Automatic Speech Recognition
NeMo
English
speech
audio
NeMo
parakeet
TDT
FastConformer
quantization
ternary
bitnet
quantization-aware-training
knowledge-distillation
Eval Results (legacy)
Instructions to use rajb3/kea-t with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- NeMo
How to use rajb3/kea-t with NeMo:
import nemo.collections.asr as nemo_asr asr_model = nemo_asr.models.ASRModel.from_pretrained("rajb3/kea-t") transcriptions = asr_model.transcribe(["file.wav"]) - Notebooks
- Google Colab
- Kaggle
Commit History
11 days ago
-
Model card and RUNTIME.md: add the INT8x2 research runtime's results (#2)
- Model card and RUNTIME.md: add the INT8x2 research runtime's results (7ace49ca203cb781b1f3fbbf04e40f945920dbee)c4a1f10 -
Model card: lead with size, accuracy and speed; move details to RUNTIME.md, EVALUATION.md and TRAINING.md (#1)
- Model card: lead with size, accuracy and speed; move details to RUNTIME.md, EVALUATION.md and TRAINING.md (97e07f7c016b660980f25652910d05274ab7af7a)c0f2896
22 days ago
-
Drop the former-name note
Verified signaturedb7d419 -
Rename model to Kea-T (repo moved from rajb3/parakeet-tdt-0.6b-v2-ternary)
Verified signaturebc1c883
27 days ago
-
Model card: add Used in section linking LocalFlow
Verified signature4cd127f
28 days ago
-
Publish RTX 5090 runtime speed, VRAM and energy results
Verified signatureDocument Compact and Expanded ternary runtimes, our optimized BF16 reference, and the off-the-shelf ONNX comparison. Preserve model metadata and recognition results.32750f3
29 days ago
-
Ternary Parakeet-TDT-0.6B-v2: 180.8 MB export, loader, model card, NOTICE, CC-BY-4.0
Verified signaturebb5a35d -
initial commit
Verified signaturef91bcb6