-
Notifications
You must be signed in to change notification settings - Fork 641
quantized w2l #1623
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
quantized w2l #1623
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/1623
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (2 Unrelated Failures)As of commit 5c3cc32 with merge base 6a1c7a2 ( FLAKY - The following job failed but was likely due to flakiness present on trunk:
BROKEN TRUNK - The following job failed but was present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
This pull request was exported from Phabricator. Differential Revision: D52809537 |
ef58ff2
to
b912426
Compare
Summary: With the features add prior to this diff, Quantized W2L should successfully work now Differential Revision: D52809537
This pull request was exported from Phabricator. Differential Revision: D52809537 |
b912426
to
6551916
Compare
Summary: With the features add prior to this diff, Quantized W2L should successfully work now Differential Revision: D52809537
This pull request was exported from Phabricator. Differential Revision: D52809537 |
Summary: With the features add prior to this diff, Quantized W2L should successfully work now Differential Revision: D52809537
6551916
to
d7244fb
Compare
This pull request was exported from Phabricator. Differential Revision: D52809537 |
d7244fb
to
32f29c8
Compare
Summary: With the features add prior to this diff, Quantized W2L should successfully work now Differential Revision: D52809537
This pull request was exported from Phabricator. Differential Revision: D52809537 |
Summary: With the features add prior to this diff, Quantized W2L should successfully work now Differential Revision: D52809537
32f29c8
to
73ed5fb
Compare
This pull request was exported from Phabricator. Differential Revision: D52809537 |
73ed5fb
to
15fb70e
Compare
Summary: With the features add prior to this diff, Quantized W2L should successfully work now Reviewed By: digantdesai Differential Revision: D52809537
This pull request was exported from Phabricator. Differential Revision: D52809537 |
Summary: With the features add prior to this diff, Quantized W2L should successfully work now Reviewed By: digantdesai Differential Revision: D52809537
15fb70e
to
4023a49
Compare
This pull request was exported from Phabricator. Differential Revision: D52809537 |
4023a49
to
98ab145
Compare
Summary: With the features add prior to this diff, Quantized W2L should successfully work now Reviewed By: digantdesai Differential Revision: D52809537
This pull request was exported from Phabricator. Differential Revision: D52809537 |
Summary: With the features add prior to this diff, Quantized W2L should successfully work now Reviewed By: digantdesai Differential Revision: D52809537
98ab145
to
b8458e5
Compare
This pull request was exported from Phabricator. Differential Revision: D52809537 |
Summary: With the features add prior to this diff, Quantized W2L should successfully work now Reviewed By: digantdesai Differential Revision: D52809537
b8458e5
to
c6b6c9f
Compare
This pull request was exported from Phabricator. Differential Revision: D52809537 |
Summary: With the features add prior to this diff, Quantized W2L should successfully work now Reviewed By: digantdesai Differential Revision: D52809537
c6b6c9f
to
f4d39ed
Compare
This pull request was exported from Phabricator. Differential Revision: D52809537 |
Summary: With the features add prior to this diff, Quantized W2L should successfully work now Reviewed By: digantdesai Differential Revision: D52809537
f4d39ed
to
ca4ab9e
Compare
This pull request was exported from Phabricator. Differential Revision: D52809537 |
ca4ab9e
to
a804548
Compare
Summary: With the features add prior to this diff, Quantized W2L should successfully work now Reviewed By: digantdesai Differential Revision: D52809537
This pull request was exported from Phabricator. Differential Revision: D52809537 |
a804548
to
c40e403
Compare
Summary: With the features add prior to this diff, Quantized W2L should successfully work now Reviewed By: digantdesai Differential Revision: D52809537
Summary: Creating new buck targets for serialization and schema. Want to split out Targets which explicitly use the schema (things like passes and node visitors use the dataclasses instantiated in the schema) and targets which are actually serializing, like xnnpack_preprocess Reviewed By: digantdesai Differential Revision: D52809539
Summary: Some funky behavior can happen when we see fused activations under quantization. Specifically we see: ``` dequant --> op --> act --> quant ``` In serialization logic, this sometimes can create some issues. As a result, it if the activation can be fused with the previous op, then we delete the activation, and embed the activation's min/max to the op's node metadata. In serialization, we will check for this metadata and properly apply the activation if found. Reviewed By: digantdesai Differential Revision: D52809538
Summary: Continued from the previous diff, we now apply the pass to the xnnpack delegate, and modify the serialization logic to use the pass's helper to get activation constraints if the node has been fused with an activation Reviewed By: digantdesai Differential Revision: D52852651
Summary: Adding a test to guard FP32 W2L delegation and inference Reviewed By: digantdesai Differential Revision: D52219632
Summary: Quantization was not supported for our conv1d pass in the past, we now add support so that we can delegate and run quantized conv1d graphs. The logic added to the pass was primarily adding q/dq nodes around the squeeze/unsqueeze nodes so that when xnnpack reshapes intermediate tensors to perform 2d conv, the squeeze and unsqueeze would both be quantized. Reviewed By: digantdesai Differential Revision: D52809536
Summary: With the features add prior to this diff, Quantized W2L should successfully work now Reviewed By: digantdesai Differential Revision: D52809537
This pull request was exported from Phabricator. Differential Revision: D52809537 |
c40e403
to
5c3cc32
Compare
This pull request was exported from Phabricator. Differential Revision: D52809537 |
This pull request has been merged in c0c2fab. |
Summary: With the features add prior to this diff, Quantized W2L should successfully work now
Differential Revision: D52809537