Tất cả mô hình
Q

Qwen3-ASR 1.7B

bởi QwenCông khaiapache-2.0
README.md

Qwen3-ASR 1.7B

Qwen3-ASR 1.7B is a medium speech-to-text and audio Q&A model from the Alibaba Qwen team, published under the Qwen organization on Hugging Face.

Links

  • Upstream model card: https://huggingface.co/Qwen/Qwen3-ASR-1.7B
  • Upstream GitHub: https://github.com/QwenLM/Qwen2.5-ASR

Model Facts

  • Task: Speech-to-Text (Transcription and Audio Q&A)
  • Parameters: 1.7B
  • License: Apache-2.0
  • Architecture: Qwen3-ASR GGUF

LA Studio Notes

Qwen3-ASR 1.7B in LA Studio runs offline using the CrispASR runtime. It downloads GGUF models from conversion repository cstr/qwen3-asr-1.7b-GGUF. It needs one of the following model files:

  • qwen3-asr-1.7b-q4_k.gguf (Recommended, 1.3 GB)
  • qwen3-asr-1.7b-q8_0.gguf (2.5 GB)
  • qwen3-asr-1.7b-f16.gguf (4.7 GB)

It supports standard multilingual speech transcription as well as free-form audio Q&A instructions via the Audio Q&A Prompt configuration setting.

Cấu hình tham số quy trình

Cấu hình các tham số tùy chỉnh này để tinh chỉnh quy trình tiếng nói và giọng nói.

Temperaturefloat

Controls randomness of speech generation. Higher values increase variety.

1
Seedint

Random seed for reproducibility. Set to -1 for random.

-1
Max Codec Stepsint

Maximum decoding steps for the neural voice audio synthesizer.

100