Upload README.md with huggingface_hub

README.md

@@ -1,6 +1,6 @@
 ---
-title: Crittiks AI Trainer
-emoji: 🧠
+title: VertexElite AI Tool Trainer
+emoji: 🚀
 colorFrom: purple
 colorTo: blue
 sdk: gradio
@@ -10,6 +10,48 @@ pinned: true
 license: mit
 ---
 
-# Crittiks AI Tool Trainer
+# VertexElite AI Tool Trainer
 
-Train tool-calling + fact-checking models on ZeroGPU.
+Fine-tune **tool-calling + function-calling** models on ZeroGPU (A10G).
+
+Train your own VertexElite AI with LoRA fine-tuning.
+
+## Features
+
+- QLoRA fine-tuning on free A10G GPU
+- Multiple base models: Qwen2.5, Phi-3.5, TinyLlama
+- Public datasets for tool calling
+- **Custom dataset upload** (JSONL/JSON)
+
+## Datasets
+
+| Type | Dataset | Description |
+|------|---------|-------------|
+| Tool Calling | NousResearch Hermes | Function calling conversations |
+| Tool Calling | Gorilla OpenFunctions | Berkeley benchmark |
+| Tool Calling | ToolBench | Tool use instructions |
+| Custom | Upload | Your own dataset.jsonl |
+
+## Custom Dataset Format
+
+Upload a `.jsonl` or `.json` file with one of these formats:
+
+```json
+{"text": "<|im_start|>user\nHello<|im_end|>\n<|im_start|>assistant\nHi!<|im_end|>"}
+{"messages": [{"role": "user", "content": "Hello"}, {"role": "assistant", "content": "Hi!"}]}
+{"instruction": "Say hello", "input": "", "output": "Hello there!"}
+```
+
+## Usage
+
+1. Select base model
+2. Choose datasets OR upload custom JSONL
+3. Configure LoRA (rank 16-32 recommended)
+4. Click "Start Training"
+5. Test model in Test tab
+
+## Limits
+
+- ZeroGPU: 5 min per GPU session
+- ~2000 samples per session
+- For longer training, run multiple sessions