whitphx

/

test-transformersjs-whisper-tiny

like 0

This automated migration updates the model repository for Transformers.js v3.

🔧 Model Binary Migration Results

Models processed: decoder_model.onnx, decoder_model_merged.onnx, decoder_with_past_model.onnx, encoder_model.onnx
✅ Added quantization modes (missing files):

• decoder_model.onnx: q4f16
• decoder_with_past_model.onnx: q4f16
• encoder_model.onnx: q4f16

📋 Summary

• Migration type: Model binaries quantization
• Purpose: Ensure all required quantized model variants are available and valid for Transformers.js v3
• Quantization modes: fp16, q8, int8, uint8, q4, q4f16, bnb4

✅ What was done:

1. Validation: Checked existing quantized model files for validity
2. Quantization: Generated missing or invalid quantized variants using transformers.js quantization tools
3. Post-validation: Verified all generated models load correctly with Transformers.js v3
4. Cleanup: Removed any invalid models that couldn't be regenerated

🔍 Please review:

• Verify that the quantized models work correctly with your use case
• Test the models in your application to ensure compatibility
• Check that file sizes are reasonable for your deployment needs

🤖 Generated by transformers-js-v3-migrator