We used RL to train a model for MCP!
Connect any MCP client to any MCP server - you can run MCP workflows fully with local models (+ tune it further).
It works with Ollama / any MCP client that supports Qwen3 models - download it below 👇1/
Connect any MCP client to any MCP server - you can run MCP workflows fully with local models (+ tune it further).
It works with Ollama / any MCP client that supports Qwen3 models - download it below 👇1/
HF link: huggingface.co/osmosis-ai/osm…
Quickstart example link: github.com/Gulp-AI/Osmosi…
You can download it and run the model locally via Ollama/LM Studio, or host it on platforms like Fireworks AI, Groq, etc. that support Qwen3 models. 2/
Quickstart example link: github.com/Gulp-AI/Osmosi…
You can download it and run the model locally via Ollama/LM Studio, or host it on platforms like Fireworks AI, Groq, etc. that support Qwen3 models. 2/
We used Dr. GRPO on Qwen3-4B given existing function calling abilities, w/ VeRL + SGLang (i.e. training for multi-turn tool-calling). As seen above, the results are pretty good! 3/
If you end up using the model, let us know what you think!
And if you’re interested in a continuously improving custom version (for this model or any open source model), please reach out 🙂 4/
And if you’re interested in a continuously improving custom version (for this model or any open source model), please reach out 🙂 4/
• • •
Missing some Tweet in this thread? You can try to
force a refresh
