Update app.py

app.py

@@ -58,10 +58,10 @@ pipe = CogVideoXPipeline.from_pretrained(
     transformer=transformer,
     vae=vae,
     torch_dtype=torch.bfloat16
-    ).to(device)
+    ).to("cpu")
 pipe.scheduler = CogVideoXDPMScheduler.from_config(pipe.scheduler.config, timestep_spacing="trailing")
 
-# pipe.enable_model_cpu_offload()
+pipe.enable_model_cpu_offload()
 pipe.vae.enable_tiling()
 pipe.vae.enable_slicing()
 
@@ -261,9 +261,9 @@ def infer(
             use_dynamic_cfg=True,
             output_type="pt",
             guidance_scale=guidance_scale,
-            generator=torch.Generator(device=device).manual_seed(seed),
+            generator=torch.Generator(device="cpu").manual_seed(seed),
         ).frames
-        pipe_video.to(device)
+        pipe_video.to("cpu")
         del pipe_video
         gc.collect()
         torch.cuda.empty_cache()
@@ -287,9 +287,9 @@ def infer(
             use_dynamic_cfg=True,
             output_type="pt",
             guidance_scale=guidance_scale,
-            generator=torch.Generator(device=device).manual_seed(seed),
+            generator=torch.Generator(device="cpu").manual_seed(seed),
         ).frames
-        pipe_image.to(device)
+        pipe_image.to("cpu")
         del pipe_image
         gc.collect()
         torch.cuda.empty_cache()
@@ -303,9 +303,9 @@ def infer(
             use_dynamic_cfg=True,
             output_type="pt",
             guidance_scale=guidance_scale,
-            generator=torch.Generator(device=device).manual_seed(seed),
+            generator=torch.Generator(device="cpu").manual_seed(seed),
         ).frames
-        pipe.to(device)
+        pipe.to("cpu")
         gc.collect()
     return (video_pt, seed)