Add comprehensive model card

README.md

@@ -1,22 +1,170 @@
 ---
-base_model: unsloth/Meta-Llama-3.1-8B-Instruct
-tags:
-- text-generation-inference
-- transformers
-- unsloth
-- llama
-- trl
-license: apache-2.0
+license: llama3.1
 language:
 - en
+tags:
+- llama
+- llama-3
+- gguf
+- quantized
+- tax
+- indian-law
+- finance
+base_model: meta-llama/Meta-Llama-3.1-8B-Instruct
+datasets:
+- custom
+pipeline_tag: text-generation
 ---
 
-# Uploaded  model
+# Llama 3.1 8B - Indian Income Tax Act 1961 (GGUF)
+
+Fine-tuned **Llama 3.1 8B Instruct** model specialized in the **Indian Income Tax Act 1961**. Optimized for tax law queries, compliance questions, and section references.
+
+## 🎯 Model Details
+
+- **Base Model**: [meta-llama/Meta-Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Meta-Llama-3.1-8B-Instruct)
+- **Fine-tuning Method**: LoRA (r=64, alpha=64) with Unsloth
+- **Training Date**: 20251201
+- **Context Length**: 4096 tokens
+- **Format**: GGUF (ready for llama.cpp, Ollama, LM Studio, Jan, etc.)
+- **Specialization**: Indian Income Tax Act 1961
+
+## 📦 Available Quantizations
+
+| File | Size | Use Case | RAM Required | Quality |
+|------|------|----------|--------------|---------|
+| `q4_k_m` | ~4.5GB | **Recommended** - Best balance | 6-8GB | ⭐⭐⭐⭐ |
+| `q5_k_m` | ~5.5GB | High quality responses | 8-10GB | ⭐⭐⭐⭐⭐ |
+| `q8_0` | ~8GB | Near-original quality | 10-12GB | ⭐⭐⭐⭐⭐ |
+| `f16` | ~15GB | Maximum quality (if available) | 18-20GB | ⭐⭐⭐⭐⭐ |
+
+### Quantization Guide
+- **q4_k_m**: Best for most users - good quality, reasonable size
+- **q5_k_m**: Better quality with slight size increase
+- **q8_0**: Minimal quality loss, larger file
+- **f16**: Full precision, largest file
+
+## 🚀 Quick Start
+
+### Using Ollama
+
+1. Download the model file (e.g., q4_k_m)
+wget https://huggingface.co/teclabs/Llama-3.1-8b-Instruct-Ind-Tax-Act-1961_Optimal/resolve/main/llama-tax-act-q4_k_m.gguf
+
+2. Create Modelfile
+cat > Modelfile << 'EOF'
+FROM ./llama-tax-act-q4_k_m.gguf
+
+PARAMETER temperature 0.7
+PARAMETER top_p 0.9
+PARAMETER top_k 40
+PARAMETER repeat_penalty 1.1
+
+SYSTEM """You are an expert on the Indian Income Tax Act 1961.
+Provide accurate, detailed information about tax regulations, exemptions,
+deductions, and compliance requirements. Always cite relevant sections
+when applicable."""
+EOF
+
+3. Create the model
+ollama create llama-tax-act -f Modelfile
+
+4. Run it
+ollama run llama-tax-act
+
+### Using llama.cpp
+
+Download model
+wget https://huggingface.co/teclabs/Llama-3.1-8b-Instruct-Ind-Tax-Act-1961_Optimal/resolve/main/llama-tax-act-q4_k_m.gguf
+
+Run inference
+./llama-cli -m llama-tax-act-q4_k_m.gguf -p "Explain Section 80C deductions:" -n 512
+
+Download model
+wget https://huggingface.co/teclabs/Llama-3.1-8b-Instruct-Ind-Tax-Act-1961_Optimal/resolve/main/llama-tax-act-q4_k_m.gguf
+
+Run inference
+./llama-cli -m llama-tax-act-q4_k_m.gguf -p "Explain Section 80C deductions:" -n 512
+
+from llama_cpp import Llama
 
-- **Developed by:** teclabs
-- **License:** apache-2.0
-- **Finetuned from model :** unsloth/Meta-Llama-3.1-8B-Instruct
+Load model
+llm = Llama(
+model_path="./llama-tax-act-q4_k_m.gguf",
+n_ctx=4096,
+n_threads=8,
+n_gpu_layers=35 # Adjust based on your GPU
+)
 
-This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
+Generate response
+output = llm(
+"What are the tax implications under Section 54?",
+max_tokens=512,
+temperature=0.7,
+top_p=0.9,
+)
+
+print(output['choices']['text'])
+
+
+## 💡 Example Queries
+
+Q: What are the deductions available under Chapter VI-A?
+Q: Explain Section 80C and its limit for FY 2023-24
+Q: What is the difference between Section 80C and 80D?
+Q: How is capital gains tax calculated under Section 112A?
+Q: What are the exemptions available under Section 10?
+
+
+## 📊 Training Details
+
+- **Training Hardware**: NVIDIA A100 80GB
+- **Training Time**: ~8 minutes (including quantization)
+- **Dataset**: Custom corpus from Income Tax Act 1961
+- **Epochs**: Optimized for convergence
+- **Learning Rate**: 2e-4 with cosine schedule
+- **Precision**: BF16 training, quantized for deployment
+
+## ⚙️ Technical Specifications
+
+- **Architecture**: Llama 3.1 (8B parameters)
+- **Vocabulary**: 128,256 tokens
+- **Max Context**: 4096 tokens
+- **Attention**: Grouped-Query Attention (GQA)
+- **Activation**: SwiGLU
+- **LoRA Rank**: 64 (higher than standard for better quality)
+
+## 📈 Performance
+
+- **Inference Speed** (q4_k_m on RTX 3090): ~40-50 tokens/sec
+- **Inference Speed** (q4_k_m on M1 Max): ~25-35 tokens/sec
+- **Quality**: Specialized responses with section references
+
+## ⚠️ Limitations
+
+- Trained on Income Tax Act 1961 as of training date (20251201)
+- May not reflect latest amendments after this date
+- Should be used as reference only, not legal advice
+- Always verify with official sources
+
+## 📜 License
+
+This model inherits the [Llama 3.1 Community License](https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/LICENSE).
+
+## ��� Acknowledgments
+
+- **Meta AI** for Llama 3.1 base model
+- **Unsloth AI** for efficient fine-tuning framework
+- **ggerganov** for llama.cpp and GGUF format
+
+## 📧 Contact
+
+For questions or issues, please open an issue on the [repository](https://huggingface.co/teclabs/Llama-3.1-8b-Instruct-Ind-Tax-Act-1961_Optimal/discussions).
+
+## 🔄 Updates
+
+**2025-12-01**: Initial release with q4_k_m, q5_k_m, q8_0 quantizations
+
+---
 
-[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
+**Disclaimer**: This model is for educational and research purposes. Tax laws are complex and subject to change. Always consult qualified tax professionals for advice.