Speech to text · NVIDIA
Canary-Qwen 2.5B
FastConformer encoder plus Qwen3-1.7B decoder; English ASR with LLM post-processing.
Datasheet
Maker | NVIDIA |
|---|---|
Parameters | 2.56 BVerifiedsafetensors.total from the HF API Source |
Licence | CC-BY-4.0Verified |
Access | Public download |
Task (HF pipeline) | automatic-speech-recognition |
First published on HF | 2025-06-26 |
Last updated on HF | 2026-04-21 |
Weights in memory | 5.1 GB at 16-bit · 1.6 GB at ~4.85 bitsEstimateParameters × bits ÷ 8. Excludes context cache and runtime overhead; not every model has a 4-bit build. |
Benchmark
6.37%
Open ASR Leaderboard, average WER (English short-form) · #1 of 7 scored in this category
avg 6.3686, RTFx 881.67. Source
Used in
The description summarises the model card at https://huggingface.co/nvidia/canary-qwen-2.5b. Parameters and licence are read from the Hugging Face API; the weights-in-memory line is our arithmetic, not a measurement.