| Qwen3-ASR-ComfyUI | |
| Qwen3-ASR-1.7B, Qwen3-ASR-0.6B and Qwen3-ForcedAligner-0.6B, Copyright 2026 The Alibaba Qwen team, licensed under | |
| the Apache License 2.0 (see LICENSE). https://huggingface.co/Qwen/Qwen3-ASR-1.7B · | |
| https://huggingface.co/Qwen/Qwen3-ASR-0.6B · https://huggingface.co/Qwen/Qwen3-ForcedAligner-0.6B | |
| Changes made by PedroMarinhoDev with tools/convert.py from https://github.com/PedroMarinhoDev/ComfyUI-Drama3: | |
| the checkpoints' tensors are stored unchanged in one file per model, renamed (thinker.audio_tower. to audio_tower., | |
| thinker.model. to decoder.model.), without the LM head, which equals the embedding table, and with the model | |
| config in the file's metadata. Qwen3-ForcedAligner-0.6B keeps its timestamp head, renamed timestamp_head.weight. | |