--- license: apache-2.0 base_model: openai/whisper-large-v3 pipeline_tag: automatic-speech-recognition library_name: onnxruntime-genai language: - multilingual tags: - onnx - onnxruntime-genai - whisper - automatic-speech-recognition --- # whisper-large-v3-genai ## Overview `openai/whisper-large-v3` exported to ONNX Runtime GenAI format with separate encoder and decoder ONNX files plus external data files. ## Source | Field | Value | |---|---| | Upstream model | [openai/whisper-large-v3](https://huggingface.co/openai/whisper-large-v3) | | Upstream source revision | `06f233fe06e710322aca913c1bc4249a0d71fce1` | | Packaging source revision | `e745b66a7380580b604215ee3dd83e28d9f1a2b9` | | Export tool/script | ONNX Runtime GenAI model builder (`onnxruntime-genai`) | | Quantization recipe | FP32 weights in GenAI encoder/decoder ONNX layout | ## Precision and Packaging Export tooling, precision, and quantization are recorded in the **Source** table above. This packaging mirror does not publish independent parity benchmarks; validate on your target execution provider before production use. ## Model Details - Source model: [`openai/whisper-large-v3`](https://huggingface.co/openai/whisper-large-v3) - Parameters: 1.5B - Format: ONNX Runtime GenAI encoder/decoder split - Runtime target: ONNX Runtime GenAI ## Files | File | Description | |---|---| | `encoder.onnx` + `encoder.onnx.data` | Encoder model | | `decoder.onnx` + `decoder.onnx.data` | Decoder model | | `genai_config.json` | ONNX Runtime GenAI session configuration | | `audio_processor_config.json` | Audio feature extraction configuration | | `tokenizer.json` + `tokenizer_config.json` | Tokenizer files | ## Intended Use Run the converted Whisper large-v3 automatic speech recognition model with ONNX Runtime GenAI. ## Training Data This repository does not document training data. See the upstream source model for training data information. ## Evaluation ## Limitations - Performance and accuracy may differ from the upstream model depending on runtime, execution provider, and decoding configuration. - Training data and evaluation details are inherited from the upstream model documentation, not reproduced here. ## License Apache-2.0, matching the upstream [`openai/whisper-large-v3`](https://huggingface.co/openai/whisper-large-v3) model.
