bert-tiny-chinese-onnx

Model ckiplab/bert-tiny-chinese, converted to ONNX format for use with ONNX Runtime.

Original model

Parameter	Value
Architecture	BERT (BertForMaskedLM)
Number of layers	4
Hidden size	312
Attention heads	12
Intermediate size	1248
Vocabulary size	21128
Max sequence length	512
Activation function	GELU
Language	Traditional Chinese (zh)

For details: ckiplab/ckip-transformers

ONNX format

Export was performed using torch.onnx.export with dynamic axes for batch and sequence length.

Files:

• model.onnx – full model (encoder + pooler)
• config.json – model configuration

Usage

import onnxruntime as ort
import numpy as np

session = ort.InferenceSession("model.onnx")
inputs = {
    "input_ids": np.array([[101,  ... , 102]], dtype=np.int64),
    "attention_mask": np.array([[1, ..., 1]], dtype=np.int64),
    "token_type_ids": np.array([[0, ..., 0]], dtype=np.int64),
}
outputs = session.run(None, inputs)

License

The original model is distributed under GPL-3.0. This ONNX export is provided under Apache License 2.0.