---
language: en
tags:
  - llama
  - fine-tuning
  - lora
  - education
  - question-answering
license: apache-2.0
models:
  - ibrahimBlyc/LA_Llama
datasets:
  - ibrahimBlyc/LA_dataset_blyc
library_name: transformers
pipeline_tag: text-generation
model_creator: ibrahimBlyc
model_type: llama
---


# Model Card: Fine-tuned LLaMA 3.2 Model

## Model Description

This model is a fine-tuned version of LLaMA 3.2, designed specifically for tasks in the domain of **learning analytics** and **education systems improvement**. It has been trained on a carefully curated dataset that includes question-answer pairs and dialogue data, ensuring high-quality responses tailored to educational and analytical contexts.

### Key Features:
- **Base Model**: LLaMA 3.2
- **Fine-tuning Approach**: Supervised fine-tuning with a question-answer structured dataset.
- **Domains Covered**: Education systems, learning analytics, review/meta-analysis literature, and strategies for academic success.

---

## Training Data

The fine-tuning dataset was crafted with precision to ensure the quality and relevance of the model's responses. The dataset contains thousands of entries with two primary formats:

1. **ShareGPT-style dialogues**:
   - Full discussions between a human and another actor (e.g., an AI) structured as interactive conversations.

2. **Alpaca-style question-answer pairs**:
   - Data structured with concise input and output information in a Q&A format.

### Dataset Creation Process:

#### **1. Literature-Based Question-Answer Pairs:**
- **Lens.org Collection**:
  - Papers filtered using keywords such as "review" and "meta-analysis".
  - Abstract sections were extracted for concise summaries of objectives, methods, and conclusions.
  - A Python program utilizing the Gemini API was used to generate relevant questions for each abstract.
  - **Data Size**: 14,000 question-answer pairs.

- **Scopus.com Collection**:
  - Focused on the keyword "learning analytics."
  - An additional **8,000 question-answer pairs** were generated using the same methodology.

#### **2. ChatGPT Recommendations for Education System Improvements:**
- High-quality recommendations generated by ChatGPT on topics such as:
  - Reducing dropout rates.
  - Combating academic failure.
  - Supporting student success.
- **Data Size**: 544 question-answer pairs.

#### Example of Dataset:
```json
[
  {
    "instruction": "What are the key factors influencing student success?",
    "output": "Key factors include teacher effectiveness, parental involvement, and access to educational resources."
  },
  {
    "instruction": "How can dropout rates be reduced?",
    "output": "Dropout rates can be reduced by implementing early intervention programs, providing mentorship opportunities, and addressing socio-economic barriers."
  }
]
```

### Dataset Highlights:
- Over **22,500 entries** spanning multiple sub-domains within education and learning analytics.
- Data curated to ensure clarity, relevance, and high-quality question-answer pairs.

---

## Model Performance

### **Intended Use Cases**
- **Education Research**: Assisting researchers and educators in analyzing learning trends and strategies.
- **Learning Analytics**: Providing insights into educational systems, success factors, and intervention strategies.
- **Academic Assistance**: Answering domain-specific questions in education.

### **Limitations**
- The model is fine-tuned for education and learning analytics; its performance in unrelated domains may vary.
- Limited coverage of topics outside the dataset's scope.

---

## Ethical Considerations
- The model may reflect biases present in the training data, such as those inherent in academic literature or AI-generated content.
- Users should verify critical outputs, especially in high-stakes scenarios such as policy-making or educational interventions.

---

## Citation
If you use this model in your research or applications, please cite:

```
@misc{llama3_finetuned_education,
  title={Fine-tuned LLaMA 3.2 for Learning Analytics},
  author={Ibrahim Belayachi},
  year={2025},
  howpublished={\url{https://huggingface.co/ibrahimBlyc/Llama_be_LA_}},
  note={Fine-tuned on education and learning analytics datasets}
}
```

---

## Contact
For questions or feedback, please contact Ibrahim Belayachi at ibrahim.belayachi@etu.utc.fr.