Chart Overlays
Compare this run against its peers
Add peer runs here to layer their train and val traces onto the current metric charts.
running Apr 23, 11:56 Started 10d 1h ago Python 3.11.7
train_joint.py --config /cluster/home/drothenpiele/models/stable_diffusion_mari/mari/.euler-launches/euler_launch_55da2d57-3198-47ba-be27-7a9cab62110e/joint_vkitti.yamllr 4e-5batch 8epochs 200precision fp16model stable-diffusion-2-1 seed 42wd 0grad accum 1warmup 0.05000ema nograd clip 1
ID 64455826
mari/runs/2026-04-23_11-56-07_6596Metrics 23/23
depth.train 6
diag 3
depth_abs_rel
depth_delta1
depth_mae
loss 3
depth
depth_head
vae_rec
rgb.train 11
diag 2
image_psnr
image_ssim
loss 7
dehaze
preserve
total
visibility
visibility_aux
visibility_rank
visibility_tv
stat 2
visibility_mean
visibility_std
sys.train 6
gpu_mem_total_gb
gpu_mem_used_gb
gpu_mem_util_pct
gpu_util_pct
lr
visibility_aux_scale
23 selected
X-Axis
Y-Scale
Series
Group
Namespace
Depth Abs Rel
↓ kind=diag V
2026-04-23_11-56-07_6596 val (smoothed) —
2026-04-23_11-56-07_6596 val (raw) —
Depth Delta1
↑ kind=diag V
2026-04-23_11-56-07_6596 val (smoothed) —
2026-04-23_11-56-07_6596 val (raw) —
Depth Mae
kind=diag V
2026-04-23_11-56-07_6596 val (smoothed) —
2026-04-23_11-56-07_6596 val (raw) —
Depth
↓ kind=loss T
2026-04-23_11-56-07_6596 train (smoothed) —
2026-04-23_11-56-07_6596 train (raw) —
Depth Head
kind=loss T
2026-04-23_11-56-07_6596 train (smoothed) —
2026-04-23_11-56-07_6596 train (raw) —
Vae Rec
kind=loss T
2026-04-23_11-56-07_6596 train (smoothed) —
2026-04-23_11-56-07_6596 train (raw) —
Image Psnr
↑ dB kind=diag V
2026-04-23_11-56-07_6596 val (smoothed) —
2026-04-23_11-56-07_6596 val (raw) —
Image Ssim
↑ kind=diag V
2026-04-23_11-56-07_6596 val (smoothed) —
2026-04-23_11-56-07_6596 val (raw) —
Dehaze
↓ kind=loss T
2026-04-23_11-56-07_6596 train (smoothed) —
2026-04-23_11-56-07_6596 train (raw) —
Preserve
↓ kind=loss T
2026-04-23_11-56-07_6596 train (smoothed) —
2026-04-23_11-56-07_6596 train (raw) —
Total
↓ kind=loss T
2026-04-23_11-56-07_6596 train (smoothed) —
2026-04-23_11-56-07_6596 train (raw) —
Visibility
↓ kind=loss T
2026-04-23_11-56-07_6596 train (smoothed) —
2026-04-23_11-56-07_6596 train (raw) —
Visibility Aux
↓ kind=loss T
2026-04-23_11-56-07_6596 train (smoothed) —
2026-04-23_11-56-07_6596 train (raw) —
Visibility Rank
↓ kind=loss T
2026-04-23_11-56-07_6596 train (smoothed) —
2026-04-23_11-56-07_6596 train (raw) —
Visibility Tv
↓ kind=loss T
2026-04-23_11-56-07_6596 train (smoothed) —
2026-04-23_11-56-07_6596 train (raw) —
Visibility Mean
kind=stat T V
2026-04-23_11-56-07_6596 train (smoothed) —
2026-04-23_11-56-07_6596 val (smoothed) —
2026-04-23_11-56-07_6596 train (raw) —
2026-04-23_11-56-07_6596 val (raw) —
Visibility Std
kind=stat T V
2026-04-23_11-56-07_6596 train (smoothed) —
2026-04-23_11-56-07_6596 val (smoothed) —
2026-04-23_11-56-07_6596 train (raw) —
2026-04-23_11-56-07_6596 val (raw) —
Gpu Mem Total Gb
T
2026-04-23_11-56-07_6596 train (smoothed) —
2026-04-23_11-56-07_6596 train (raw) —
Gpu Mem Used Gb
T
2026-04-23_11-56-07_6596 train (smoothed) —
2026-04-23_11-56-07_6596 train (raw) —
Gpu Mem Util Pct
T
2026-04-23_11-56-07_6596 train (smoothed) —
2026-04-23_11-56-07_6596 train (raw) —
Gpu Util Pct
T
2026-04-23_11-56-07_6596 train (smoothed) —
2026-04-23_11-56-07_6596 train (raw) —
Lr
T
2026-04-23_11-56-07_6596 train (smoothed) —
2026-04-23_11-56-07_6596 train (raw) —
Visibility Aux Scale
T
2026-04-23_11-56-07_6596 train (smoothed) —
2026-04-23_11-56-07_6596 train (raw) —
No output snapshots found for this run.
Outputs are generated during training and saved to outputs/epoch_N_step_M/ directories.
Producer launch exports are available. Manage launch-owned exports here for quick reference.
Open LaunchInherited Launch Exports
These exports are published by the run's producer launch.
Published
The producer launch does not publish any exports yet.
Run-Owned Exports
Publish direct filesystem paths here.
Published
No run-owned exports are published yet.
Raw Artifacts
Run-owned exports are typically direct paths, so there are no captured artifacts to publish from here.