-
Notifications
You must be signed in to change notification settings - Fork 142
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Whisper Example on ORT master #293
Comments
Thanks for opening this issue. There was a recent change in the ORT that makes the I have updated the README and dependencies to version 1.15.0 in this PR #305 If needed, we could update the user script to support ORT versions > 1.15.0. |
Hi @jambayk Thanks for the response. Is it possible to update the user script as you suggested to support ORT > 1.15? |
## Describe your changes This PR adds support for the latest ORT version and multilingual support. Also added transformer optimization pass in the `cpu,int8` config. It was previously ommited due to a bug in ORT which was already resolved in version `1.15.0`. We currently don't have tests for the latest ORT and multilingual support since there are no publically available ort-nightly builds. Will look into this in the future. Fixes: #293 Fixes: #341 ## Checklist before requesting a review - [ ] Add unit tests for this change. - [ ] Make sure all tests can pass. - [ ] Update documents if necessary. - [ ] Format your code by running `pre-commit run --all-files` - [ ] Is this a user-facing change? If yes, give a description of this change to be included in the release notes. ## (Optional) Issue link
The documentation specifies that we still use a specific nightly build of ORT to optimize the Whisper model. However, the dummy input issue of microsoft/onnxruntime#15936 and microsoft/onnxruntime@e518933 is blocking execution of the optimized model through the OpenVINO Execution Provider for the same reason i.e, OpenVINO model compilation removes any dummy inputs in the graph but since it is present in the Olive optimized model (through the ORT 1.15 nightly), it fails to go through. Hence, OVEP requires the fixes for the dummy input issue in ORT master.
When I use ORT master to optimize Whisper through Olive, I get this error
out = model(inputs.encoder_input_ids, inputs.encoder_attention_mask, inputs.decoder_input_ids) AttributeError: 'WhisperEncoderDecoderInitInputs' object has no attribute 'encoder_attention_mask'
The text was updated successfully, but these errors were encountered: