Skip to content

Downgrade min deployment target for INT8 torchao per channel quantization to IO16 #14127

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
Sep 10, 2025
Merged

Downgrade min deployment target for INT8 torchao per channel quantization to IO16 #14127

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
7 changes: 5 additions & 2 deletions backends/apple/coreml/recipes/coreml_recipe_provider.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -69,6 +69,7 @@ def create_recipe(
recipe_type, activation_dtype=torch.float32, **kwargs
)
elif recipe_type == CoreMLRecipeType.TORCHAO_INT4_WEIGHT_ONLY_PER_CHANNEL:
self._validate_and_set_deployment_target(kwargs, ct.target.iOS18, "torchao")
return self._build_torchao_quantized_recipe(
recipe_type,
weight_dtype=torch.int4,
Expand All @@ -77,6 +78,7 @@ def create_recipe(
)
elif recipe_type == CoreMLRecipeType.TORCHAO_INT4_WEIGHT_ONLY_PER_GROUP:
group_size = kwargs.pop("group_size", 32)
self._validate_and_set_deployment_target(kwargs, ct.target.iOS18, "torchao")
return self._build_torchao_quantized_recipe(
recipe_type,
weight_dtype=torch.int4,
Expand All @@ -85,11 +87,14 @@ def create_recipe(
**kwargs,
)
elif recipe_type == CoreMLRecipeType.TORCHAO_INT8_WEIGHT_ONLY_PER_CHANNEL:
self._validate_and_set_deployment_target(kwargs, ct.target.iOS16, "torchao")
return self._build_torchao_quantized_recipe(
recipe_type, weight_dtype=torch.int8, is_per_channel=True, **kwargs
)
elif recipe_type == CoreMLRecipeType.TORCHAO_INT8_WEIGHT_ONLY_PER_GROUP:
group_size = kwargs.pop("group_size", 32)
# override minimum_deployment_target to ios18 for torchao (GH issue #13122)
self._validate_and_set_deployment_target(kwargs, ct.target.iOS18, "torchao")
return self._build_torchao_quantized_recipe(
recipe_type,
weight_dtype=torch.int8,
Expand Down Expand Up @@ -312,8 +317,6 @@ def _build_torchao_quantized_recipe(
ao_quantization_configs=[config],
)

# override minimum_deployment_target to ios18 for torchao (GH issue #13122)
self._validate_and_set_deployment_target(kwargs, ct.target.iOS18, "torchao")
lowering_recipe = self._get_coreml_lowering_recipe(**kwargs)

return ExportRecipe(
Expand Down
2 changes: 1 addition & 1 deletion backends/apple/coreml/test/test_coreml_recipes.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -501,7 +501,7 @@ def test_minimum_deployment_target_validation(self):
(CoreMLRecipeType.TORCHAO_INT4_WEIGHT_ONLY_PER_GROUP, ct.target.iOS18, {}),
(
CoreMLRecipeType.TORCHAO_INT8_WEIGHT_ONLY_PER_CHANNEL,
ct.target.iOS18,
ct.target.iOS16,
{},
),
(CoreMLRecipeType.TORCHAO_INT8_WEIGHT_ONLY_PER_GROUP, ct.target.iOS18, {}),
Expand Down
Loading