Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Hey, thanks for this project! I started exploring/experimenting it with some ASR models (Parakeet TDT, Sortformer), and the results look great so far :-) . While testing, I hit a crash: when an input is invalid, the whole app exits instead of getting an error back.
based on my experiments, the cause is mlx-c's default error handler, which prints the message and calls
exit(-1). The EP never replaces it, so its own error handling never gets a chance to run. I just add a handler that just logs the message... The failing mlx-c call then returns a non-zero status, and ORT gets a normalORT_EP_FAIL.if you want to reproduce: run the Parakeet TDT encoder with an input that's too long (
T=45000). The CPU EP returns a broadcast error. The MLX EP currently exits with status 255. With this change it returnsExecutionProviderFailure, and the same session keeps working and still matches CPU afterwards.Btw my simple tests:
For 40 sec audio:
CPU: 2.82 s
TDT, encoder on MLX, decoder on CPU = 0.64 s
TDT, all on MLX = 0.89 s
4.5 sec sortformer audio=
cpu= 15 secs
mx= 5.0 secs!
strange:
Nemotron streaming (for 40 secs audio)
cpu = 10.5 secs
mlx= 10.3 secs