated during fine-tuning. It is designed to be used with the base model `meta-llama/Meta-Llama-3.1-8B-Instruct`.
+The model can be used for question answering.
 
 ### Downstream Use
 
-This model can be further fine-tuned or used as a starting point for research on knowledge incorporation into LLMs.
+The model can be further fine-tuned for domain-specific question answering.
 
 ### Out-of-Scope Use
 
-This model should not be used for generating harmful, biased, or misleading content.  Its performance on general question-answering benchmarks might be impacted after fine-tuning with specific knowledge.
+The model may not perform well on questions outside the knowledge it has been fine-tuned on, or if the training data was biased.
 
 ## Bias, Risks, and Limitations
 
-This model inherits the biases present in the base Llama-3.1-8B-Instruct model.  Furthermore, the focused fine-tuning may introduce biases related to the new knowledge incorporated. The paper highlights potential performance decline on external question-answering benchmarks and a tendency to over-represent answers related to prominent entities in the training data.
+The model may exhibit biases present in the training data. The model's performance may degrade on external question-answering benchmarks after fine-tuning, especially if the training data is biased towards certain entities.
 
 ### Recommendations
 
-Users should be aware of the potential biases and limitations of the model. Careful attention should be paid to the composition and balance of the training data to mitigate biases and preserve general question-answering capabilities.
-
+Users should be aware of potential biases in the model's responses and the limitations of its knowledge.
 
 ## How to Get Started with the Model
 
-See the Github repository for detailed instructions on training and using the LoRA adapter with the base Llama model.
+[More Information Needed]
 
 ## Training Details
 
 ### Training Data
 
-The model is fine-tuned on a dataset generated using the head-to-tail pipeline with DBpedia as the knowledge source. The data includes known facts, potentially known facts, and unknown facts categorized based on the base model's pre-training knowledge. See the "Data" section of the Github README for details.
+The training data consists of questions and answers generated using the head-to-tail pipeline with a Dbpedia script.  See the paper and Github repository for more details.
+Model was trained on 3000 Unknown questions with 10 additional Paraphrased question per Unknown
 
 ### Training Procedure
 
-The model is trained using the LoRA technique.  Refer to the `lora_train_llama.py` script in the Github repository for training parameters and instructions.
+The model was fine-tuned using LoRA.
 
+#### Training Hyperparameters
 
+    LR = 1e-3
+    BS = 8
+    EPOCHS = 10
+    LoRA:
+    lora_rank = 1
+    lora_alpha = 2
+    use_rslora = True
+    lora_dropout = 0.1
+    bias = "none"
+    target_modules = ["down_proj", "gate_proj", "up_proj"]
+    task_type = "CAUSAL_LM"
 
 ## Evaluation
 
-The paper evaluates the model's performance using a reliability score and investigates different knowledge integration scenarios. See the paper for detailed results and analysis.
-
+For evaluation you can use [notebooks](https://github.com/AIRI-Institute/knowledge-packing/tree/main/notebooks) from github repository
 
 ## Environmental Impact
 
-The environmental impact information is not available in the original README. Users can estimate the carbon emissions using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+[More Information Needed]
+
 
 ## Citation
 
@@ -88,4 +101,8 @@ The environmental impact information is not available in the original README. Us
       primaryClass={cs.CL},
       url={https://arxiv.org/abs/2502.14502}, 
 }
-```
+```
+
+**APA:**
+
+Pletenev, S., Marina, M., Moskovskiy, D., Konovalov, V., Braslavski, P., Panchenko, A., & Salnikov, M. (2025). How Much Knowledge Can You Pack into a LoRA Adapter without Harming LLM?.