# 🔐 SETUP INSTRUCTIONS FOR HUGGING FACE SPACE

## ⚠️ IMPORTANT: Add Your HuggingFace Token

The Llama 3.1 model is **gated** and requires authentication.

### Steps to Add Token:

1. **Get your HuggingFace token:**
   - Go to: https://huggingface.co/settings/tokens
   - Copy your token (starts with `hf_...`)

2. **Request access to Llama 3.1:**
   - Go to: https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct
   - Click "Request Access"
   - Wait for approval (usually instant)

3. **Add token to Space secrets:**
   - Go to: https://huggingface.co/spaces/sayande/agri-critique/settings
   - Scroll to "Repository secrets"
   - Click "New secret"
   - Name: `HF_TOKEN`
   - Value: `<your token here>`
   - Click "Add"

4. **Restart the Space:**
   - The space will automatically rebuild
   - Wait ~2-3 minutes for it to load

### ✅ After Setup:

Your space will be able to:
- Load Llama 3.1 8B base model
- Load your fine-tuned LoRA adapters
- Run full RAG pipeline
- Answer questions with evidence
- Validate answers for errors

### 🌐 Your Space:
https://huggingface.co/spaces/sayande/agri-critique

---

## 📊 System Overview:

**Model:** Llama 3.1 8B + LoRA (sayande/agri-critique-llama)
**Dataset:** FAISS index (sayande/agri-critique-index)  
**Hardware:** CPU (free) - upgrade to GPU T4 for faster performance

**Performance:**
- CPU: ~30-60s per query
- GPU T4: ~5-10s per query