Update README.md

README.md

@@ -1,70 +1,57 @@
-# Fine-tuned SmolLM2-135M with LoRA on CAMEL-AI Physics Dataset
+# fine-tuned-smolLM2-135M-with-LoRA-on-camel-ai-physics
 
-## Model Overview
-
-This model is a fine-tuned version of [HuggingFaceTB/SmolLM2-135M](https://huggingface.co/HuggingFaceTB/SmolLM2-135M) using Low-Rank Adaptation (LoRA) on the decontaminated [CAMEL-AI Physics](https://huggingface.co/datasets/camel-ai/physics) dataset. The dataset was decontaminated to ensure no overlap with the evaluation dataset `mmlu:college_physics`, ensuring a fair evaluation process.
-
-The fine-tuning leveraged PEFT (Parameter-Efficient Fine-Tuning) to optimize a smaller set of parameters, making it a lightweight yet effective fine-tuning approach.
+This model is a fine-tuned version of [HuggingFaceTB/SmolLM2-135M](https://huggingface.co/HuggingFaceTB/SmolLM2-135M) on the dataset [akhilfau/physics_decontaminated_2](https://huggingface.co/datasets/akhilfau/physics_decontaminated_2). This dataset was created by decontaminating the [camel-ai/physics](https://huggingface.co/datasets/camel-ai/physics) dataset from [mmlu:college_physics](https://huggingface.co/datasets/lighteval/mmlu).
 
 ---
 
-## Model Details
+## Model Performance
 
-- **Base Model**: [HuggingFaceTB/SmolLM2-135M](https://huggingface.co/HuggingFaceTB/SmolLM2-135M)
-- **Dataset Used for Fine-tuning**: [akhilfau/physics_decontaminated_2](https://huggingface.co/datasets/akhilfau/physics_decontaminated_2)
-- **Fine-tuning Methodology**: LoRA (Low-Rank Adaptation)
-- **Framework Versions**:
-  - PEFT: `0.13.2`
-  - Transformers: `4.46.2`
-  - PyTorch: `2.4.1+cu121`
-  - Datasets: `3.1.0`
-  - Tokenizers: `0.20.3`
+This model was evaluated on **MMLU: college_physics** using **LightEval**. The evaluation compared the base model (HuggingFaceTB/SmolLM2-135M) and the fine-tuned model (akhilfau/fine-tuned-smolLM2-135M-with-LoRA-on-camel-ai-physics). Results are as follows:
 
----
+### Evaluation Results:
 
-## Dataset Information
+| Model Name                                          | Task                        | Metric | Accuracy ± Stderr |
+|-----------------------------------------------------|-----------------------------|--------|-------------------|
+| **HuggingFaceTB/SmolLM2-135M**                     | mmlu:college_physics        | acc    | 0.2157 ± 0.0409   |
+| **akhilfau/fine-tuned-smolLM2-135M-with-LoRA-on-camel-ai-physics** | mmlu:college_physics        | acc    | 0.2843 ± 0.0449   |
 
-The training dataset ([akhilfau/physics_decontaminated_2](https://huggingface.co/datasets/akhilfau/physics_decontaminated_2)) was generated by decontaminating the [CAMEL-AI Physics](https://huggingface.co/datasets/camel-ai/physics) dataset to remove any overlap with the evaluation dataset `mmlu:college_physics`. This ensures that the fine-tuned model's performance on `mmlu:college_physics` is not biased due to data leakage.
+---
 
-- **Training Dataset**: Physics-related text data, where:
-  - `message_1`: The problem statement (e.g., a physics question).
-  - `message_2`: The solution or explanation.
+## Model Description
 
-The decontamination process used n-gram matching to eliminate any overlap with `mmlu:college_physics`.
+The fine-tuned model leverages **LoRA (Low-Rank Adaptation)** for parameter-efficient fine-tuning. The base model is SmolLM2-135M, which uses the **LlamaForCausalLM** architecture, and it was fine-tuned to enhance its understanding of physics-related questions and answers using the **akhilfau/physics_decontaminated_2** dataset.
 
 ---
 
-## Intended Use Cases
+## Training and Evaluation Data
 
-- Solving physics-related questions and problems in a Q&A format.
-- Educational purposes in the field of physics.
-- Benchmarking and comparison with other lightweight language models.
+### Dataset Details:
 
-### Limitations
+- **Training Dataset:** [akhilfau/physics_decontaminated_2](https://huggingface.co/datasets/akhilfau/physics_decontaminated_2)
+- **Evaluation Dataset:** [mmlu:college_physics](https://huggingface.co/datasets/lighteval/mmlu/viewer/college_physics)
 
-- The model is trained on a decontaminated dataset to ensure fairness during evaluation, but this process may exclude some valid training examples.
-- The model may require additional alignment or fine-tuning for tasks with different formats, such as multiple-choice questions (MCQs).
+The training dataset was decontaminated to ensure no overlap with the evaluation dataset for fair performance testing.
 
 ---
 
 ## Training Procedure
 
-### Hyperparameters
+### Training Hyperparameters
 
-The following hyperparameters were used during training:
-
-- **Learning Rate**: `0.0005`
-- **Train Batch Size**: `4`
-- **Eval Batch Size**: `4`
-- **Seed**: `42`
-- **Optimizer**: AdamW with `betas=(0.9, 0.999)` and `epsilon=1e-08`
-- **Learning Rate Scheduler**: `cosine`
-- **Number of Epochs**: `8`
+| Hyperparameter           | Value              |
+|---------------------------|--------------------|
+| Learning Rate            | 0.0005            |
+| Train Batch Size         | 4                 |
+| Eval Batch Size          | 4                 |
+| Seed                     | 42                |
+| Optimizer                | AdamW with betas=(0.9, 0.999), epsilon=1e-8 |
+| LR Scheduler Type        | Cosine            |
+| Number of Epochs         | 8                 |
 
 ### Training Results
 
 | Training Loss | Epoch | Step  | Validation Loss |
-|:-------------:|:-----:|:-----:|:---------------:|
+|---------------|-------|-------|-----------------|
 | 1.0151        | 1.0   | 4000  | 1.0407          |
 | 1.0234        | 2.0   | 8000  | 1.0087          |
 | 0.9995        | 3.0   | 12000 | 0.9921          |
@@ -76,38 +63,16 @@ The following hyperparameters were used during training:
 
 ---
 
-## Evaluation
-
-The model was evaluated using the physics subset of the `mmlu:college_physics` dataset. The training dataset was explicitly decontaminated to ensure fair evaluation.
-
----
-
-## Model Usage
-
-You can load this model from the Hugging Face Hub as follows:
-
-```python
-from transformers import AutoTokenizer, AutoModelForCausalLM
+## Intended Use
 
-# Load the fine-tuned model
-model_name = "akhilfau/fine-tuned-smolLM2-135M-with-LoRA-on-camel-ai-physics"
-tokenizer = AutoTokenizer.from_pretrained(model_name)
-model = AutoModelForCausalLM.from_pretrained(model_name)
-
-# Prepare a question
-input_text = "What is the Schrödinger equation?"
-inputs = tokenizer(input_text, return_tensors="pt")
-
-# Generate a response
-output = model.generate(**inputs, max_length=100)
-print(tokenizer.decode(output[0], skip_special_tokens=True))
-```
+This model is specifically fine-tuned for physics-related reasoning tasks and QA tasks. It may perform well on datasets that require understanding physics-related problems and concepts. Evaluation results show a measurable improvement compared to the base model on MMLU college physics tasks.
 
 ---
 
-## Acknowledgments
-
-- The decontamination process was implemented using tools from the [Cosmopedia Repository](https://github.com/huggingface/cosmopedia).
-- The model fine-tuning leveraged PEFT for efficient adaptation of the base model.
+## Framework Versions
 
-For any issues or contributions, feel free to open a pull request or an issue on the Hugging Face repository.
+- **PEFT**: 0.13.2
+- **Transformers**: 4.46.2
+- **Pytorch**: 2.4.1+cu121
+- **Datasets**: 3.1.0
+- **Tokenizers**: 0.20.3