set patch_size_t as None by default

Author: zRzRzRzRzRzRzR

src/diffusers/models/transformers/cogvideox_transformer_3d.py

@@ -230,7 +230,7 @@ def __init__(
         sample_height: int = 60,
         sample_frames: int = 49,
         patch_size: int = 2,
-        patch_size_t: int = 2,
+        patch_size_t: Optional[int] = None,
         temporal_compression_ratio: int = 4,
         max_text_seq_length: int = 226,
         activation_fn: str = "gelu-approximate",

src/diffusers/pipelines/cogvideo/pipeline_cogvideox.py

@@ -368,12 +368,12 @@ def prepare_extra_step_kwargs(self, generator, eta):
             extra_step_kwargs["generator"] = generator
         return extra_step_kwargs
 
+    # Copied from diffusers.pipelines.latte.pipeline_latte.LattePipeline.check_inputs
     def check_inputs(
         self,
         prompt,
         height,
         width,
-        num_frames,
         negative_prompt,
         callback_on_step_end_tensor_inputs,
         prompt_embeds=None,
@@ -382,15 +382,6 @@ def check_inputs(
         if height % 8 != 0 or width % 8 != 0:
             raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
 
-        latent_frames = (num_frames - 1) // self.vae_scale_factor_temporal + 1
-        if (
-            self.transformer.config.patch_size_t is not None
-            and latent_frames % self.transformer.config.patch_size_t != 0
-        ):
-            raise ValueError(
-                f"Number of latent frames must be divisible by `{self.transformer.config.patch_size_t}` but got {latent_frames=}."
-            )
-
         if callback_on_step_end_tensor_inputs is not None and not all(
             k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
         ):
@@ -611,7 +602,6 @@ def __call__(
             prompt,
             height,
             width,
-            num_frames,
             negative_prompt,
             callback_on_step_end_tensor_inputs,
             prompt_embeds,
@@ -744,8 +734,7 @@ def __call__(
                     progress_bar.update()
 
         if not output_type == "latent":
-            breakpoint()
-            video = self.decode_latents(latents)
+            video = self.decode_latents(latents[:,1:])
             video = self.video_processor.postprocess_video(video=video, output_type=output_type)
         else:
             video = latents