Optimized ONNX Transform with Merging and Thread Pooling #545
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
Signed-off-by: abhishek-singh591 <[email protected]>
Signed-off-by: abhishek-singh591 <[email protected]>
Signed-off-by: abhishek-singh591 <[email protected]>
Signed-off-by: abhishek-singh591 <[email protected]>
Signed-off-by: abhishek-singh591 <[email protected]>
…netune.md readme file. (#520) - Introduced the handling of custom dataset via --dataset_config argument. This argument expects a json file which has parameters to enable custom preprocessing for any dataset. - Updated the docs to reflect the changes in the interface of custom dataset usage. --------- Signed-off-by: meetkuma <[email protected]> Signed-off-by: abhishek-singh591 <[email protected]>
This PR addresses the missing copyright headers in several source files that were previously overlooked. ### Changes include: - Added standard copyright notice to all applicable files - Ensured consistency in formatting and placement of headers - Verified that no functional code changes were introduced ### Files Updated: - `QEfficient/transformers/models/gemma3/modeling_gemma3.py` - `docs/conf.py` Signed-off-by: Abukhoyer Shaik <[email protected]> Signed-off-by: abhishek-singh591 <[email protected]>
…and displaying non scaled loss value on console. (#527) Signed-off-by: Swati Allabadi <[email protected]> Co-authored-by: Swati Allabadi <[email protected]> Signed-off-by: abhishek-singh591 <[email protected]>
Signed-off-by: Mohit Soni <[email protected]> Signed-off-by: abhishek-singh591 <[email protected]>
### Purpose of this PR: This update aims to reduce test execution time for causal language model inference. Previously, tests were run using full-scale models with one or two layers, which was inefficient and time-consuming. Refactoring CLI api testing for independent testing and redundant conftest code. ### What’s Changed: Introduced dummy models with significantly smaller configurations by adjusting parameters such as `max_position_embeddings, num_hidden_layers, num_attention_heads, hidden_size, intermediate_size, vocab_size and additional_params`. These lightweight models are used exclusively for testing purposes to ensure faster execution without compromising test coverage. And CLI testing has two test scripts one is for export, compile, and execute, another is for infer cli api. **Note:** This optimization is applied only to causal language models. --------- Signed-off-by: Abukhoyer Shaik <[email protected]> Signed-off-by: abhishek-singh591 <[email protected]>
Fix for qeff vision models export is failing for dual qpc method when we providing onnx_dir option to export API. --------- Signed-off-by: Dipankar Sarkar <[email protected]> Signed-off-by: Dipankar Sarkar <[email protected]> Signed-off-by: abhishek-singh591 <[email protected]>
Signed-off-by: Abukhoyer Shaik <[email protected]> Signed-off-by: abhishek-singh591 <[email protected]>
…el loss/metrics (#531) Enable test cases for Intermediate step level loss/metric matching in single and DDP set up. Nested dictionary structure for mapping the reference losses at different test scenarios. The test scenarios with the ref values are listed in a separate reference file. The test scenarios at present include single device testing for below models: Llama, Bert on Alpaca and GSM8k dataset. **REFERNCE DATA based on SDK - 1.21.0.23** --------- Signed-off-by: Ann Kuruvilla <[email protected]> Signed-off-by: Ann Kuruvilla <[email protected]> Signed-off-by: abhishek-singh591 <[email protected]>
abhishek-singh591
requested review from
quic-rishinr,
ochougul,
quic-hemagnih and
quic-amitraj
as code owners
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.
Optimized ONNX Transform with Merging and Thread Pooling
This PR follows up on 539 – Optimized ONNX transform class via multithreading.
It merges the FP16 and Split ONNX transform classes into a single implementation to eliminate redundant tensor loading and iteration. Additionally, the transform logic has been refactored to use a thread pool, replacing the previous sequential loop to parallelize tensor operations.
Performance Benchmarks