Model_API/Model/model_onnx.py

import onnx
import onnxruntime
import torch
from transformers import BertForTokenClassification

from .config_train import model_load_path, onnx_path, tokenizer


# Convert Model to ONNX
def convert_to_onnx(model_path, tokenizer):
    """Convert the fine-tuned BERT token classification model to ONNX."""
    model = BertForTokenClassification.from_pretrained(model_path)
    model.eval()
    
    # Dummy input
    dummy_sentence = "Tôi muốn đi cắm trại ngắm hoàng hôn trên biển cùng gia đình"
    inputs = tokenizer(dummy_sentence, return_tensors="pt", padding=True, truncation=True)
    dummy_input_ids = inputs["input_ids"]
    dummy_attention_mask = inputs["attention_mask"]
    
    # Export ONNX model
    torch.onnx.export(
        model, 
        (inputs["input_ids"], inputs["attention_mask"]),  # Tuple of model inputs
        onnx_path,
        export_params=True,
        opset_version=14,  # Use Opset 14 or higher
        input_names=["input_ids", "attention_mask"],
        output_names=["logits"],
        dynamic_axes={"input_ids": {0: "batch_size", 1: "sequence_length"},
                      "attention_mask": {0: "batch_size", 1: "sequence_length"},
                      "logits": {0: "batch_size", 1: "sequence_length"}},
    )
    print(f"✅ ONNX model saved to {onnx_path}")


convert_to_onnx(model_load_path, tokenizer)