Qualcomm AI Engine Direct - GA Model Enablement (Roberta) #11354
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/11354
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit fe5089a with merge base e02ca41 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
|
@pytorchbot label "release notes: qualcomm" |
pytorch-bot
bot
added
the
release notes: qualcomm
chenweng-quic
changed the title
|
Hi @cccclai, we have 4 GA BERT variants depend on this PR, will submit them after this got merged. It's just FYI, please take your time, no rush. |
|
And looks like it's not triggering CI, could you help on it? |
|
@cccclai has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
There is a lint error, can you fix it? |
- Add end to end script for Roberta - Handle extreme values in replace_inf function for masked_fill
chenweng-quic
force-pushed
the
dev1/chenweng/roberta
branch
from
32dd145 to
fe5089a
Compare
|
@cccclai has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
@cccclai has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
) ### Summary - support e2e script / test case for GA RoBERTa model - perf: 16a8w 79.2ms/inf (SM8750) (seq_len=107) - acc: ~= 50% ### Test plan ``` python backends/qualcomm/tests/test_qnn_delegate.py -k TestExampleOssScript.test_roberta --device <device_id> --host <host> --model <soc_model> --build_folder build-android --executorch_root . --sentence_dataset wikisent2.txt --artifact roberta_artifact ```
Summary
Test plan