valiant9000/bert4ner-base-chinese-onnx

bert4ner-base-chinese (ONNX)

ONNX export of shibing624/bert4ner-base-chinese at revision 5d660ed2aa9da482bf2d99c6bc8cf2ce66758f6a, for use with Transformers.js / ONNX Runtime. Weights are unchanged (FP32); only the format was converted.

• Labels: O, B-/I-PER, B-/I-ORG, B-/I-LOC, B-/I-TIME (see config.json)
• Layout: onnx/model.onnx plus tokenizer.json (BertNormalizer, vocab identical to bert-base-chinese)
• Inputs: input_ids, attention_mask, token_type_ids; output logits

Export

optimum-cli export onnx -m <local snapshot of the pinned revision> --task token-classification out/

Tool versions: optimum 2.1.0, optimum-onnx 0.1.0, transformers 4.57.6, torch 2.15.0+cpu. Exported logits match the PyTorch model within 4e-6 (max abs) with identical argmax labels on a check sentence.

License

Apache-2.0, inherited from the original model. See LICENSE.txt and NOTICE.txt. All credit for training goes to the original author (shibing624, nerpy).