• I am running the latest code from the JamePeng/llama-cpp-python branch.
• I carefully followed the README.md.
• I have verified the issue is not a duplicate.
• I have tested it using the official binary llama-cli or llama-server provided by llama.cpp, and the problem (exists/does not exist) still exists.
• I reviewed the Discussions, and have a new bug or useful enhancement to share.

Environment & Hardware Configuration

Please provide your specific setup. Use the suggested commands for your OS to verify.

• OS Version: Windows
• CPU: AMD Ryzen 5 3600 6-Core Processor
• RAM: 16 GB
• GPU: NVIDIA RTX 3060 12GB
• Python: 3.12

Model & Logic Context

• Model source: HuggingFace
• Model: Qwen3.5-VL

Failure Timing & Logs

The issue is occurring during Runtime.

Analysis & Brainstorming

I've noticed that tool_choice isn't taken into account in multimodal models such as Qwen3.5. The tools are passed and executed for every request as if tool_choice were set to “required” by default. I suspect the reason might be that the code doesn't pass this parameter when rendering the chat template.

I've made sure to set tool_choice to None. I've also tried other tool_choice like "auto".