lilfugu-transformers-8bit

8-bit quantized (bitsandbytes) version of lilfugu-transformers for CUDA / Linux. See the main model card for details.

2.2 GB — smallest CUDA variant.

Usage

pip install -U qwen-asr bitsandbytes

from qwen_asr import Qwen3ASRModel

model = Qwen3ASRModel.from_pretrained("holotherapper/lilfugu-transformers-8bit", device_map="auto")
result = model.transcribe("audio.wav", language="Japanese")
print(result[0].text)