push_to_hub.py

from transformers import Wav2Vec2ForCTC, Wav2Vec2CTCTokenizer, Wav2Vec2FeatureExtractor, Wav2Vec2Processor
import json

# Path to your local model directory and vocab file
local_model_path = './wav2vec2-base-mal'  # Directory with model checkpoints
vocab_path = './vocab.json'  # Path to your vocab.json file

# Hugging Face model ID (replace with your username)
model_id = "aoxo/wav2vec2-base-mal"

# Load vocab
with open(vocab_path, 'r') as f:
    vocab_dict = json.load(f)

# Create custom tokenizer
tokenizer = Wav2Vec2CTCTokenizer(
    vocab_path, 
    unk_token="[UNK]", 
    pad_token="[PAD]", 
    word_delimiter_token="|"
)

# Create feature extractor
feature_extractor = Wav2Vec2FeatureExtractor(
    feature_size=1, 
    sampling_rate=16000, 
    padding_value=0.0, 
    do_normalize=True, 
    return_attention_mask=False
)

# Create processor
processor = Wav2Vec2Processor(
    feature_extractor=feature_extractor, 
    tokenizer=tokenizer
)

# Load the model from the checkpoint directory
model = Wav2Vec2ForCTC.from_pretrained(local_model_path)

# Push to Hugging Face Hub
model.push_to_hub(model_id)
processor.push_to_hub(model_id)
tokenizer.push_to_hub(model_id)

print(f"Model, processor, and tokenizer successfully pushed to {model_id}")