Downloads · 30 days
20
29% of all-time downloads
bearcove/qwen3-asr-forced-aligner-0.6b-tq4_1s-text-tq4_1s-audio
qwen3-asr-forced-aligner-0.6b-tq4_1s-text-tq4_1s-audio is a machine learning model from bearcove. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
A TurboQuant-packed build of the Qwen3-ASR Forced Aligner (0.6B) for the bee / helix on-device Metal runtime. Given an audio clip and its known transcript, it emits per-word start/end timestamps (80 ms time-bucket res…
Downloads · 30 days
20
29% of all-time downloads
All-time downloads
70
Public
Parameters
575M
585 MB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors585 MB · 98%
How the weights are stored.
U8571M · 99%
From the Hugging Face model README
A TurboQuant-packed build of the Qwen3-ASR Forced Aligner (0.6B) for the bee / helix on-device Metal runtime. Given an audio clip and its known transcript, it emits per-word start/end timestamps (80 ms time-bucket resolution).
This is not a generic 🤗 Transformers checkpoint — model.safetensors
holds TurboQuant-quantized weights in helix's packed layout, with the
quantization index embedded in the safetensors header metadata. It loads with
the bee/helix engine, not AutoModel.
| tower | format | notes |
|---|---|---|
| text | tq4_1s | 4-bit TurboQuant blocks (32-value WHT block, dual fp16 scales) |
| audio | tq4_1s | 4-bit TurboQuant blocks |
Quantized from the bf16 Qwen/Qwen3-ASR-ForcedAligner-0.6B weights with
hx packer. Boundary accuracy was validated against a tq8-text oracle on a
116-clip cut corpus: every clip agrees with the tq8 reference to within one
80 ms time bucket (zero multi-bucket errors), at ~34% smaller on disk
(~558 MiB vs ~845 MiB).
model.safetensors — packed TQ weights + embedded turboquant.indexconfig.json — aligner config (classify_num, timestamp_segment_time,
token ids, text hidden_size)tokenizer.json, tokenizer_config.json, vocab.json, merges.txt — Qwen3 tokenizerchat_template.json, generation_config.json, preprocessor_config.json — auxUsed by bee to forced-align a carrier sentence during hotword enrollment, so the spoken hotword can be cut out at its true word boundaries.