-
update the model
-
Delete onnx
-
Rename onnx/decoder_model_int8.onnx to onnx/decoder_model_quantized.onnx
-
Rename onnx/decoder_with_past_model_int8.onnx to onnx/decoder_with_past_model_quantized.onnx
-
Rename onnx/encoder_model_int8.onnx to onnx/encoder_model_quantized.onnx
-
update wrongmodel
-
Delete onnx
-
update wrong file before
-
Delete encoder_model_int8.onnx
-
Delete decoder_with_past_model_int8.onnx
-
Delete decoder_model_int8.onnx
-
Delete onnx
-
upload model into new folder
-
add model configs and tokenizers for transformers.js compatibility
Add missing metadata and configuration files required by @xenova/transformers pipeline:
- config.json & generation_config.json
- preprocessor_config.json (Mel-spectrogram extraction)
- tokenizer.json & tokenizer_config.json -
Update README.md
-
upload dynamic INT8 quantized Whisper-tiny ONNX models
Initial upload of INT8 quantized ONNX weights exported from openai/whisper-tiny.
Key Details:
- Files uploaded: encoder_model_int8.onnx (~9.6 MB), decoder_model_int8.onnx (~47.5 MB), decoder_with_past_model_int8.onnx (~46.3 MB)
- Quantization: Dynamic Unsigned INT8 (QUInt8) via ONNX Runtime
- Total Repository Footprint: ~103.4 MB
- Active Runtime Footprint: ~56 MB (Encoder + KV-cache Decoder)
- Target Deployment: In-browser execution using ONNX Runtime WebAssembly (EdgeSpeech-WASM) -
initial commit
