JLNTV_IPD/CogVideo
4
0
Fork 0
mirror of https://github.com/THUDM/CogVideo.git synced 2026-06-01 00:38:16 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix patch_size_t padding calculation for frame count alignment

Browse Source
This commit is contained in:
Mr-Neutr0n 2026-02-07 05:15:41 +05:30
parent 7a1af71545
commit 6c74057905
1 changed files with 5 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
finetune/models/cogvideox_t2v/lora_trainer.py
View File

@ -109,10 +109,12 @@ class CogVideoXT2VLoraTrainer(Trainer):
patch_size_t = self.state.transformer_config.patch_size_t patch_size_t = self.state.transformer_config.patch_size_t
if patch_size_t is not None: if patch_size_t is not None:
ncopy = latent.shape[2] % patch_size_t remainder = latent.shape[2] % patch_size_t
ncopy = (patch_size_t - remainder) % patch_size_t
# Copy the first frame ncopy times to match patch_size_t # Copy the first frame ncopy times to match patch_size_t
first_frame = latent[:, :, :1, :, :] # Get first frame [B, C, 1, H, W] if ncopy > 0:
latent = torch.cat([first_frame.repeat(1, 1, ncopy, 1, 1), latent], dim=2) first_frame = latent[:, :, :1, :, :] # Get first frame [B, C, 1, H, W]
latent = torch.cat([first_frame.repeat(1, 1, ncopy, 1, 1), latent], dim=2)
assert latent.shape[2] % patch_size_t == 0 assert latent.shape[2] % patch_size_t == 0
batch_size, num_channels, num_frames, height, width = latent.shape batch_size, num_channels, num_frames, height, width = latent.shape