andreas/ComfyUI
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

LTXV lowvram fixes.

Browse Source
This commit is contained in:
comfyanonymous
2024-11-22 17:17:11 -05:00
parent bc6be6c11e
commit e5c3f4b87f
4 changed files with 13 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
comfy/ldm/lightricks/vae/causal_conv3d.py
View File

@@ -2,6 +2,8 @@ from typing import Tuple, Union
import torch
import torch.nn as nn
import comfy.ops
ops = comfy.ops.disable_weight_init
class CausalConv3d(nn.Module):
@@ -29,7 +31,7 @@ class CausalConv3d(nn.Module):
width_pad = kernel_size[2] // 2
padding = (0, height_pad, width_pad)
self.conv = nn.Conv3d(
self.conv = ops.Conv3d(
in_channels,
out_channels,
kernel_size,

4
comfy/ldm/lightricks/vae/causal_video_autoencoder.py
View File

@@ -628,10 +628,10 @@ class processor(nn.Module):
self.register_buffer("channel", torch.empty(128))
def un_normalize(self, x):
return (x * self.get_buffer("std-of-means").view(1, -1, 1, 1, 1)) + self.get_buffer("mean-of-means").view(1, -1, 1, 1, 1)
return (x * self.get_buffer("std-of-means").view(1, -1, 1, 1, 1).to(x)) + self.get_buffer("mean-of-means").view(1, -1, 1, 1, 1).to(x)
def normalize(self, x):
return (x - self.get_buffer("mean-of-means").view(1, -1, 1, 1, 1)) / self.get_buffer("std-of-means").view(1, -1, 1, 1, 1)
return (x - self.get_buffer("mean-of-means").view(1, -1, 1, 1, 1).to(x)) / self.get_buffer("std-of-means").view(1, -1, 1, 1, 1).to(x)
class VideoVAE(nn.Module):
def __init__(self):

11
comfy/ldm/lightricks/vae/conv_nd_factory.py
View File

@@ -4,7 +4,8 @@ import torch
from .dual_conv3d import DualConv3d
from .causal_conv3d import CausalConv3d
import comfy.ops
ops = comfy.ops.disable_weight_init
def make_conv_nd(
dims: Union[int, Tuple[int, int]],
@@ -19,7 +20,7 @@ def make_conv_nd(
causal=False,
):
if dims == 2:
return torch.nn.Conv2d(
return ops.Conv2d(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=kernel_size,
@@ -41,7 +42,7 @@ def make_conv_nd(
groups=groups,
bias=bias,
)
return torch.nn.Conv3d(
return ops.Conv3d(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=kernel_size,
@@ -71,11 +72,11 @@ def make_linear_nd(
bias=True,
):
if dims == 2:
return torch.nn.Conv2d(
return ops.Conv2d(
in_channels=in_channels, out_channels=out_channels, kernel_size=1, bias=bias
)
elif dims == 3 or dims == (2, 1):
return torch.nn.Conv3d(
return ops.Conv3d(
in_channels=in_channels, out_channels=out_channels, kernel_size=1, bias=bias
)
else: