Interleaved Thinking, minimax:tool_call parsing
#29
by
0xSero
- opened
Hello, for anyone hoping to selfhost this, here's a proxy that will make it compatible with Openai api clients, and anthropic api
It makes it much smarter if you're doing and minimax:tool_call right
This was made based off: https://huggingface.co/MiniMaxAI/MiniMax-M2/blob/main/docs/tool_calling_guide.md confirming it works in codex & claude code & chatbox
I believe the /responses api also works with interleaved thinking, which vLLM supports. Has anyone tested this?