Public/automatic
1
0
Fork 0
mirror of https://github.com/vladmandic/automatic synced 2026-09-19 17:24:32 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'dev' into feature/chroma-support

Browse Source
This commit is contained in:
Enes Sadık Özbek
2025-06-18 10:08:38 +03:00
committed by GitHub
parent 4b3ce06916 e657cf790d
commit 7025ac8c36
2 changed files with 12 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/intel/ipex/hijacks.py
+7
View File
@@ -254,8 +254,15 @@ torch.Tensor.original_Tensor_to = torch.Tensor.to
@wraps(torch.Tensor.to)
def Tensor_to(self, device=None, *args, **kwargs):
if check_cuda(device):
if not device_supports_fp64 and kwargs.get("dtype", None) == torch.float64:
kwargs["dtype"] = torch.float32
return self.original_Tensor_to(return_xpu(device), *args, **kwargs)
else:
if not device_supports_fp64:
if kwargs.get("dtype", None) == torch.float64 and torch.device(device).type == "xpu":
kwargs["dtype"] = torch.float32
elif device == torch.float64 and self.device.type == "xpu":
device = torch.float32
return self.original_Tensor_to(device, *args, **kwargs)
original_Tensor_cuda = torch.Tensor.cuda
modules/sdnq/__init__.py
+5 -2
View File
@@ -56,8 +56,11 @@ def sdnq_quantize_layer(layer, weights_dtype="int8", torch_dtype=None, group_siz
output_channel_size, channel_size = layer.weight.shape
if use_quantized_matmul:
use_quantized_matmul = weights_dtype in quantized_matmul_dtypes and channel_size >= 32 and output_channel_size >= 32
if use_quantized_matmul and not dtype_dict[weights_dtype]["is_integer"]:
use_quantized_matmul = output_channel_size % 16 == 0 and channel_size % 16 == 0
if use_quantized_matmul:
if dtype_dict[weights_dtype]["is_integer"]:
use_quantized_matmul = output_channel_size % 8 == 0 and channel_size % 8 == 0
else:
use_quantized_matmul = output_channel_size % 16 == 0 and channel_size % 16 == 0
if group_size == 0:
if is_linear_type: