Qwen 3.6: worse adherence?
Just swapped Qwen 3.5 for the 3.6 variant (FP8, RTX 6000 Pro) using the same recommended generation settings. My stack is vLLM (v0.19.0) + Open WebUI (v0.8.12) in a RAG setup where the model has access to several document retrieval tools. ​After some i…