Skip to content

rebasing main commit#891

Closed
quic-akuruvil wants to merge 2 commits intoft_experimental_v1from
main
Closed

rebasing main commit#891
quic-akuruvil wants to merge 2 commits intoft_experimental_v1from
main

Conversation

@quic-akuruvil
Copy link
Contributor

rebasing main commit

vbaddi and others added 2 commits March 26, 2026 13:23
This PR restores FP16ClipTransform for embedding models
(`QEFFAutoModel`) in the default (non-proxy) path, while preserving
existing proxy-gated behavior for other model categories.

### What changed
- Added per-model support for always-on ONNX transforms in proxy
configuration.
- Set embedding models to always keep FP16ClipTransform enabled by
default.
- Embedding accuracy on HW depends on FP16 clipping, so clip must remain
enabled for embedding even when proxy is disabled.

### Tests verified

- `python -m pytest -q tests/unit_test/models/test_model_quickcheck.py
-k "test_text_embedding_fp16_clip_transform_and_export"`

cc: @anujgupt-github @quic-rishinr @quic-hemagnih

---------

Signed-off-by: vbaddi <vbaddi@qti.qualcomm.com>
CI test update to reduce the time taken for each models

---------

Signed-off-by: Abukhoyer Shaik <abukhoye@qti.qualcomm.com>
Signed-off-by: Abukhoyer SHaik <abukhoye@qti.qualcomm.com>
Signed-off-by: Rishin Raj <rishinr@qti.qualcomm.com>
Signed-off-by: Ann Kuruvilla <akuruvil@qti.qualcomm.com>
Co-authored-by: Rishin Raj <rishinr@qti.qualcomm.com>
Co-authored-by: Ann Kuruvilla <akuruvil@qti.qualcomm.com>
@quic-akuruvil
Copy link
Contributor Author

Not required

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants