diff --git a/src/diffusers/pipelines/cogvideo/pipeline_cogvideox.py b/src/diffusers/pipelines/cogvideo/pipeline_cogvideox.py
index 99ae9025cd..4636844cd8 100644
--- a/src/diffusers/pipelines/cogvideo/pipeline_cogvideox.py
+++ b/src/diffusers/pipelines/cogvideo/pipeline_cogvideox.py
@@ -718,14 +718,15 @@ class CogVideoXPipeline(DiffusionPipeline, CogVideoXLoraLoaderMixin):
                 timestep = t.expand(latent_model_input.shape[0])
 
                 # predict noise model_output
-                noise_pred = self.transformer(
-                    hidden_states=latent_model_input,
-                    encoder_hidden_states=prompt_embeds,
-                    timestep=timestep,
-                    image_rotary_emb=image_rotary_emb,
-                    attention_kwargs=attention_kwargs,
-                    return_dict=False,
-                )[0]
+                with self.transformer.cache_context("cond_uncond"):
+                    noise_pred = self.transformer(
+                        hidden_states=latent_model_input,
+                        encoder_hidden_states=prompt_embeds,
+                        timestep=timestep,
+                        image_rotary_emb=image_rotary_emb,
+                        attention_kwargs=attention_kwargs,
+                        return_dict=False,
+                    )[0]
                 noise_pred = noise_pred.float()
 
                 # perform guidance
diff --git a/src/diffusers/pipelines/cogvideo/pipeline_cogvideox_fun_control.py b/src/diffusers/pipelines/cogvideo/pipeline_cogvideox_fun_control.py
index e37574ec9c..121c4280f0 100644
--- a/src/diffusers/pipelines/cogvideo/pipeline_cogvideox_fun_control.py
+++ b/src/diffusers/pipelines/cogvideo/pipeline_cogvideox_fun_control.py
@@ -784,14 +784,15 @@ class CogVideoXFunControlPipeline(DiffusionPipeline, CogVideoXLoraLoaderMixin):
                 timestep = t.expand(latent_model_input.shape[0])
 
                 # predict noise model_output
-                noise_pred = self.transformer(
-                    hidden_states=latent_model_input,
-                    encoder_hidden_states=prompt_embeds,
-                    timestep=timestep,
-                    image_rotary_emb=image_rotary_emb,
-                    attention_kwargs=attention_kwargs,
-                    return_dict=False,
-                )[0]
+                with self.transformer.cache_context("cond_uncond"):
+                    noise_pred = self.transformer(
+                        hidden_states=latent_model_input,
+                        encoder_hidden_states=prompt_embeds,
+                        timestep=timestep,
+                        image_rotary_emb=image_rotary_emb,
+                        attention_kwargs=attention_kwargs,
+                        return_dict=False,
+                    )[0]
                 noise_pred = noise_pred.float()
 
                 # perform guidance
diff --git a/src/diffusers/pipelines/cogvideo/pipeline_cogvideox_image2video.py b/src/diffusers/pipelines/cogvideo/pipeline_cogvideox_image2video.py
index 59d7c4cad5..25e62cb2b3 100644
--- a/src/diffusers/pipelines/cogvideo/pipeline_cogvideox_image2video.py
+++ b/src/diffusers/pipelines/cogvideo/pipeline_cogvideox_image2video.py
@@ -831,15 +831,16 @@ class CogVideoXImageToVideoPipeline(DiffusionPipeline, CogVideoXLoraLoaderMixin)
                 timestep = t.expand(latent_model_input.shape[0])
 
                 # predict noise model_output
-                noise_pred = self.transformer(
-                    hidden_states=latent_model_input,
-                    encoder_hidden_states=prompt_embeds,
-                    timestep=timestep,
-                    ofs=ofs_emb,
-                    image_rotary_emb=image_rotary_emb,
-                    attention_kwargs=attention_kwargs,
-                    return_dict=False,
-                )[0]
+                with self.transformer.cache_context("cond_uncond"):
+                    noise_pred = self.transformer(
+                        hidden_states=latent_model_input,
+                        encoder_hidden_states=prompt_embeds,
+                        timestep=timestep,
+                        ofs=ofs_emb,
+                        image_rotary_emb=image_rotary_emb,
+                        attention_kwargs=attention_kwargs,
+                        return_dict=False,
+                    )[0]
                 noise_pred = noise_pred.float()
 
                 # perform guidance
diff --git a/src/diffusers/pipelines/cogvideo/pipeline_cogvideox_video2video.py b/src/diffusers/pipelines/cogvideo/pipeline_cogvideox_video2video.py
index c4dc7e574f..46deec9b99 100644
--- a/src/diffusers/pipelines/cogvideo/pipeline_cogvideox_video2video.py
+++ b/src/diffusers/pipelines/cogvideo/pipeline_cogvideox_video2video.py
@@ -799,14 +799,15 @@ class CogVideoXVideoToVideoPipeline(DiffusionPipeline, CogVideoXLoraLoaderMixin)
                 timestep = t.expand(latent_model_input.shape[0])
 
                 # predict noise model_output
-                noise_pred = self.transformer(
-                    hidden_states=latent_model_input,
-                    encoder_hidden_states=prompt_embeds,
-                    timestep=timestep,
-                    image_rotary_emb=image_rotary_emb,
-                    attention_kwargs=attention_kwargs,
-                    return_dict=False,
-                )[0]
+                with self.transformer.cache_context("cond_uncond"):
+                    noise_pred = self.transformer(
+                        hidden_states=latent_model_input,
+                        encoder_hidden_states=prompt_embeds,
+                        timestep=timestep,
+                        image_rotary_emb=image_rotary_emb,
+                        attention_kwargs=attention_kwargs,
+                        return_dict=False,
+                    )[0]
                 noise_pred = noise_pred.float()
 
                 # perform guidance