whisper-large-v3

Whisper Large V3 — Safetensors Version

This is a custom packaged version of OpenAI's Whisper Large V3 model, converted to safetensors format for safer and faster loading. This repo includes both the model weights and tokenizer files required for ASR (Automatic Speech Recognition) tasks.

Files Included

• model.safetensors: Model weights in safetensors format
• tokenizer_config.json: Tokenizer configuration
• vocab.json: Vocabulary file
• merges.txt: BPE merges
• special_tokens_map.json: Special token mapping

Usage (Python)

from transformers import WhisperForConditionalGeneration, WhisperTokenizer

model = WhisperForConditionalGeneration.from_pretrained("Zvatlov/whisper-large-v3")
tokenizer = WhisperTokenizer.from_pretrained("Zvatlov/whisper-large-v3")

Example Inference

import torch
from transformers import WhisperProcessor, WhisperForConditionalGeneration

processor = WhisperProcessor.from_pretrained("Zvatlov/whisper-large-v3")
model = WhisperForConditionalGeneration.from_pretrained("Zvatlov/whisper-large-v3")

# Load audio
from datasets import load_dataset
ds = load_dataset("hf-internal-testing/librispeech_asr_demo", "clean", split="validation")
input_audio = ds[0]["audio"]["array"]

# Prepare input
inputs = processor(input_audio, return_tensors="pt")
with torch.no_grad():
    generated_ids = model.generate(inputs["input_features"])

# Decode output
transcription = processor.batch_decode(generated_ids, skip_special_tokens=True)
print(transcription[0])

Notes

• Tensor type: FP16
• Total parameters: ~1.5B
• This model is not fine-tuned; original Whisper weights were packaged for safe and fast loading.