{"id":26135825,"url":"https://github.com/programmer-rd-ai/intellihack_outlierrejects_task3","last_synced_at":"2025-03-11T00:06:09.451Z","repository":{"id":281475000,"uuid":"945387118","full_name":"Programmer-RD-AI/Intellihack_OutlierRejects_Task3","owner":"Programmer-RD-AI","description":"LLM Fine-tuning Challenge Enhancing Qwen 2.5 3B for AI Research QA","archived":false,"fork":false,"pushed_at":"2025-03-09T17:21:46.000Z","size":2221,"stargazers_count":1,"open_issues_count":1,"forks_count":0,"subscribers_count":1,"default_branch":"main","last_synced_at":"2025-03-11T00:03:27.330Z","etag":null,"topics":["ai","fine","fine-tuning","llm","qwen","tuning"],"latest_commit_sha":null,"homepage":"https://drive.google.com/drive/folders/18qAq284GfgvIfAZYQAJiMeMkcELDkEq9?usp=sharing","language":"Jupyter Notebook","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"gpl-3.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/Programmer-RD-AI.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2025-03-09T09:51:53.000Z","updated_at":"2025-03-10T01:23:53.000Z","dependencies_parsed_at":"2025-03-11T00:03:35.406Z","dependency_job_id":null,"html_url":"https://github.com/Programmer-RD-AI/Intellihack_OutlierRejects_Task3","commit_stats":null,"previous_names":["programmer-rd-ai/llm-fine-tuning-challenge-enhancing-qwen-2.5-3b-for-ai-research-qa","programmer-rd-ai/intellihack_outliearrejects_task3","programmer-rd-ai/intellihack_outlierrejects_task3"],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Programmer-RD-AI%2FIntellihack_OutlierRejects_Task3","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Programmer-RD-AI%2FIntellihack_OutlierRejects_Task3/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Programmer-RD-AI%2FIntellihack_OutlierRejects_Task3/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Programmer-RD-AI%2FIntellihack_OutlierRejects_Task3/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/Programmer-RD-AI","download_url":"https://codeload.github.com/Programmer-RD-AI/Intellihack_OutlierRejects_Task3/tar.gz/refs/heads/main","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":242945834,"owners_count":20210762,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["ai","fine","fine-tuning","llm","qwen","tuning"],"created_at":"2025-03-11T00:03:29.503Z","updated_at":"2025-03-11T00:06:09.423Z","avatar_url":"https://github.com/Programmer-RD-AI.png","language":"Jupyter Notebook","funding_links":[],"categories":[],"sub_categories":[],"readme":"# LLM Fine-tuning Challenge: Enhancing Qwen 2.5-3B for AI Research QA\n\nThis project demonstrates a comprehensive approach to fine-tuning the Qwen 2.5-3B model for specialized AI research question-answering. The implementation focuses on creating an efficient domain-specific QA system that can accurately answer questions about technical AI infrastructure concepts, particularly those related to distributed file systems and performance optimization.\n\n## 📥 Model Access\n\nDue to their large file sizes, the trained models are not included in this repository but can be accessed via the following links:\n\n- **Fine-tuned Model**: [Download from Google Drive](https://drive.google.com/drive/folders/1-1ntQq51rkWsvrOXY1HA73zlD8mZh8Kj?usp=sharing)\n- **Quantized Model**: [Download from Google Drive](https://drive.google.com/drive/folders/1-vSd2iDNG_9hxZ_H79MVPYO9HYKs8nRh?usp=sharing)\n- **Complete Project Files**: [Access on Google Drive](https://drive.google.com/drive/folders/18qAq284GfgvIfAZYQAJiMeMkcELDkEq9?usp=sharing) (where the project was run and tested)\n\n## 📋 Project Overview\n\nThe project implements a complete pipeline for:\n\n1. Processing technical research documents\n2. Generating high-quality synthetic QA pairs\n3. Fine-tuning Qwen 2.5-3B using QLoRA\n4. Building a retrieval-augmented generation (RAG) system\n5. Evaluating model performance using multiple metrics\n\n## 🧩 Components\n\n### Document Processing\n\n- Extracts structured information from technical markdown documents\n- Segments text into meaningful chunks for context preservation\n- Handles specialized formatting and technical content\n\n### QA Generation\n\n- Creates synthetic question-answer pairs from processed documents\n- Employs instruction templates optimized for technical QA formatting\n- Generates training and validation datasets\n\n### Fine-tuning Pipeline\n\n- Implements QLoRA (Quantized Low-Rank Adaptation) for efficient fine-tuning\n- Optimizes hyperparameters for the technical domain\n- Uses BitsAndBytes for quantization\n- Tracks training with Weights \u0026 Biases integration\n\n### RAG System\n\n- FAISS-based vector store for semantic document retrieval\n- Optimized embeddings for technical content\n- Context-aware question answering\n\n### Evaluation Framework\n\n- Multiple metrics including ROUGE, BLEU, and custom accuracy measures\n- Comprehensive evaluation of model output quality\n\n## 🚀 Usage\n\n### Prerequisites\n\n```bash\n# Clone the repository\ngit clone https://github.com/yourusername/LLM-Fine-tuning-Challenge-Enhancing-Qwen-2.5-3B-for-AI-Research-QA.git\ncd LLM-Fine-tuning-Challenge-Enhancing-Qwen-2.5-3B-for-AI-Research-QA\n\n# Install dependencies\nuv sync\n\n# Run\nuv run llm_fine_tuning_challenge_enhancing_qwen_2_5_3b_for_ai_research_qa.py\n```\n\n## 📊 Results\n\nThe fine-tuned model demonstrates significant improvements over the base model for technical AI research questions:\n\n- Higher accuracy in addressing complex technical concepts\n- Improved response quality for system architecture questions\n- Better context maintenance for multi-part technical explanations\n\n## 🧪 Dataset\n\nThe model is trained using the Q3 dataset containing detailed technical documentation about:\n\n- Fire-Flyer File System (3FS) architecture\n- Chain Replication with Apportioned Queries (CRAQ)\n- Performance optimizations for distributed systems\n- AI infrastructure components\n\n## 📃 License\n\nThis project is licensed under the GPL-3.0 License - see the LICENSE file for details.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fprogrammer-rd-ai%2Fintellihack_outlierrejects_task3","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fprogrammer-rd-ai%2Fintellihack_outlierrejects_task3","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fprogrammer-rd-ai%2Fintellihack_outlierrejects_task3/lists"}