Skip to content

Open-Sora: Democratizing Efficient Video Production for All

Python

debug

Shen-Chenhuice0b1928e618
diff --git a/scripts/inference-debug.py b/scripts/inference-debug.py
index 6f21ab3..4b739ad 100644
--- a/scripts/inference-debug.py
+++ b/scripts/inference-debug.py
@@ -163,9 +163,9 @@ def main():
# load model using booster
print("loading:", cfg.model["from_pretrained"])
booster.load_model(vae, os.path.join(cfg.model["from_pretrained"], "model"))
- booster.load_optimizer(optimizer, os.path.join(cfg.load, "optimizer"))
+ booster.load_optimizer(optimizer, os.path.join(cfg.model["from_pretrained"], "optimizer"))
if lr_scheduler is not None:
- booster.load_lr_scheduler(lr_scheduler, os.path.join(cfg.load, "lr_scheduler"))
+ booster.load_lr_scheduler(lr_scheduler, os.path.join(cfg.model["from_pretrained"], "lr_scheduler"))
# running_states = load_json(os.path.join(cfg.load, "running_states.json"))
dist.barrier()
# start_epoch, start_step, sampler_start_idx = running_states["epoch"], running_states["step"], running_states["sample_start_index"]
diff --git a/scripts/train-vae.py b/scripts/train-vae.py
index dc750b1..b9ca67e 100644
--- a/scripts/train-vae.py
+++ b/scripts/train-vae.py
@@ -181,10 +181,10 @@ def main():
# running_states = load_json(os.path.join(cfg.load, "running_states.json"))
dist.barrier()
# start_epoch, start_step, sampler_start_idx = running_states["epoch"], running_states["step"], running_states["sample_start_index"]
- logger.info(f"Loaded checkpoint {cfg.load} at epoch {start_epoch} step {start_step}")
- logger.info(f"Training for {cfg.epochs} epochs with {num_steps_per_epoch} steps per epoch")
+ # logger.info(f"Loaded checkpoint {cfg.load} at epoch {start_epoch} step {start_step}")
+ # logger.info(f"Training for {cfg.epochs} epochs with {num_steps_per_epoch} steps per epoch")
- dataloader.sampler.set_start_index(sampler_start_idx)
+ # dataloader.sampler.set_start_index(sampler_start_idx)
# # define loss function
# loss_function = VEA3DLoss(kl_weight=cfg.kl_weight, perceptual_weight=cfg.perceptual_weight).to(device, dtype)