[Bug] lfm2.5-thinking:1.2b — Tool calls ignored and data hallucinated; #73
Description
Bug : Inconsistent and broken tool call behavior
lfm2.5-thinking:1.2b exhibits multiple failure modes when tool calling is involved:
1. Wrong query / wrong arguments: The model executes the tool call but passes incorrect arguments — for example using a wrong column name that does not exist in the schema, causing the tool to failure, however complete schema is provided in the system prompt.
2. No response after successful tool call: Sometimes the model calls the tool correctly and receives a valid result, but then stops and returns no response to the user. The tool result is silently dropped.
Expected: Model calls the tool with correct arguments, receives the result, and always completes a full response based on that result.
Actual: Behavior is inconsistent across runs — wrong arguments, silent drops after tool result. There is no reliable pattern and the failures are non-deterministic.
**Attaching the 2. No response after successful tool call screenshot : **
