lllyasviel/stable-diffusion-webui-forge
1
0
Fork 0
mirror of https://github.com/lllyasviel/stable-diffusion-webui-forge.git synced 2026-03-03 20:19:49 +00:00
Code Issues Packages Projects Releases Wiki Activity

Revert partially "use safer codes"

Browse Source
This commit is contained in:
layerdiffusion
2024-08-31 11:07:28 -07:00
parent 70a555906a
commit 3a9cf1f8e5
1 changed files with 5 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
packages_3rdparty/gguf/quants.py vendored
View File

@@ -151,7 +151,7 @@ class __Quant(ABC):
rows = data.reshape((-1, data.shape[-1])).view(torch.uint8)
n_blocks = rows.numel() // cls.type_size
blocks = rows.reshape((n_blocks, cls.type_size))
parameter.data = blocks.clone(memory_format=torch.contiguous_format)
parameter.data = blocks.contiguous()
cls.bake_inner(parameter)
parameter.baked = True
return
@@ -312,7 +312,7 @@ class Q4_0(__Quant, qtype=GGMLQuantizationType.Q4_0):
d, x = quick_split(blocks, [2])
d = d.view(torch.float16).to(parameter.computation_dtype).view(torch.uint8)
x = change_4bits_order(x).view(torch.uint8)
parameter.data = torch.cat([d, x], dim=-1).clone(memory_format=torch.contiguous_format)
parameter.data = torch.cat([d, x], dim=-1).contiguous()
return
@classmethod
@@ -389,7 +389,7 @@ class Q4_1(__Quant, qtype=GGMLQuantizationType.Q4_1):
m = m.view(torch.float16).to(parameter.computation_dtype).view(torch.uint8)
qs = change_4bits_order(qs).view(torch.uint8)
parameter.data = torch.cat([d, m, qs], dim=-1).clone(memory_format=torch.contiguous_format)
parameter.data = torch.cat([d, m, qs], dim=-1).contiguous()
return
@@ -601,7 +601,7 @@ class Q8_0(__Quant, qtype=GGMLQuantizationType.Q8_0):
d, x = quick_split(blocks, [2])
x = x.view(torch.int8)
d = d.view(torch.float16).to(parameter.computation_dtype).view(torch.int8)
parameter.data = torch.cat([d, x], dim=-1).clone(memory_format=torch.contiguous_format)
parameter.data = torch.cat([d, x], dim=-1).contiguous()
return
@classmethod
@@ -808,7 +808,7 @@ class Q4_K(__Quant, qtype=GGMLQuantizationType.Q4_K):
dm = dm.view(torch.uint8).reshape((n_blocks, -1))
qs = qs.view(torch.uint8)
parameter.data = torch.cat([d, dm, qs], dim=-1).clone(memory_format=torch.contiguous_format)
parameter.data = torch.cat([d, dm, qs], dim=-1).contiguous()
return
@classmethod