Eval bug: Model output directed to Reasoning block instead of standard response. #20265
Description
Name and Version
command is : llama-server -m Qwen3.5-9B-Claude-4.6-HighIQ-INSTRUCT-HERETIC-UNCENSORED.i1-Q4_K_M.gguf --host 0.0.0.0
version: llama-b8244-bin-win-cuda-12.4-x64
Operating systems
Windows
GGML backends
CUDA
Hardware
RTX 5070
Models
Qwen3.5-9B-Claude-4.6-HighIQ-INSTRUCT-HERETIC-UNCENSORED.i1-Q4_K_M.gguf
Problem description & steps to reproduce
Model output directed to Reasoning block instead of standard response.
First Bad Commit
No response