ComfyUI Node
Qwen Thinking Loader (Merged or Sharded)
A ComfyUI node in Qwen/Thinking with 6 inputs and 3 outputs.
Qwen Thinking Loader (Merged or Sharded)
- model
- tokenizer
- model_out
◄model_folder▾►
◄devicecuda►
◄dtypebf16►
◄compile_modeltrue►
◄use_multi_gputrue►
◄keep_loadedtrue►
CategoryQwen/Thinking
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| model_folder | COMBO | Select your Qwen model folder | |
| device | COMBO | cuda | 2 options: cuda, cpu |
| dtype | COMBO | bf16 | 4 options: bf16, fp16, fp32, auto |
| compile_model | BOOLEAN | true | Enable torch.compile for faster inference after first run |
| use_multi_gpu | BOOLEAN | true | Use all visible GPUs (device_map='auto'). Turn off for single GPU only. |
| keep_loaded | BOOLEAN | true | Keep model in memory for faster repeated use |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| model | QWEN_THINKING_MODEL | — |
| tokenizer | QWEN_TOKENIZER | — |
| model_out | QWEN_THINKING_MODEL | — |