Ideas (1)
BatchNorm in a small MNIST CNN
mnist-cnn-batchnorm completed e2ecvmnistsmoke 58s agoInsert BatchNorm2d after each convolution of a two-block MNIST CNN and compare against the identical network without it. Same data, same hyperparameters, same seed, same number of epochs - the only difference is the normalisation layer. Used as the end-to-end test of the automation pipeline, so it is deliberately small and fast.
Hypothesis: BatchNorm raises MNIST test accuracy over the plain CNN at an equal training budget, without increasing wall-clock time meaningfully.
repo: /Users/makartkar/.agents/.claude/worktrees/ml-experiment-automation-116c14/clis/mlexp/examples/mnist_cnn/mnist-cnn
| baseline: plain-cnn
| Run | Status | Pipeline | Compute | Metrics | Updated |
|---|---|---|---|---|---|
| e2e-resume-001 | completed |
Idea
Environment
Baseline
Setup
Implement
GPUs
Venv
Data
Tests
Train
Monitor
Eval
Report
|
slurm / capella 1x H100 MIG job 3922209 |
loss 0.0539
val_loss 0.0426
val_acc 0.9872
variant idea
test_loss 0.040121
test_acc 0.9866
test_error_pct 1.34
eval_seconds 0.26
device cuda
n_test 10000
base: variant=baseline test_loss=0.043111 test_acc=0.9862 test_error_pct=1.38 eval_seconds=0.23 device=cuda n_test=10000
|
57s ago |
| e2e-dresden-005 | completed |
Idea
Environment
Baseline
Setup
Implement
GPUs
Venv
Data
Tests
Train
Monitor
Eval
Report
|
slurm / capella 1x H100 MIG job 3922209 |
loss 0.0534
val_loss 0.0428
val_acc 0.9882
variant idea
test_loss 0.041019
test_acc 0.9857
test_error_pct 1.43
eval_seconds 0.34
device cuda
n_test 10000
base: variant=baseline test_loss=0.044817 test_acc=0.9854 test_error_pct=1.46 eval_seconds=0.38 device=cuda n_test=10000
|
10m ago |
| e2e-local-006 | completed |
Idea
Environment
Baseline
Setup
Implement
GPUs
Venv
Data
Tests
Train
Monitor
Eval
Report
|
local 0x cpu |
loss 0.0538
val_loss 0.0433
val_acc 0.9876
variant idea
test_loss 0.04103
test_acc 0.9867
test_error_pct 1.33
eval_seconds 0.45
device mps
n_test 10000
base: variant=baseline test_loss=0.04256 test_acc=0.9856 test_error_pct=1.44 eval_seconds=0.22 device=mps n_test=10000
|
47m ago |
Recent activity
58s ago
report
e2e-resume-001 report written to /Users/makartkar/.local/state/mlexp/runs/e2e-resume-001/report.md
1m ago
evaluate
e2e-resume-001 eval_baseline metrics: {'variant': 'baseline', 'test_loss': 0.043111, 'test_acc': 0.9862, 'test_error_pct': 1.38, 'eval_seconds': 0.23, 'device': 'cuda', 'n_test': 10000}
1m ago
evaluate
e2e-resume-001 'eval_baseline' still running (4s): RUN_MARKER step=eval_baseline exit_code=0 end_epoch=1786969732
1m ago
evaluate
e2e-resume-001 launched 'eval_baseline' (pid 1545847) -> /data/horse/ws/arma647i-mlexp/runs/e2e-resume-001/eval_baseline.log
1m ago
evaluate
e2e-resume-001 eval_idea metrics: {'variant': 'idea', 'test_loss': 0.040121, 'test_acc': 0.9866, 'test_error_pct': 1.34, 'eval_seconds': 0.26, 'device': 'cuda', 'n_test': 10000}
2m ago
evaluate
e2e-resume-001 'eval_idea' still running (4s): RUN_MARKER step=eval_idea exit_code=0 end_epoch=1786969689
2m ago
evaluate
e2e-resume-001 launched 'eval_idea' (pid 1541805) -> /data/horse/ws/arma647i-mlexp/runs/e2e-resume-001/eval_idea.log
3m ago
train
e2e-resume-001 baseline training started -> /data/horse/ws/arma647i-mlexp/runs/e2e-resume-001/train_baseline.log
4m ago
train
e2e-resume-001 training started -> /data/horse/ws/arma647i-mlexp/runs/e2e-resume-001/train_idea.log
4m ago
testing
e2e-resume-001 launched 'smoke_run' (pid 1527521) -> /data/horse/ws/arma647i-mlexp/runs/e2e-resume-001/smoke_run.log
5m ago
testing
e2e-resume-001 'unit_tests' still running (4s): RUN_MARKER step=unit_tests exit_code=0 end_epoch=1786969483
5m ago
testing
e2e-resume-001 launched 'unit_tests' (pid 1519682) -> /data/horse/ws/arma647i-mlexp/runs/e2e-resume-001/unit_tests.log
6m ago
data_prep
e2e-resume-001 data cache already prepared (/data/horse/ws/arma647i-mlexp/.data/mnist/mnist.npz)