fun_asr

Model type

fun_asr

Class

FunASRForConditionalGeneration

Task

fun-asr-speech-language

Source

models/fun_asr.py

Description

Fun-ASR-Nano composite model for speech recognition.

Contains:

  • audio_tower: Audio encoder + adaptor (fbank → LLM-dim features)

  • embedding: Text/audio embedding fusion (no adaptor)

  • decoder: Text decoder with KV cache (Qwen3-based)

The 3-model split for ONNX export is handled by the fun-asr-speech-language task.

HuggingFace class: FunASRForConditionalGeneration

Usage

mobius build --model <MODEL_ID> output_dir/

from mobius import build

model = build("<MODEL_ID>")