nithinraok/parakeet-tdt-v2-encoder

Model Card for Model ID

Model Details

Model Description

Usage:

from transformers import ParakeetEncoder, AutoProcessor
from datasets import load_dataset, Audio
import torch

device = "cuda" if torch.cuda.is_available() else "cpu"

model_id = "nithinraok/parakeet-tdt-v2-encoder"
processor = AutoProcessor.from_pretrained(model_id)
model = ParakeetEncoder.from_pretrained(model_id).to(device)

ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
ds = ds.cast_column("audio", Audio(sampling_rate=processor.feature_extractor.sampling_rate))

# just take the first 2 samples
ds = ds.select(range(2))

model.eval()

for sample in ds:
    inputs = processor(sample["audio"]["array"])
    inputs.to(device, dtype=model.dtype)
    with torch.no_grad():
        outputs = model(**inputs)
    print(outputs.last_hidden_state.shape)
    print("sum:", outputs.last_hidden_state.sum())