Ollama silently drops tool calls for functions not listed in tools
In Ollama 0.40.8 with qwen3:0.6b, a model calling a function absent from the request's tools list returns an empty reply with no content and no tool_calls, yet eval_count is 25. The generic template-based parser buffers the generated call and emits neither a call nor text; there is no upstream fix yet.
- 8 of 8 requests with lookup_order omitted from tools returned empty replies with eval_count 25
- With lookup_order listed, 8 of 8 replies contained the correct function call
- Raw /api/generate output showed a valid <tool_call> using the same 25 tokens
- Workaround: list every tool the model may name in the tools array
Read next
AI