| import torch | |
| from configuration_ltgbert import LtgBertConfig # Adjust this if you have a custom config class in modeling_ltgbert.py | |
| from modeling_ltgbert import LtgBertForMaskedLM # Import your Hugging Face wrapper | |
| # 1. Initialize Config and Model | |
| config = LtgBertConfig( | |
| attention_probs_dropout_prob=0.1, | |
| classifier_dropout=None, | |
| hidden_dropout_prob=0.1, | |
| hidden_size=384, | |
| intermediate_size=1024, | |
| layer_norm_eps=1e-07, | |
| max_position_embeddings=512, | |
| num_attention_heads=6, | |
| num_hidden_layers=12, | |
| output_all_encoded_layers=True, | |
| pad_token_id=4, | |
| position_bucket_size=32, | |
| vocab_size=6144 | |
| ) | |
| model = LtgBertForMaskedLM(config) | |
| # 2. Load the Custom Model Weights | |
| model_weights_path = "model_weights.pth" | |
| state_dict = torch.load(model_weights_path, map_location="cpu") | |
| model.load_state_dict(state_dict) | |
| # 3. Save the Model in Hugging Face Format | |
| output_dir = "./" | |
| model.save_pretrained(output_dir,safe_serialization=False) | |