INFO:root:Extracted global_step 10000 from checkpoint path /root/work/Isaac-GR00T/.venv/lib/python3.12/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1. check_for_updates() Flash Attention 2 only supports torch.float16 and torch.bfloat16 dtypes, but the current dype in Qwen3VLForConditionalGeneration is torch.float32. You should run training or inference using Automatic Mixed-Precision via the `with torch.autocast(device_type='torch_device'):` decorator, or load the model with the `dtype` argument. Example: `model = AutoModel.from_pretrained("openai/whisper-tiny", attn_implementation="flash_attention_2", dtype=torch.float16)` Flash Attention 2 only supports torch.float16 and torch.bfloat16 dtypes, but the current dype in Qwen3VLModel is torch.float32. You should run training or inference using Automatic Mixed-Precision via the `with torch.autocast(device_type='torch_device'):` decorator, or load the model with the `dtype` argument. Example: `model = AutoModel.from_pretrained("openai/whisper-tiny", attn_implementation="flash_attention_2", dtype=torch.float16)` Flash Attention 2 only supports torch.float16 and torch.bfloat16 dtypes, but the current dype in Qwen3VLVisionModel is torch.float32. You should run training or inference using Automatic Mixed-Precision via the `with torch.autocast(device_type='torch_device'):` decorator, or load the model with the `dtype` argument. Example: `model = AutoModel.from_pretrained("openai/whisper-tiny", attn_implementation="flash_attention_2", dtype=torch.float16)` Flash Attention 2 only supports torch.float16 and torch.bfloat16 dtypes, but the current dype in Qwen3VLTextModel is torch.float32. You should run training or inference using Automatic Mixed-Precision via the `with torch.autocast(device_type='torch_device'):` decorator, or load the model with the `dtype` argument. Example: `model = AutoModel.from_pretrained("openai/whisper-tiny", attn_implementation="flash_attention_2", dtype=torch.float16)` Total number of DiT parameters: 1091722240 INFO:gr00t.model.gr00t_n1d7.gr00t_n1d7:Using AlternateVLDiT for diffusion model Total number of SelfAttentionTransformer parameters: 201433088 Loading checkpoint shards: 0%| | 0/3 [00:00, type=, format=, state_key=None), ActionConfig(rep=, type=, format=, state_key=None), ActionConfig(rep=, type=, format=, state_key=None), ActionConfig(rep=, type=, format=, state_key=None)]), 'language': ModalityConfig(delta_indices=[0], modality_keys=['annotation.human.task_description'], sin_cos_embedding_keys=None, mean_std_embedding_keys=None, action_configs=None)} INFO:root:Dataset length: 900 INFO:root:Running evaluation on trajectories: [0, 300, 600] INFO:root:Running trajectory: 0 INFO:root:Using 143 steps (requested: 300, trajectory length: 143) INFO:root:inferencing at step: 0 INFO:root:inferencing at step: 16 INFO:root:inferencing at step: 32 INFO:root:inferencing at step: 48 INFO:root:inferencing at step: 64 INFO:root:inferencing at step: 80 INFO:root:inferencing at step: 96 INFO:root:inferencing at step: 112 INFO:root:inferencing at step: 128 INFO:root:Unnormalized Action MSE across single traj: 0.00448231864720583 INFO:root:Unnormalized Action MAE across single traj: 0.03181375935673714 INFO:root:state_joints vs time (143, 13) INFO:root:gt_action_joints vs time (143, 13) INFO:root:pred_action_joints vs time (143, 13) INFO:root:MSE for trajectory 0: 0.00448231864720583, MAE: 0.03181375935673714 INFO:root:Running trajectory: 300 INFO:root:Using 108 steps (requested: 300, trajectory length: 108) INFO:root:inferencing at step: 0 INFO:root:inferencing at step: 16 INFO:root:inferencing at step: 32 INFO:root:inferencing at step: 48 INFO:root:inferencing at step: 64 INFO:root:inferencing at step: 80 INFO:root:inferencing at step: 96 INFO:root:Unnormalized Action MSE across single traj: 0.002719493815675378 INFO:root:Unnormalized Action MAE across single traj: 0.0312812440097332 INFO:root:state_joints vs time (108, 13) INFO:root:gt_action_joints vs time (108, 13) INFO:root:pred_action_joints vs time (108, 13) INFO:root:MSE for trajectory 300: 0.002719493815675378, MAE: 0.0312812440097332 INFO:root:Running trajectory: 600 INFO:root:Using 270 steps (requested: 300, trajectory length: 270) INFO:root:inferencing at step: 0 INFO:root:inferencing at step: 16 INFO:root:inferencing at step: 32 INFO:root:inferencing at step: 48 INFO:root:inferencing at step: 64 INFO:root:inferencing at step: 80 INFO:root:inferencing at step: 96 INFO:root:inferencing at step: 112 INFO:root:inferencing at step: 128 INFO:root:inferencing at step: 144 INFO:root:inferencing at step: 160 INFO:root:inferencing at step: 176 INFO:root:inferencing at step: 192 INFO:root:inferencing at step: 208 INFO:root:inferencing at step: 224 INFO:root:inferencing at step: 240 INFO:root:inferencing at step: 256 INFO:root:Unnormalized Action MSE across single traj: 0.001021101139485836 INFO:root:Unnormalized Action MAE across single traj: 0.0182601697742939 INFO:root:state_joints vs time (270, 13) INFO:root:gt_action_joints vs time (270, 13) INFO:root:pred_action_joints vs time (270, 13) INFO:root:MSE for trajectory 600: 0.001021101139485836, MAE: 0.0182601697742939 INFO:root:Average MSE across all trajs: 0.0027409710455685854 INFO:root:Average MAE across all trajs: 0.027118390426039696 INFO:root:Done