wangbo/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2026-05-24 07:57:29 +08:00
Code Issues Actions 16 Packages Projects Releases Wiki Activity

Remove unused code.

Browse Source
This commit is contained in:
Talmaj Marinc 2026-05-19 13:45:46 +02:00
parent a859dc81a8
commit 395fbaa8a8
1 changed files with 7 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
comfy/image_encoders/dino2.py
View File

@ -184,8 +184,7 @@ class RotaryPositionEmbedding2D(torch.nn.Module):
class Dino2Encoder(torch.nn.Module): class Dino2Encoder(torch.nn.Module):
def __init__(self, dim, num_heads, layer_norm_eps, num_layers, dtype, device, operations, use_swiglu_ffn, def __init__(self, dim, num_heads, layer_norm_eps, num_layers, dtype, device, operations, use_swiglu_ffn,
qknorm_start: int = -1, rope: "RotaryPositionEmbedding2D | None" = None, qknorm_start: int = -1):
rope_start: int = -1):
super().__init__() super().__init__()
self.layer = torch.nn.ModuleList([ self.layer = torch.nn.ModuleList([
Dino2Block( Dino2Block(
@ -195,8 +194,6 @@ class Dino2Encoder(torch.nn.Module):
) )
for i in range(num_layers) for i in range(num_layers)
]) ])
self.rope = rope
self.rope_start = rope_start
def forward(self, x, intermediate_output=None): def forward(self, x, intermediate_output=None):
# Backward-compat path used by ``ClipVisionModel`` (no DA3 extensions). # Backward-compat path used by ``ClipVisionModel`` (no DA3 extensions).
@ -282,7 +279,6 @@ class Dino2Embeddings(torch.nn.Module):
return torch.cat((class_pos, patch_pos), dim=1).to(x.dtype) return torch.cat((class_pos, patch_pos), dim=1).to(x.dtype)
def forward(self, pixel_values): def forward(self, pixel_values):
_, _, H, W = pixel_values.shape
x = self.patch_embeddings(pixel_values) x = self.patch_embeddings(pixel_values)
x = torch.cat((self.cls_token.to(device=x.device, dtype=x.dtype).expand(x.shape[0], -1, -1), x), dim=1) x = torch.cat((self.cls_token.to(device=x.device, dtype=x.dtype).expand(x.shape[0], -1, -1), x), dim=1)
if x.shape[1] - 1 == self.position_embeddings.shape[1] - 1: if x.shape[1] - 1 == self.position_embeddings.shape[1] - 1:
@ -351,11 +347,16 @@ class Dinov2Model(torch.nn.Module):
dim, heads, layer_norm_eps, num_layers, dtype, device, operations, dim, heads, layer_norm_eps, num_layers, dtype, device, operations,
use_swiglu_ffn=use_swiglu_ffn, use_swiglu_ffn=use_swiglu_ffn,
qknorm_start=self.qknorm_start, qknorm_start=self.qknorm_start,
rope=self.rope, rope_start=self.rope_start,
) )
self.layernorm = operations.LayerNorm(dim, eps=layer_norm_eps, dtype=dtype, device=device) self.layernorm = operations.LayerNorm(dim, eps=layer_norm_eps, dtype=dtype, device=device)
def forward(self, pixel_values, attention_mask=None, intermediate_output=None): def forward(self, pixel_values, attention_mask=None, intermediate_output=None):
if self.alt_start != -1:
raise RuntimeError(
"Dinov2Model.forward() is the backward-compatible CLIP-vision path and does not "
"apply DA3 extensions (RoPE, alternating attention, camera-token injection). "
"Use get_intermediate_layers_da3() for Depth Anything 3 models."
)
x = self.embeddings(pixel_values) x = self.embeddings(pixel_values)
x, i = self.encoder(x, intermediate_output=intermediate_output) x, i = self.encoder(x, intermediate_output=intermediate_output)
x = self.layernorm(x) x = self.layernorm(x)