invincibleambuj/Ambuj-Tripathi-Llama-8B-LoRA
Ambuj-Tripathi-Llama-8B-LoRA is an 8-billion parameter Llama 3.1-based model, fine-tuned by Ambuj Kumar Tripathi using QLoRA for specialized Indian legal question-answering. This model focuses on the Constitution of India, Indian Penal Code, and Code of Criminal Procedure, demonstrating fine-tuning on complex domain-specific text. It is designed for educational, research, and AI skill-development purposes, particularly for studying hallucination patterns in legal QA tasks.
Loading preview...
Overview
This model, Ambuj-Tripathi-Llama-8B-LoRA, is an 8-billion parameter Llama 3.1-based language model fine-tuned by Ambuj Kumar Tripathi. It utilizes QLoRA (4-bit quantization + LoRA rank 64) on the Meta-Llama-3.1-8B-Instruct base model. The fine-tuning process involved 1,820 steps over two epochs, achieving a final loss of 0.566, indicating consistent convergence.
Key Capabilities & Training
- Domain-Specific Expertise: Specialized in Indian legal question-answering, trained on 14,543 QA pairs covering the Constitution of India, Indian Penal Code (IPC), and Code of Criminal Procedure (CrPC).
- Efficient Fine-tuning: Achieved with Unsloth, TRL, and PEFT frameworks, making it suitable for resource-constrained environments (trained on Kaggle Free Tier).
- Local Inference: Optimized for CPU-first inference, with GGUF format (Q4_K_M) available for easy deployment in tools like LM Studio, Ollama, and llama.cpp.
Intended Use & Limitations
- Educational & Research Focus: Primarily built for educational, research, and AI skill-development purposes, specifically to explore fine-tuning on complex domain-specific text and study hallucination patterns in legal QA tasks.
- Not for Legal Advice: Crucially, this model is NOT intended to provide legal advice and should not be used as a substitute for a qualified legal professional.
- Known Limitations: May hallucinate specific Section/Article numbers and is only trained on a subset of Indian legal acts (Constitution, IPC, CrPC), excluding others like BNS/BNSS/IT Act/Consumer Protection. Verification against official sources is always necessary.