Qwen3-ASR-1.7B MLX 8-bit (pinned mirror)

Byte-identical mirror of aufklarer/Qwen3-ASR-1.7B-MLX-8bit (an 8-bit MLX quantization of Qwen/Qwen3-ASR-1.7B, Apache-2.0), frozen at the revision validated against the Qwen3-ASR technical report's benchmarks: LibriSpeech test-clean WER 1.66%; FLEURS es 3.1 / de 3.8 / fr 5.0 WER, zh 2.34 / ko 1.99 / ja 5.86 CER (paper-style scoring).

Mirrored so a consuming app can pin provenance and verify SHA-256 checksums rather than tracking a third-party branch tip.

SHA-256:

1b76b3b6c655fc54595da025f7a96474ad9fa86363303fbdd61a7d8483ccfaf7  config.json
8831e4f1a044471340f7c0a83d7bd71306a5b867e95fd870f74d0c5308a904d5  merges.txt
bf304b009cc7eca79283056f787b44c952d24ac22cec787b39732bba3c23c13c  model.safetensors
0a5d0ec11188602242ff81a9969883d0fdeb98cd5d85cd1413089d897c201af5  model.safetensors.index.json
4942d005604266809309cabc9f4e9cb89ce855d59b14681fdc0e1cc62ea26c4c  tokenizer_config.json
ca10d7e9fb3ed18575dd1e277a2579c16d108e32f27439684afa0e10b1440910  vocab.json

Downloads last month: 2

Safetensors

Model size

0.8B params

Tensor type

BF16

U32

MLX

Hardware compatibility

8-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for ryancod/Qwen3-ASR-1.7B-MLX-8bit

Base model

Qwen/Qwen3-ASR-1.7B

Quantized

(43)

this model