Skip to content

Commit

Permalink
remove print
Browse files Browse the repository at this point in the history
  • Loading branch information
kijai committed Dec 19, 2024
1 parent efec814 commit da2fb54
Showing 1 changed file with 9 additions and 9 deletions.
18 changes: 9 additions & 9 deletions hyvideo/diffusion/pipelines/pipeline_hunyuan_video.py
Original file line number Diff line number Diff line change
Expand Up @@ -690,17 +690,17 @@ def __call__(
))
for c in context_queue:
partial_latent_model_input = latent_model_input[:, :, c, :, :]
print("partial_latent_model_input", partial_latent_model_input.shape)
#print("partial_latent_model_input", partial_latent_model_input.shape)
with torch.autocast(
device_type="cuda", dtype=self.base_dtype, enabled=True):
noise_pred[:, :, c, :, :] += self.transformer( # For an input image (129, 192, 336) (1, 256, 256)
partial_latent_model_input, # [2, 16, 33, 24, 42]
t_expand, # [2]
text_states=input_prompt_embeds, # [2, 256, 4096]
text_mask=input_prompt_mask, # [2, 256]
text_states_2=input_prompt_embeds_2, # [2, 768]
freqs_cos=freqs_cos, # [seqlen, head_dim]
freqs_sin=freqs_sin, # [seqlen, head_dim]
noise_pred[:, :, c, :, :] += self.transformer(
partial_latent_model_input,
t_expand,
text_states=input_prompt_embeds,
text_mask=input_prompt_mask,
text_states_2=input_prompt_embeds_2,
freqs_cos=freqs_cos,
freqs_sin=freqs_sin,
guidance=guidance_expand,
stg_block_idx=stg_block_idx,
stg_mode=stg_mode,
Expand Down

0 comments on commit da2fb54

Please sign in to comment.