Skip to content
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

#12376: Support for non-32 Height in Width Sharded Conv2d #12382

Merged
merged 4 commits into from
Sep 12, 2024

Conversation

sankarmanoj-tt
Copy link
Contributor

@sankarmanoj-tt sankarmanoj-tt commented Sep 9, 2024

Ticket

#12376

Problem description

Support for non-32 Height in Width Sharded Conv2d

Checklist

  • Post commit CI passes
  • New/Existing tests provide coverage for changes

(256, 2048, 8, 8, 3, 3, 1, 1, 8),
(512, 2048, 16, 16, 3, 3, 1, 1, 4),
(768, 768, 8, 8, 3, 3, 1, 1, 1),
(768, 768, 16, 16, 3, 3, 1, 1, 1),
(1280, 1280, 16, 16, 3, 3, 1, 1, 1),
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why removed these test cases?

@@ -204,13 +204,13 @@ OptimizedConvParallelizationConfig determine_conv_op_parallel_config_from_conv_o
TT_ASSERT(conv_output_mem_config.shard_spec.has_value());
const auto& shard_spec = conv_output_mem_config.shard_spec.value();
const auto& shard_shape = shard_spec.shape;
TT_ASSERT(shard_shape[0] % 32 == 0);
// TT_ASSERT(shard_shape[0] % 32 == 0);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

delete

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

does this affect height and block sharding variants?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fixed.

@@ -317,18 +317,21 @@ def run_conv_with_split(
assert_with_pcc(torch_output_tensor, torch_out_golden_tensor, pcc=pcc)


@skip_for_grayskull()
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do other tests cover these test scenarios on grayskull?
If not, why would these test scenarios note be relevant for grayskull?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We are not planning on supporting these new features on gray skull.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Are the cases in the already existing test (before this PR) supported by grayskull?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Most of them were skipped, due to an OOM error. The width sharded Conv2d which is tested by this code was implemented to support a wormhole only model.

@sankarmanoj-tt sankarmanoj-tt merged commit 9c52688 into main Sep 12, 2024
6 checks passed
@sankarmanoj-tt sankarmanoj-tt deleted the smanoj/conv_ws_non32h branch September 12, 2024 16:37
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