Update README.md

README.md

@@ -1,199 +1,146 @@
 ---
-library_name: transformers
-tags: []
+language: en
+license: apache-2.0
+datasets:
+- derek-thomas/ScienceQA
+- allenai/ai2_arc
+tags:
+- education
+- stem
+- computer science
+- data science
+- engineering
+- biology
+- chemistry
 ---
 
-# Model Card for Model ID
 
-<!-- Provide a quick summary of what the model is/does. -->
+# STEMerald-2b
 
+**Model name:** STEMerald-2b
 
+**Model description:** 
+STEMerald-2b is a fine-tuned version of the Gemma-2b model, designed specifically for answering university-level STEM multiple-choice questions. This model leverages advanced fine-tuning techniques, including Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO), to enhance its accuracy and reliability in providing educational support.
 
-## Model Details
-
-### Model Description
-
-<!-- Provide a longer summary of what this model is. -->
-
-This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
-
-- **Developed by:** [More Information Needed]
-- **Funded by [optional]:** [More Information Needed]
-- **Shared by [optional]:** [More Information Needed]
-- **Model type:** [More Information Needed]
-- **Language(s) (NLP):** [More Information Needed]
-- **License:** [More Information Needed]
-- **Finetuned from model [optional]:** [More Information Needed]
-
-### Model Sources [optional]
-
-<!-- Provide the basic links for the model. -->
-
-- **Repository:** [More Information Needed]
-- **Paper [optional]:** [More Information Needed]
-- **Demo [optional]:** [More Information Needed]
-
-## Uses
-
-<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
-
-### Direct Use
-
-<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
-
-[More Information Needed]
-
-### Downstream Use [optional]
-
-<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
-
-[More Information Needed]
-
-### Out-of-Scope Use
-
-<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
-
-[More Information Needed]
-
-## Bias, Risks, and Limitations
-
-<!-- This section is meant to convey both technical and sociotechnical limitations. -->
-
-[More Information Needed]
-
-### Recommendations
-
-<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
-
-Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+<p align="center">
+    <img src="STEMerald_pic.jpeg" alt="STEMerald picture" width="400"/>
+</p>
 
-## How to Get Started with the Model
-
-Use the code below to get started with the model.
-
-[More Information Needed]
-
-## Training Details
-
-### Training Data
-
-<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
-
-[More Information Needed]
-
-### Training Procedure
-
-<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
-
-#### Preprocessing [optional]
-
-[More Information Needed]
-
-
-#### Training Hyperparameters
-
-- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
-
-#### Speeds, Sizes, Times [optional]
+## Model Details
 
-<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+**Base Model:** [Gemma-2b](https://arxiv.org/abs/2403.08295)
 
-[More Information Needed]
+**Architecture:** Decoder-only Language Model (Causal)
 
-## Evaluation
+**Parameters:** 2.51 billion
 
-<!-- This section describes the evaluation protocols and provides the results. -->
+**Quantized Version:** STEMerald-2b-4bit (with 4-bit NormalFloat)
 
-### Testing Data, Factors & Metrics
+**Training Framework:** PyTorch with Hugging Face Transformers
 
-#### Testing Data
+## Datasets
 
-<!-- This should link to a Dataset Card if possible. -->
+The model was fine-tuned on a variety of datasets tailored for STEM education, including:
 
-[More Information Needed]
+- **EPFL Preference Pairs Dataset:** 1522 university-level STEM questions with 26k preference pairs, annotated by students using ChatGPT-3.5 with Chain-of-Thought (CoT).
+- **Stack Exchange Dataset:** Questions and answers from various topics such as math, computer science, and engineering.
+- **Orca-Math:** 200k grade-school math word problems to enhance reasoning capabilities.
+- **EPFL MCQA Dataset**: Dataset of multiple-choice questions with explanation (for CoT) extracted from the winning pairs of EPFL preference pairs.
+- **ScienceQA:** Multiple-choice questions on biology, physics, chemistry, economics, earth science, and engineering practices.
+- **AI2 Reasoning Challenge (ARC):** Grade-school level multiple-choice science questions.
 
-#### Factors
+## Training Process
 
-<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+The training process for STEMerald-2b involved multiple steps:
 
-[More Information Needed]
+1. **Supervised Fine-Tuning (SFT):** Initial training on datasets like Orca-Math to improve reasoning abilities.
+2. **Direct Preference Optimization (DPO):** Training on preference pairs from EPFL and Stack Exchange datasets to align model outputs with preferred answers.
+3. **MCQA Fine-Tuning:** Specialization for multiple-choice question answering using datasets like ScienceQA and ARC.
 
-#### Metrics
+## Performance
 
-<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+The performance of STEMerald-2b was evaluated using various metrics:
 
-[More Information Needed]
+- **Accuracy:** The model achieved high accuracy across multiple test sets, demonstrating its effectiveness in answering STEM questions.
+- **Qualitative Evaluation:** The model's answers were evaluated for logical consistency, truthfulness, clarity, and coherence with the final answer.
 
 ### Results
 
-[More Information Needed]
-
-#### Summary
-
-
-
-## Model Examination [optional]
-
-<!-- Relevant interpretability work for the model goes here -->
-
-[More Information Needed]
-
-## Environmental Impact
-
-<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
-
-Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
-
-- **Hardware Type:** [More Information Needed]
-- **Hours used:** [More Information Needed]
-- **Cloud Provider:** [More Information Needed]
-- **Compute Region:** [More Information Needed]
-- **Carbon Emitted:** [More Information Needed]
+| Model Version                     | Accuracy (Non-Quantized) | Accuracy (Quantized) |
+|-----------------------------------|--------------------------|----------------------|
+| it-ORCA-DPO-MCQA _(STEMerald-2b)_ | 0.750                    | 0.720                |
+| it-DPO-MCQA                       | 0.744                    | 0.720                |
+| it-MCQA                           | 0.736                    | 0.700                |
+| it-ORCA-MCQA                      | 0.722                    | 0.714                |
+| MCQA                              | 0.702                    | 0.654                |
+| DPO-MCQA                          | 0.694                    | 0.674                |
+| Gemma-it-OneShot                  | 0.546                    | 0.520                |
+| Gemma-it                          | 0.518                    | 0.518                |
 
-## Technical Specifications [optional]
+Micro-averaged accuracy over three MCQA test sets(EPFL MCQA, ScienceQA and ARC).
 
-### Model Architecture and Objective
+## Use Cases
 
-[More Information Needed]
+STEMerald-2b can be utilized as a STEM course assistant, providing support in areas such as:
 
-### Compute Infrastructure
+- Answering university-level multiple-choice STEM questions.
+- Offering detailed explanations and reasoning for answers.
+- Enhancing student engagement and learning efficiency during independent studies.
 
-[More Information Needed]
+## Ethical Considerations
 
-#### Hardware
+While STEMerald-2b aims to provide accurate and helpful responses, it is important to consider potential ethical implications:
 
-[More Information Needed]
+- **Over-Reliance:** Students might become overly dependent on the model for answers, potentially affecting their independent learning and problem-solving skills.
+- **Accuracy:** Although efforts were made to ensure the truthfulness of responses, there is still a possibility of incorrect answers. Teacher supervision is crucial.
 
-#### Software
+## Limitations
 
-[More Information Needed]
+- The model's performance may vary based on the specific context and nature of the questions.
+- Quantization reduces memory footprint but may slightly affect accuracy.
 
-## Citation [optional]
+## Conclusion
 
-<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+STEMerald-2b offers a promising solution for enhancing STEM education through advanced language model capabilities. By leveraging fine-tuning techniques and comprehensive datasets, it aims to provide accurate and accessible learning support for students.
 
-**BibTeX:**
+## How to Use
 
-[More Information Needed]
+You can use the model directly with the `transformers` library:
 
-**APA:**
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
 
-[More Information Needed]
+tokenizer = AutoTokenizer.from_pretrained("matsant01/STEMerald-2b")
+model = AutoModelForCausalLM.from_pretrained("matsant01/STEMerald-2b")
 
-## Glossary [optional]
+input_text = "Question: What is the derivative of x^2? \nOptions: A. 4x B. 2*x^2 C. 2x D. 2\nAnswer:"
+inputs = tokenizer(input_text, return_tensors="pt")
+outputs = model.generate(**inputs)
+print(tokenizer.decode(outputs[0], skip_special_tokens=True))
+```
 
-<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+For the quantized version, use:
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 
-[More Information Needed]
+quantization_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_compute_dtype=torch.bfloat16,
+    bnb_4bit_quant_type="nf4"
+)
 
-## More Information [optional]
+tokenizer = AutoTokenizer.from_pretrained("matsant01/STEMerald-2b-4bit")
+model = AutoModelForCausalLM.from_pretrained("matsant01/STEMerald-2b-4bit", quantization_config=quantization_config)
+```
 
-[More Information Needed]
 
-## Model Card Authors [optional]
+## Acknowledgements
 
-[More Information Needed]
+We acknowledge the contributions of the EPFL and Stack Exchange communities for their invaluable datasets, and the Hugging Face team for their support and tools that made this project possible.
 
-## Model Card Contact
+## Contact
 
-[More Information Needed]
+For any questions or feedback, please contact:
+- [Antonio Mari](https://github.com/antoniomari) (antonio.mari@epfl.ch)
+- [Matteo Santelmo](https://github.com/matsant01) (matteo.santelmo@epfl.ch)
+- [Stefano Viel](https://github.com/stefanoviel) (stefano.viel@epfl.ch)