Base architecture: EncDecCTCModelBPE (NeMo)
Calibration: 32 Persian clips from Reza2kn/persian-asr-eval-v0 (held out from eval).
Hardware target: NVIDIA Blackwell tensor cores.

Eval — `Reza2kn/persian-asr-eval-v0` (FLEURS-fa, 200 clips)

Variant	WER ↓	CER ↓	per-clip latency	peak VRAM
NVFP4 (this repo)	18.63%	5.85%	137 ms	4239 MiB

Usage

python

import nemo.collections.asr as nemo_asr
m = nemo_asr.models.ASRModel.restore_from("MohammadGholizadeh_parakeet-ctc-1.1b-persian.nemo-NVFP4.nemo").cuda().eval()
transcripts = m.transcribe(["clip.wav"])
print(transcripts[0])

License

Inherits the base model's license.

Reza2kn/MohammadGholizadeh_parakeet-ctc-1.1b-persian-NVFP4

MohammadGholizadeh_parakeet-ctc-1.1b-persian-NVFP4

Eval — Reza2kn/persian-asr-eval-v0 (FLEURS-fa, 200 clips)

Usage

License

Reza2kn/
MohammadGholizadeh_parakeet-ctc-1.1b-persian-NVFP4

Eval — `Reza2kn/persian-asr-eval-v0` (FLEURS-fa, 200 clips)