Skip to content

adding Int8 weight only and dynamic configs to safetensors #3474

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
liangel-02 merged 1 commit into from
Dec 10, 2025
Merged

adding Int8 weight only and dynamic configs to safetensors #3474

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
6 changes: 6 additions & 0 deletions test/prototype/safetensors/test_safetensors_support.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,7 +20,9 @@
from torchao.quantization.quant_api import (
Float8DynamicActivationFloat8WeightConfig,
Int4WeightOnlyConfig,
Int8DynamicActivationInt8WeightConfig,
Int8DynamicActivationIntxWeightConfig,
Int8WeightOnlyConfig,
IntxWeightOnlyConfig,
)
from torchao.utils import is_sm_at_least_89
Expand Down Expand Up @@ -50,6 +52,8 @@ class TestSafeTensors(TestCase):
(Int4WeightOnlyConfig(int4_packing_format="tile_packed_to_4d"), False),
(IntxWeightOnlyConfig(), False),
(Int8DynamicActivationIntxWeightConfig(), False),
(Int8WeightOnlyConfig(version=2), False),
(Int8DynamicActivationInt8WeightConfig(version=2), False),
],
)
def test_safetensors(self, config, act_pre_scale=False):
Expand Down Expand Up @@ -95,6 +99,8 @@ def test_safetensors(self, config, act_pre_scale=False):
(Int4WeightOnlyConfig(int4_packing_format="tile_packed_to_4d"), False),
(IntxWeightOnlyConfig(), False),
(Int8DynamicActivationIntxWeightConfig(), False),
(Int8WeightOnlyConfig(version=2), False),
(Int8DynamicActivationInt8WeightConfig(version=2), False),
],
)
def test_safetensors_sharded(self, config, act_pre_scale=False):
Expand Down
11 changes: 10 additions & 1 deletion torchao/prototype/safetensors/safetensors_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,28 +10,37 @@
Float8Tensor,
Int4Tensor,
Int4TilePackedTo4dTensor,
Int8Tensor,
IntxUnpackedToInt8Tensor,
MappingType,
)
from torchao.quantization.quantize_.common import KernelPreference
from torchao.quantization.quantize_.workflows import QuantizeTensorToFloat8Kwargs
from torchao.quantization.quantize_.workflows import (
QuantizeTensorToFloat8Kwargs,
QuantizeTensorToInt8Kwargs,
)

ALLOWED_CLASSES = {
"Float8Tensor": Float8Tensor,
"Int4Tensor": Int4Tensor,
"Int4TilePackedTo4dTensor": Int4TilePackedTo4dTensor,
"IntxUnpackedToInt8Tensor": IntxUnpackedToInt8Tensor,
"Int8Tensor": Int8Tensor,
"Float8MMConfig": torchao.float8.inference.Float8MMConfig,
"QuantizeTensorToFloat8Kwargs": QuantizeTensorToFloat8Kwargs,
"QuantizeTensorToInt8Kwargs": QuantizeTensorToInt8Kwargs,
"PerRow": torchao.quantization.PerRow,
"PerTensor": torchao.quantization.PerTensor,
"KernelPreference": KernelPreference,
"MappingType": MappingType,
}

ALLOWED_TENSORS_SUBCLASSES = [
"Float8Tensor",
"Int4Tensor",
"Int4TilePackedTo4dTensor",
"IntxUnpackedToInt8Tensor",
"Int8Tensor",
]

__all__ = [
Expand Down
Loading