Qwen 3 VL 4B Instruct or Thinking?
#8
by Privac - opened
Is the text encoder used the Instruct or Thinking version of the model? Also why the LLM inference parameters found in the Comfy workflows are never using the right ones?
https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct#generation-hyperparameters
VL
export greedy='false'
export top_p=0.8
export top_k=20
export temperature=0.7
export repetition_penalty=1.0
export presence_penalty=1.5
export out_seq_length=16384
Text
export greedy='false'
export top_p=1.0
export top_k=40
export repetition_penalty=1.0
export presence_penalty=2.0
export temperature=1.0
export out_seq_length=32768
Also in this case only when enabling thinking mode the output will follow the instruction provided of "Then output a single expanded prompt paragraph." while otherwise it will not