Mirrored_Repos/InvokeAI
1
0
Fork 0
mirror of https://github.com/invoke-ai/InvokeAI synced 2024-08-30 20:32:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'development' into fix-high-step-count

Browse Source
This commit is contained in:
Lincoln Stein
2022-10-21 06:55:17 -04:00
committed by GitHub
parent 595d15455a d4d1014c9f
commit ddb007af65
205 changed files with 8144 additions and 2888 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
ldm/modules/diffusionmodules/model.py
View File

@ -49,9 +49,15 @@ class Upsample(nn.Module):
padding=1)
def forward(self, x):
cpu_m1_cond = True if hasattr(torch.backends, 'mps') and torch.backends.mps.is_available() and \
x.size()[0] * x.size()[1] * x.size()[2] * x.size()[3] % 2**27 == 0 else False
if cpu_m1_cond:
x = x.to('cpu') # send to cpu
x = torch.nn.functional.interpolate(x, scale_factor=2.0, mode="nearest")
if self.with_conv:
x = self.conv(x)
if cpu_m1_cond:
x = x.to('mps') # return to mps
return x
@ -117,6 +123,14 @@ class ResnetBlock(nn.Module):
padding=0)
def forward(self, x, temb):
if hasattr(torch.backends, 'mps') and torch.backends.mps.is_available():
x_size = x.size()
if (x_size[0] * x_size[1] * x_size[2] * x_size[3]) % 2**29 == 0:
self.to('cpu')
x = x.to('cpu')
else:
self.to('mps')
x = x.to('mps')
h = self.norm1(x)
h = silu(h)
h = self.conv1(h)
@ -245,7 +259,7 @@ class AttnBlock(nn.Module):
def make_attn(in_channels, attn_type="vanilla"):
assert attn_type in ["vanilla", "linear", "none"], f'attn_type {attn_type} unknown'
print(f"making attention of type '{attn_type}' with {in_channels} in_channels")
print(f" | Making attention of type '{attn_type}' with {in_channels} in_channels")
if attn_type == "vanilla":
return AttnBlock(in_channels)
elif attn_type == "none":
@ -521,7 +535,7 @@ class Decoder(nn.Module):
block_in = ch*ch_mult[self.num_resolutions-1]
curr_res = resolution // 2**(self.num_resolutions-1)
self.z_shape = (1,z_channels,curr_res,curr_res)
print("Working with z of shape {} = {} dimensions.".format(
print(" | Working with z of shape {} = {} dimensions.".format(
self.z_shape, np.prod(self.z_shape)))
# z to block_in

6
ldm/modules/diffusionmodules/util.py
View File

@ -66,7 +66,7 @@ def make_ddim_timesteps(
c = num_ddpm_timesteps // num_ddim_timesteps
if c < 1:
c = 1
ddim_timesteps = np.asarray(list(range(0, num_ddpm_timesteps, c)))
ddim_timesteps = (np.arange(0, num_ddim_timesteps) * c).astype(int)
elif ddim_discr_method == 'quad':
ddim_timesteps = (
(
@ -83,8 +83,8 @@ def make_ddim_timesteps(
# assert ddim_timesteps.shape[0] == num_ddim_timesteps
# add one to get the final alpha values right (the ones from first scale to data during sampling)
# steps_out = ddim_timesteps + 1
steps_out = ddim_timesteps
steps_out = ddim_timesteps + 1
# steps_out = ddim_timesteps
if verbose:
print(f'Selected timesteps for ddim sampler: {steps_out}')

2
ldm/modules/encoders/modules.py
View File

@ -5,7 +5,7 @@ import clip
from einops import rearrange, repeat
from transformers import CLIPTokenizer, CLIPTextModel
import kornia
from ldm.dream.devices import choose_torch_device
from ldm.invoke.devices import choose_torch_device
from ldm.modules.x_transformer import (
Encoder,