wangbo/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2026-01-15 08:40:50 +08:00
Code Issues Actions 10 Packages Projects Releases Wiki Activity

Don't scale RoPE for lite model as that just doesn't work...

Browse Source
This commit is contained in:
kijai 2025-11-27 02:40:59 +02:00
parent a72c7edd3c
commit 68c712ef7e
1 changed files with 10 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
comfy/ldm/kandinsky5/model.py
View File

@ -307,24 +307,23 @@ class Kandinsky5(nn.Module):
h_start += rope_options.get("shift_y", 0.0)
w_start += rope_options.get("shift_x", 0.0)
else:
# this is experimental as the original code only had two fixed scales for 512p and 1024p
if t == 1: # image model
self.rope_scale_factor = (1.0, 1.0, 1.0)
else:
if self.model_dim == 4096: # pro video model,this is experimental as the original code only had two fixed scales for 512p and 1024p
spatial_size = h * w
scale_16384 = (1.0, 3.16, 3.16)
scale_9216 = (1.0, 2.0, 2.0)
if spatial_size <= 6144:
self.rope_scale_factor = scale_9216
rope_scale_factor = scale_9216
elif spatial_size >= 14080:
self.rope_scale_factor = scale_16384
rope_scale_factor = scale_16384
else:
t = (spatial_size - 14080) / (6144 - 14080)
self.rope_scale_factor = tuple(a + (b - a) * t for a, b in zip(scale_16384, scale_9216))
rope_scale_factor = tuple(a + (b - a) * t for a, b in zip(scale_16384, scale_9216))
else:
rope_scale_factor = self.rope_scale_factor
t_len = (t_len - 1.0) // self.rope_scale_factor[0] + 1.0
h_len = (h_len - 1.0) // self.rope_scale_factor[1] + 1.0
w_len = (w_len - 1.0) // self.rope_scale_factor[2] + 1.0
t_len = (t_len - 1.0) // rope_scale_factor[0] + 1.0
h_len = (h_len - 1.0) // rope_scale_factor[1] + 1.0
w_len = (w_len - 1.0) // rope_scale_factor[2] + 1.0
img_ids = torch.zeros((steps_t, steps_h, steps_w, 3), device=device, dtype=dtype)
img_ids[:, :, :, 0] = img_ids[:, :, :, 0] + torch.linspace(t_start, t_start + (t_len - 1), steps=steps_t, device=device, dtype=dtype).reshape(-1, 1, 1)
@ -345,8 +344,8 @@ class Kandinsky5(nn.Module):
visual_embed = self.visual_embeddings(x)
visual_shape = visual_embed.shape[:-1]
visual_embed = visual_embed.flatten(1, -2)
blocks_replace = patches_replace.get("dit", {})
transformer_options["total_blocks"] = len(self.visual_transformer_blocks)
transformer_options["block_type"] = "double"