Vibe-coded proof-of-concept for finetuning LLMs with evolution strategy (ES) and distributed GPU workers communicating via http.
webinterface.mp4
- Distributed evolution strategy with dynamic population scaling
- Lightweight Python client
- Web interface for real-time monitoring
- Lightweight model state broadcasting
- Python 3.12
- vLLM
# Create and activate environment
conda create -n myenv python=3.12 -y
conda activate myenv
# Install dependencies
pip install --upgrade uv
uv pip install vllm --torch-backend=auto
## Usage
```python vllm_client.py --url https://odesapi.schumann.pub --project_id 1```