Skip to content

[pull] master from comfyanonymous:master #38

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
pull merged 1 commit into from
May 4, 2025
Merged

[pull] master from comfyanonymous:master #38

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 9 additions & 9 deletions comfy/ldm/lumina/model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,7 +8,7 @@
import torch.nn.functional as F
import comfy.ldm.common_dit

from comfy.ldm.modules.diffusionmodules.mmdit import TimestepEmbedder, RMSNorm
from comfy.ldm.modules.diffusionmodules.mmdit import TimestepEmbedder
from comfy.ldm.modules.attention import optimized_attention_masked
from comfy.ldm.flux.layers import EmbedND

Expand Down Expand Up @@ -64,8 +64,8 @@ def __init__(
)

if qk_norm:
self.q_norm = RMSNorm(self.head_dim, elementwise_affine=True, **operation_settings)
self.k_norm = RMSNorm(self.head_dim, elementwise_affine=True, **operation_settings)
self.q_norm = operation_settings.get("operations").RMSNorm(self.head_dim, elementwise_affine=True, device=operation_settings.get("device"), dtype=operation_settings.get("dtype"))
self.k_norm = operation_settings.get("operations").RMSNorm(self.head_dim, elementwise_affine=True, device=operation_settings.get("device"), dtype=operation_settings.get("dtype"))
else:
self.q_norm = self.k_norm = nn.Identity()

Expand Down Expand Up @@ -242,11 +242,11 @@ def __init__(
operation_settings=operation_settings,
)
self.layer_id = layer_id
self.attention_norm1 = RMSNorm(dim, eps=norm_eps, elementwise_affine=True, **operation_settings)
self.ffn_norm1 = RMSNorm(dim, eps=norm_eps, elementwise_affine=True, **operation_settings)
self.attention_norm1 = operation_settings.get("operations").RMSNorm(dim, eps=norm_eps, elementwise_affine=True, device=operation_settings.get("device"), dtype=operation_settings.get("dtype"))
self.ffn_norm1 = operation_settings.get("operations").RMSNorm(dim, eps=norm_eps, elementwise_affine=True, device=operation_settings.get("device"), dtype=operation_settings.get("dtype"))

self.attention_norm2 = RMSNorm(dim, eps=norm_eps, elementwise_affine=True, **operation_settings)
self.ffn_norm2 = RMSNorm(dim, eps=norm_eps, elementwise_affine=True, **operation_settings)
self.attention_norm2 = operation_settings.get("operations").RMSNorm(dim, eps=norm_eps, elementwise_affine=True, device=operation_settings.get("device"), dtype=operation_settings.get("dtype"))
self.ffn_norm2 = operation_settings.get("operations").RMSNorm(dim, eps=norm_eps, elementwise_affine=True, device=operation_settings.get("device"), dtype=operation_settings.get("dtype"))

self.modulation = modulation
if modulation:
Expand Down Expand Up @@ -431,7 +431,7 @@ def __init__(

self.t_embedder = TimestepEmbedder(min(dim, 1024), **operation_settings)
self.cap_embedder = nn.Sequential(
RMSNorm(cap_feat_dim, eps=norm_eps, elementwise_affine=True, **operation_settings),
operation_settings.get("operations").RMSNorm(cap_feat_dim, eps=norm_eps, elementwise_affine=True, device=operation_settings.get("device"), dtype=operation_settings.get("dtype")),
operation_settings.get("operations").Linear(
cap_feat_dim,
dim,
Expand All @@ -457,7 +457,7 @@ def __init__(
for layer_id in range(n_layers)
]
)
self.norm_final = RMSNorm(dim, eps=norm_eps, elementwise_affine=True, **operation_settings)
self.norm_final = operation_settings.get("operations").RMSNorm(dim, eps=norm_eps, elementwise_affine=True, device=operation_settings.get("device"), dtype=operation_settings.get("dtype"))
self.final_layer = FinalLayer(dim, patch_size, self.out_channels, operation_settings=operation_settings)

assert (dim // n_heads) == sum(axes_dims)
Expand Down
Loading