lllyasviel/stable-diffusion-webui-forge
1
0
Fork 0
mirror of https://github.com/lllyasviel/stable-diffusion-webui-forge.git synced 2026-04-30 03:01:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

speed up a bit

Browse Source
This commit is contained in:
layerdiffusion
2024-08-19 05:06:46 -07:00
parent d38e560e42
commit d03fc5c2b1
1 changed files with 8 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
backend/operations.py
View File

@@ -354,7 +354,7 @@ try:
if self.bias is not None and self.bias.dtype != x.dtype: if self.bias is not None and self.bias.dtype != x.dtype:
# Maybe this can also be set to all non-bnb ops since the cost is very low. # Maybe this can also be set to all non-bnb ops since the cost is very low.
# And it only invokes one time, and most linear does not have bias # And it only invokes one time, and most linear does not have bias
self.bias.data = self.bias.data.to(x.dtype) self.bias = utils.tensor2parameter(self.bias.to(x.dtype))
if not self.parameters_manual_cast: if not self.parameters_manual_cast:
return functional_linear_4bits(x, self.weight, self.bias) return functional_linear_4bits(x, self.weight, self.bias)
@@ -409,7 +409,13 @@ class ForgeOperationsGGUF(ForgeOperations):
return self return self
def forward(self, x): def forward(self, x):
weight, bias, signal = weights_manual_cast(self, x, weight_fn=dequantize_tensor, bias_fn=dequantize_tensor) if self.bias is not None and self.bias.dtype != x.dtype:
self.bias = utils.tensor2parameter(dequantize_tensor(self.bias).to(x.dtype))
if self.weight is not None and self.weight.dtype != x.dtype and getattr(self.weight, 'gguf_cls', None) is None:
self.weight = utils.tensor2parameter(self.weight.to(x.dtype))
weight, bias, signal = weights_manual_cast(self, x, weight_fn=dequantize_tensor, bias_fn=None, skip_bias_dtype=True)
with main_stream_worker(weight, bias, signal): with main_stream_worker(weight, bias, signal):
return torch.nn.functional.linear(x, weight, bias) return torch.nn.functional.linear(x, weight, bias)