Skip to content

update lama export DS specs to be more accurate. #14737

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.

Sign up for GitHub

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

Merged
merged 1 commit into from
Oct 2, 2025
Merged

update lama export DS specs to be more accurate. #14737

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 7 additions & 2 deletions extension/llm/export/builder.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -142,9 +142,14 @@ def __init__(
{1: torch.export.Dim("token_dim", max=self.max_seq_len - 1)},
)
else:
# Two input arguments: tokens and input_pos but input_pos is static shape
# Two input arguments: tokens and input_pos but input_pos is static shape.

# A runtime assertion is added by torch.ops.llama.update_cache requires that
# L['tokens'].size()[1] + input_pos[0].item() < self.max_seq_len
# This consttaint L['tokens'].size()[1] to be elf.max_seq_len-1
# run with TORCH_LOGS=+dynamic for details
self.dynamic_shapes = (
{1: torch.export.Dim("token_dim", max=self.max_seq_len)},
{1: torch.export.Dim("token_dim", max=self.max_seq_len - 1)},
{"input_pos": {0: 1}},
)

Expand Down
2 changes: 1 addition & 1 deletion extension/llm/export/test/test_builder.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -88,7 +88,7 @@ def test_get_dynamic_shape_with_dynamic_shape_enabled_with_kv_cache(self) -> Non
# Check first element (tokens dimension)
self.assertIsInstance(result[0], dict)
self.assertIn(1, result[0])
self.assertEqual(result[0][1].max, self.max_seq_len)
self.assertEqual(result[0][1].max, self.max_seq_len - 1)

# Check second element (input_pos dimension)
self.assertIsInstance(result[1], dict)
Expand Down
Loading