1 year ago

tools
ollama run rosemarla/qwen-go-experiment:fp16

Applications

Claude Code
Claude Code ollama launch claude --model rosemarla/qwen-go-experiment:fp16
Codex App
Codex App ollama launch codex-app --model rosemarla/qwen-go-experiment:fp16
OpenClaw
OpenClaw ollama launch openclaw --model rosemarla/qwen-go-experiment:fp16
Hermes Agent
Hermes Agent ollama launch hermes --model rosemarla/qwen-go-experiment:fp16
Codex
Codex ollama launch codex --model rosemarla/qwen-go-experiment:fp16
OpenCode
OpenCode ollama launch opencode --model rosemarla/qwen-go-experiment:fp16

Models

View all →

Readme

this is experimentation of finetuning qwen3-0.6 to play go in chat interface. it can render small game board, start a new game, process moves and make new moves with varying success rate.

this experimental version renders tiny go board 4x4 and doesn’t really play go. at most it may succeed in random legal moves.

prompts to try: new game. a4.

training data used for this model can be generated using following python code, and it can be adjusted to different board sizes.

https://notebin.io/nevent1qvzqqqq98ypzp2rwf5z0juhm84lm7cxephre4eqws6rtfhczfz7pa0rgc3pjfd8qqyt8wumn8ghj7un9d3shjtnwda6x2cnfdchxjmcqypfghq6r8gs5ddz408cfmr480j37hzcvxc4zqlprr6t4t0yt0ws86tx6p2e