AI/diffusers
1
0
Fork 0
mirror of https://github.com/huggingface/diffusers.git synced 2026-01-27 17:22:53 +03:00
Code Activity

make style for https://github.com/huggingface/diffusers/pull/10368 (#10370)

Browse Source 
* fix bug for torch.uint1-7 not support in torch<2.6

* up

---------

Co-authored-by: baymax591 <cbai@mail.nwpu.edu.cn>
This commit is contained in:
YiYi Xu
2024-12-23 19:52:21 -10:00
committed by Aryan
parent cd4d0d8ffb
commit 11b1151840
1 changed files with 23 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

39
src/diffusers/quantizers/torchao/torchao_quantizer.py
View File

@@ -23,7 +23,7 @@ from typing import TYPE_CHECKING, Any, Dict, List, Union
from packaging import version
from ...utils import get_module_from_name, is_torch_available, is_torchao_available, logging
from ...utils import get_module_from_name, is_torch_available, is_torch_version, is_torchao_available, logging
from ..base import DiffusersQuantizer
@@ -35,21 +35,28 @@ if is_torch_available():
import torch
import torch.nn as nn
SUPPORTED_TORCH_DTYPES_FOR_QUANTIZATION = (
# At the moment, only int8 is supported for integer quantization dtypes.
# In Torch 2.6, int1-int7 will be introduced, so this can be visited in the future
# to support more quantization methods, such as intx_weight_only.
torch.int8,
torch.float8_e4m3fn,
torch.float8_e5m2,
torch.uint1,
torch.uint2,
torch.uint3,
torch.uint4,
torch.uint5,
torch.uint6,
torch.uint7,
)
if is_torch_version(">=", "2.5"):
SUPPORTED_TORCH_DTYPES_FOR_QUANTIZATION = (
# At the moment, only int8 is supported for integer quantization dtypes.
# In Torch 2.6, int1-int7 will be introduced, so this can be visited in the future
# to support more quantization methods, such as intx_weight_only.
torch.int8,
torch.float8_e4m3fn,
torch.float8_e5m2,
torch.uint1,
torch.uint2,
torch.uint3,
torch.uint4,
torch.uint5,
torch.uint6,
torch.uint7,
)
else:
SUPPORTED_TORCH_DTYPES_FOR_QUANTIZATION = (
torch.int8,
torch.float8_e4m3fn,
torch.float8_e5m2,
)
if is_torchao_available():
from torchao.quantization import quantize_