Update LTX-Video/ltx_video/pipelines/pipeline_ltx_video.py
Browse files
LTX-Video/ltx_video/pipelines/pipeline_ltx_video.py
CHANGED
|
@@ -977,7 +977,7 @@ class LTXVideoPipeline(DiffusionPipeline):
|
|
| 977 |
print(f"[2ADUC DEBUG LTX *causal_video_autoencoder.py*]=======")
|
| 978 |
print(f"skip_initial_inference_steps {skip_initial_inference_steps}")
|
| 979 |
print(f"skip_final_inference_steps {skip_final_inference_steps}")
|
| 980 |
-
|
| 981 |
|
| 982 |
if self.allowed_inference_steps is not None:
|
| 983 |
for timestep in [round(x, 4) for x in timesteps.tolist()]:
|
|
@@ -1121,7 +1121,7 @@ class LTXVideoPipeline(DiffusionPipeline):
|
|
| 1121 |
print(f"[5ADUC DEBUG LTX *causal_video_autoencoder.py*]=======")
|
| 1122 |
print(f"skip_initial_inference_steps {skip_initial_inference_steps}")
|
| 1123 |
print(f"skip_final_inference_steps {skip_final_inference_steps}")
|
| 1124 |
-
|
| 1125 |
|
| 1126 |
|
| 1127 |
# Update the latents with the conditioning items and patchify them into (b, n, c)
|
|
@@ -1143,7 +1143,7 @@ class LTXVideoPipeline(DiffusionPipeline):
|
|
| 1143 |
print(f"[6ADUC DEBUG LTX *causal_video_autoencoder.py*]=======")
|
| 1144 |
print(f"skip_initial_inference_steps {skip_initial_inference_steps}")
|
| 1145 |
print(f"skip_final_inference_steps {skip_final_inference_steps}")
|
| 1146 |
-
|
| 1147 |
# 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
|
| 1148 |
extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
|
| 1149 |
|
|
@@ -1151,7 +1151,7 @@ class LTXVideoPipeline(DiffusionPipeline):
|
|
| 1151 |
print(f"[7ADUC DEBUG LTX *causal_video_autoencoder.py*]=======")
|
| 1152 |
print(f"skip_initial_inference_steps {skip_initial_inference_steps}")
|
| 1153 |
print(f"skip_final_inference_steps {skip_final_inference_steps}")
|
| 1154 |
-
|
| 1155 |
# 7. Denoising loop
|
| 1156 |
num_warmup_steps = max(
|
| 1157 |
len(timesteps) - num_inference_steps * self.scheduler.order, 0
|
|
@@ -1347,9 +1347,9 @@ class LTXVideoPipeline(DiffusionPipeline):
|
|
| 1347 |
print(f"[8ADUC DEBUG LTX *causal_video_autoencoder.py*]=======")
|
| 1348 |
print(f"skip_initial_inference_steps {skip_initial_inference_steps}")
|
| 1349 |
print(f"skip_final_inference_steps {skip_final_inference_steps}")
|
| 1350 |
-
|
| 1351 |
|
| 1352 |
-
|
| 1353 |
self.transformer = self.transformer.cpu()
|
| 1354 |
if self._execution_device == "cuda":
|
| 1355 |
torch.cuda.empty_cache()
|
|
|
|
| 977 |
print(f"[2ADUC DEBUG LTX *causal_video_autoencoder.py*]=======")
|
| 978 |
print(f"skip_initial_inference_steps {skip_initial_inference_steps}")
|
| 979 |
print(f"skip_final_inference_steps {skip_final_inference_steps}")
|
| 980 |
+
print(f"latents {latents.shape}")
|
| 981 |
|
| 982 |
if self.allowed_inference_steps is not None:
|
| 983 |
for timestep in [round(x, 4) for x in timesteps.tolist()]:
|
|
|
|
| 1121 |
print(f"[5ADUC DEBUG LTX *causal_video_autoencoder.py*]=======")
|
| 1122 |
print(f"skip_initial_inference_steps {skip_initial_inference_steps}")
|
| 1123 |
print(f"skip_final_inference_steps {skip_final_inference_steps}")
|
| 1124 |
+
print(f"latents {latents.shape}")
|
| 1125 |
|
| 1126 |
|
| 1127 |
# Update the latents with the conditioning items and patchify them into (b, n, c)
|
|
|
|
| 1143 |
print(f"[6ADUC DEBUG LTX *causal_video_autoencoder.py*]=======")
|
| 1144 |
print(f"skip_initial_inference_steps {skip_initial_inference_steps}")
|
| 1145 |
print(f"skip_final_inference_steps {skip_final_inference_steps}")
|
| 1146 |
+
print(f"latents {latents.shape}")
|
| 1147 |
# 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
|
| 1148 |
extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
|
| 1149 |
|
|
|
|
| 1151 |
print(f"[7ADUC DEBUG LTX *causal_video_autoencoder.py*]=======")
|
| 1152 |
print(f"skip_initial_inference_steps {skip_initial_inference_steps}")
|
| 1153 |
print(f"skip_final_inference_steps {skip_final_inference_steps}")
|
| 1154 |
+
print(f"latents {latents.shape}")
|
| 1155 |
# 7. Denoising loop
|
| 1156 |
num_warmup_steps = max(
|
| 1157 |
len(timesteps) - num_inference_steps * self.scheduler.order, 0
|
|
|
|
| 1347 |
print(f"[8ADUC DEBUG LTX *causal_video_autoencoder.py*]=======")
|
| 1348 |
print(f"skip_initial_inference_steps {skip_initial_inference_steps}")
|
| 1349 |
print(f"skip_final_inference_steps {skip_final_inference_steps}")
|
| 1350 |
+
print(f"latents {latents.shape}")
|
| 1351 |
|
| 1352 |
+
if offload_to_cpu:
|
| 1353 |
self.transformer = self.transformer.cpu()
|
| 1354 |
if self._execution_device == "cuda":
|
| 1355 |
torch.cuda.empty_cache()
|