README.md

---
title: Fathom R1 Chat
emoji: 🤖
colorFrom: indigo
colorTo: blue
sdk: docker
pinned: false
---

# Fathom R1 Chat — Full‑stack (React + FastAPI)

ChatGPT‑style UI on React + a FastAPI backend that calls **FractalAIResearch/Fathom-R1-14B** via `transformers`.

## Run with Docker (GPU)

> Requires an NVIDIA GPU + NVIDIA Container Toolkit.

```bash
docker build -t fathom-r1-chat .
docker run --gpus all -p 8000:8000           -e MODEL_ID=FractalAIResearch/Fathom-R1-14B           -e QUANTIZE=auto           fathom-r1-chat
# Open http://localhost:8000
```

### Notes
- Model is derived from **DeepSeek-R1-Distill-Qwen-14B** and targets **16K context** usage. Use the tokenizer chat template.
- For long answers, bump `max_new_tokens` in the request.
- If you need private HF access, pass `-e HUGGING_FACE_HUB_TOKEN=...`.

## Dev mode (run separately)
```bash
# backend
cd backend
python3 -m venv .venv && source .venv/bin/activate
pip install --index-url https://download.pytorch.org/whl/cu121 torch==2.4.1+cu121
pip install -r requirements.txt
uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload

# frontend (new terminal)
cd frontend
npm ci
npm run dev
```

## API
- `POST /api/chat` with `{ messages: [{role, content}, ...], max_new_tokens, temperature, top_p }` → `{ reply, model }`

## Hardware
- 14B parameter model; for comfortable generation use **>=24–40 GB VRAM** or 4/8‑bit quantization on 16–24 GB GPUs.

## License
- MIT (model card states MIT) and this template is MIT.