The problem about training log file

[TensorsSun]

Hello,
I would like to ask why the training loss and metrics are not recorded in the run_training.log file. Is this caused by some kind of bug? I can only see this information in my command line. Here is my run_training.log file after 100 epochs:

[2025-12-30 05:01:20,941][__main__][INFO] - Global Seed set to 0
[2025-12-30 05:01:20,944][__main__][INFO] - Path where all results are stored: /mnt/hwdata/xiaolong/DiffusionDriveV2/navsim/exp/training_diffusiondrive_agent/2025.12.30.05.01.11
[2025-12-30 05:01:20,944][__main__][INFO] - Building Agent
[2025-12-30 05:01:22,372][timm.models._builder][INFO] - Loading pretrained weights from Hugging Face hub (timm/resnet34.a1_in1k)
[2025-12-30 05:01:22,920][httpx][INFO] - HTTP Request: HEAD https://hf-mirror.com/timm/resnet34.a1_in1k/resolve/main/model.safetensors "HTTP/1.1 302 Found"
[2025-12-30 05:01:22,921][timm.models._hub][INFO] - [timm/resnet34.a1_in1k] Safe alternative available for 'pytorch_model.bin' (as 'model.safetensors'). Loading weights using safetensors.
[2025-12-30 05:01:22,946][timm.models._builder][INFO] - Missing keys (fc.weight, fc.bias) discovered while loading pretrained weights. This is expected if model is being adapted.
[2025-12-30 05:01:23,465][__main__][INFO] - Building Lightning Module
[2025-12-30 05:01:23,473][torch.distributed.nn.jit.instantiator][INFO] - Created a temporary directory at /tmp/tmp7rnwt3wn
[2025-12-30 05:01:23,473][torch.distributed.nn.jit.instantiator][INFO] - Writing /tmp/tmp7rnwt3wn/_remote_module_non_scriptable.py
[2025-12-30 05:01:23,487][__main__][INFO] - Using cached data without building SceneLoader
[2025-12-30 05:06:15,392][__main__][INFO] - Building Datasets
[2025-12-30 05:06:15,393][__main__][INFO] - Num training samples: 85109
[2025-12-30 05:06:15,394][__main__][INFO] - Num validation samples: 18179
[2025-12-30 05:06:15,394][__main__][INFO] - Building Trainer
[2025-12-30 05:06:15,594][__main__][INFO] - Starting Training
[2025-12-30 05:09:14,259][torch.distributed.distributed_c10d][INFO] - Added key: store_based_barrier_key:1 to store for rank: 0
[2025-12-30 05:09:14,259][torch.distributed.distributed_c10d][INFO] - Rank 0: Completed store-based barrier for key:store_based_barrier_key:1 with 8 nodes.

[Doctor-James]

It uses TensorBoard for logging by default. You can check if there is a corresponding event file.

[TensorsSun]

It uses TensorBoard for logging by default. You can check if there is a corresponding event file.

Thanks for your help! However, when I performed the evaluation, I still couldn't see the results. Where can I view the metrics? Here is the final output of my evaluation:

───────────────────────────────────────────────────────────────────────────────────
      Validate metric             DataLoader 0
───────────────────────────────────────────────────────────────────────────────────
 val/agent_box_loss_epoch       6.573030471801758
val/agent_class_loss_epoch      2.125600814819336
val/bev_semantic_loss_epoch     6.85503625869751
 val/diffusion_loss_epoch              0.0
      val/loss_epoch           18.711817286731566
 val/trajectory_loss_epoch     3.1581502086070423
───────────────────────────────────────────────────────────────────────────────────

[Doctor-James]

Which script did you run for the test? Is this the result from diffusiondrivev1?

[TensorsSun]

I used the script provided by DiffusionDrivev2, but I modified it to evaluate the trained DiffusionDrivev1 model. Here is the script:

python $NAVSIM_DEVKIT_ROOT/navsim/planning/script/run_pdm_score_fast.py \
agent=diffusiondrive_agent \
experiment_name=diffusiondrive_agent_eval \
train_test_split=navtest \
'agent.checkpoint_path="/mnt/hwdata/xiaolong/DiffusionDriveV2/navsim/exp/training_diffusiondrive_agent/2025.12.30.05.01.11/lightning_logs/version_0/checkpoints/epoch=99-step=16700.ckpt"' \
+metric_cache_path="${NAVSIM_EXP_ROOT}/metric_cache/" \
+test_cache_path="${NAVSIM_EXP_ROOT}/metric_feature_cache/"

[Doctor-James]

You might need to modify navsim/agents/diffusiondrive/transfuser_agent.py by following the logic in navsim/agents/diffusiondrivev2/diffusiondrivev2_sel_agent.py. Alternatively, you can directly use the official evaluation script.