Resonant-Light

Speech recognition model in the Qwen3-ASR format (Qwen3ASRForConditionalGeneration, safetensors). It loads in vLLM without custom code, and in transformers through the qwen-asr package.

Usage (transformers)

pip install qwen-asr

import torch
from qwen_asr import Qwen3ASRModel

model = Qwen3ASRModel.from_pretrained("Reson8Labs/Resonant-Light", dtype=torch.bfloat16, device_map="cuda")
print(model.transcribe(audio="audio.wav", language="English")[0].text)

Usage (vLLM)

pip install "vllm[audio]"
vllm serve Reson8Labs/Resonant-Light --max-model-len 4096

import base64
from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")
audio = base64.b64encode(open("audio.wav", "rb").read()).decode()
r = client.chat.completions.create(
    model="Reson8Labs/Resonant-Light",
    messages=[{"role": "user", "content": [{"type": "audio_url", "audio_url": {"url": f"data:audio/wav;base64,{audio}"}}]}],
    temperature=0,
    extra_body={"chat_template_kwargs": {"language": "English"}},
)
print(r.choices[0].message.content)

Options (chat_template_kwargs, all optional):

Option	Default	Effect
language	unset	Forces the language (English name, e.g. "Dutch"). If unset, the output starts with language <Name><asr_text>.
digits	true	Numbers as digits; false writes them as words.
filler	false	true keeps fillers such as "um" and "uh".
verbatim	true	false collapses repeated words.