Beta
Explore
Marketplace
Neural Labs
Chat
Wallet
Docs
qwen3-aligner-0.6b-ov-int8-stateless – AI Model by jinhwan000 | AlphaNeural AI
You can deploy this model and start earning money today!
jinhwan000
/
qwen3-aligner-0.6b-ov-int8-stateless
like
0
qwen3_asr
openvino
forced-alignment
qwen3
int8
Qwen/Qwen3-ForcedAligner-0.6B
finetune
apache-2.0
us
Views
No views yet
Model card
Files and Versions
Community
API
Deploy
qwen3-aligner-0.6b-ov-int8-stateless
OpenVINO IR export of
Qwen/Qwen3-ForcedAligner-0.6B
with INT8 weight compression,
stateless
language-model variant (empty past_key_values fed per call) for simple single-pass forced alignment on Intel CPU/iGPU.
Four thinker sub-models — embedding / audio (conv frontend) / audio_encoder / language_model (stateless, past_key_values.{i}.key/value inputs).
Word timestamps are decoded from timestamp-token positions: predicted token id x 80 ms (timestamp_segment_time), timestamp_token_id 151705.
Tokenizer/config files included. Warm alignment latency ~50-95 ms per utterance on an Intel Arc 130V iGPU with shape bucketing.
