ysdede/whisper-large-v3-onnx-4graph

OpenAI Whisper Large v3 (non-turbo) four-graph ONNX

Source and attribution: https://huggingface.co/openai/whisper-large-v3 (OpenAI). FP16 export for ASRJS, split into encoder, decoder initialization, decoder step and alignment. External weight files are required. Native CPU JFK greedy transcription succeeded; artifact audit passed 5153 checks with no failures. This does not establish browser acceptance or general accuracy. This is the full 32-decoder-layer model, not Turbo, and its four graph bundles total approximately 6.3 GB.