qwen3-asr-0.6b for the Ryzen AI NPU

Qwen3-ASR-0.6B, converted to run on the Ryzen AI NPU of AMD Strix Halo inside halogen-flash-server, beside the Flash model on the GPU and behind the same port. Questions go to the Discord. Bugs go to the server repository's issues.

These files run only inside the halogen-flash-server container, on the NPU. They will not load in transformers or any other runtime.

What is here

devices/                  30 MB  the NPU program for the speech decoder, and its NOTICE.md
enc/                     250 MB  the NPU program and weights for the audio encoder, and its NOTICE.md
qwen3-asr-0.6b.hnpw     1.29 GB  the decoder's weights, converted for the NPU
tokenizer/                11 MB  the tokenizer

Use it

Start halogen-flash-server 0.18.0 or later with HALOGEN_DOWNLOAD on and HALOGEN_NPU_MODELS=qwen3-asr-0.6b. The server fetches these files into /models/npu/qwen3-asr-0.6b/ and checks each one's size and checksum before it starts. It answers /v1/audio/transcriptions in the OpenAI request shape, for WAV recordings of up to 20 minutes or 25 MiB. Of the two sizes, this one is the lighter. One speech model is served at a time. With a speech model loaded beside other NPU models, images above 512x512 from flux2-klein-4b may not fit, and then answer an error that says so.

The files at the head of this repository need halogen-flash-server 0.18.12 or later; each server version downloads the revision it was released with, so an older server keeps working unchanged.

The host needs the NPU driver, XRT with its NPU plugin, and the GPU's fabric clock held while the NPU works beside the GPU. The setup and every request shape are in docs/NPU.md.

Source and license

Converted from Qwen/Qwen3-ASR-0.6B at revision 5eb144179a02acc5e5ba31e748d22b0cf3e303b0. Apache-2.0, as the source. devices/NOTICE.md and enc/NOTICE.md list the third-party notices the NPU programs carry.