Skip to content

Address comments on #158 #159

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

Open
wants to merge 3 commits into
base: main
Choose a base branch
from
Open

Address comments on #158 #159

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
2ca61ed
Address comments on #158
larryliu0820 Oct 3, 2025
cdb4613
Fix version
larryliu0820 Oct 3, 2025
f03e2ad
Only change ExecuTorch nightly
larryliu0820 Oct 3, 2025
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
4 changes: 2 additions & 2 deletions install_dev.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,7 @@

def install_torch_nightly_deps():
"""Install torch related dependencies from pinned nightly"""
EXECUTORCH_NIGHTLY_VERSION = "dev20250916"
EXECUTORCH_NIGHTLY_VERSION = "dev20251003"
TORCHAO_NIGHTLY_VERSION = "dev20250916"
# Torch nightly is aligned with pinned nightly in https://github.com/pytorch/executorch/blob/main/install_requirements.py#L74
TORCH_NIGHTLY_VERSION = "dev20250916"
Expand All @@ -15,7 +15,7 @@ def install_torch_nightly_deps():
"-m",
"pip",
"install",
f"executorch==1.0.0.{EXECUTORCH_NIGHTLY_VERSION}",
f"executorch==1.1.0.{EXECUTORCH_NIGHTLY_VERSION}",
f"torch==2.10.0.{TORCH_NIGHTLY_VERSION}",
f"torchvision==0.25.0.{TORCH_NIGHTLY_VERSION}",
f"torchaudio==2.8.0.{TORCH_NIGHTLY_VERSION}",
Expand Down
8 changes: 4 additions & 4 deletions optimum/exporters/executorch/recipes/cuda.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -114,10 +114,6 @@ def _lower_to_executorch(
)
return {pte_name: et_prog}

# Decomposes SDPA since we don't have a flash attention kernel for it yet.
with torch.nn.attention.sdpa_kernel([SDPBackend.MATH]), torch.no_grad():
exported_progs = model.export()

if (
model.config._attn_implementation == "custom_sdpa"
or model.config._attn_implementation == "custom_sdpa_ring_kv_cache"
Expand All @@ -126,4 +122,8 @@ def _lower_to_executorch(
"Custom SDPA implementation is not supported for CUDA yet. Please use 'flash_attention' instead."
)

# Decomposes SDPA since we don't have a flash attention kernel for it yet.
with torch.nn.attention.sdpa_kernel([SDPBackend.MATH]), torch.no_grad():
exported_progs = model.export()

return _lower_to_executorch(exported_progs, model.metadata)
1 change: 1 addition & 0 deletions tests/models/test_modeling_voxtral.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -351,3 +351,4 @@ def test_voxtral_export_to_executorch_cuda_recipe(self):
)
subprocess.run(cmd, shell=True, check=True)
self.assertTrue(os.path.exists(os.path.join(output_dir, "model.pte")))
self.assertTrue(os.path.exists(os.path.join(output_dir, "aoti_cuda_blob.ptd")))
Loading