| license: mit | |
| pipeline_tag: image-to-text | |
| tags: | |
| - onnx | |
| - ocr | |
| - glm-ocr | |
| base_model: zai-org/GLM-OCR | |
| # GLM-OCR ONNX (Decoder) | |
| ONNX export of the **decoder** of [zai-org/GLM-OCR](https://huggingface.co/zai-org/GLM-OCR). Exported with `scripts/export_glm_ocr_onnx.py` (Transformers 5.1.0, custom torch.onnx path). | |
| ## Contents | |
| - `glm_ocr_decoder.onnx` / `glm_ocr_decoder.onnx.data` – Decoder ONNX (inputs: `decoder_input_ids`, `encoder_hidden_states`; output: `logits`). | |
| - `tokenizer.json`, `tokenizer_config.json` – Tokenizer from zai-org/GLM-OCR. | |
| ## Note | |
| The **vision encoder** was not exported (model forward requires either `input_ids` or `inputs_embeds` when called with image inputs only). To run full OCR you need encoder hidden states from another source or the original PyTorch model for the vision part. | |
| ## Usage | |
| Load with ONNX Runtime; feed `encoder_hidden_states` (from your vision encoder or zai-org/GLM-OCR in PyTorch) and `decoder_input_ids`; get `logits` and decode with the included tokenizer. | |
| ## Source | |
| - Base model: [zai-org/GLM-OCR](https://huggingface.co/zai-org/GLM-OCR) | |
| - Export spec: See [Docs/GLM_OCR_ONNX_Export.md](https://github.com/...) in the TranslateBlue repo. | |