update

app.py

@@ -132,26 +132,6 @@ def load_image_as_tensor(image_path: str) -> torch.Tensor:
     ])
     return transform(image)
 
-def run_local(cfg: DictConfig):
-    # delay some imports in case they are not needed in non-local envs for submission
-    from experiments import build_experiment
-
-    # Get yaml names
-    hydra_cfg = hydra.core.hydra_config.HydraConfig.get()
-    cfg_choice = OmegaConf.to_container(hydra_cfg.runtime.choices)
-
-    with open_dict(cfg):
-        if cfg_choice["experiment"] is not None:
-            cfg.experiment._name = cfg_choice["experiment"]
-        if cfg_choice["dataset"] is not None:
-            cfg.dataset._name = cfg_choice["dataset"]
-        if cfg_choice["algorithm"] is not None:
-            cfg.algorithm._name = cfg_choice["algorithm"]
-
-    # launch experiment
-    experiment = build_experiment(cfg, None, None)
-    return experiment.exec_interactive(cfg.experiment.tasks[0])
-
 def enable_amp(model, precision="16-mixed"):
     original_forward = model.forward
 
@@ -193,7 +173,7 @@ load_custom_checkpoint(algo=worldmem.diffusion_model, checkpoint_path=cfg.diffus
 load_custom_checkpoint(algo=worldmem.vae, checkpoint_path=cfg.vae_path)
 load_custom_checkpoint(algo=worldmem.pose_prediction_model, checkpoint_path=cfg.pose_predictor_path)
 worldmem.to("cuda").eval()
-worldmem = enable_amp(worldmem, precision="16-mixed")
+# worldmem = enable_amp(worldmem, precision="16-mixed")
 
 actions = np.zeros((1, 25), dtype=np.float32)
 poses = np.zeros((1, 5), dtype=np.float32)
@@ -555,7 +535,6 @@ with gr.Blocks(css=css) as demo:
     )
 
 
-    # input_box.submit(update_image_and_log, inputs=[input_box], outputs=[image_display, video_display, log_output])
     submit_button.click(generate, inputs=[input_box, log_output, memory_frames, self_frames, self_actions, self_poses, self_memory_c2w, self_frame_idx], outputs=[image_display, video_display, log_output, memory_frames, self_frames, self_actions, self_poses, self_memory_c2w, self_frame_idx])
     reset_btn.click(reset, inputs=[selected_image], outputs=[log_output, memory_frames, self_frames, self_actions, self_poses, self_memory_c2w, self_frame_idx])
     image_display_1.select(lambda: on_image_click(SUNFLOWERS_IMAGE), outputs=[log_output, selected_image, image_display, memory_frames, self_frames, self_actions, self_poses, self_memory_c2w, self_frame_idx])