d10-1500iter (~1hr target)

Completed
UUID:b6af0936-5f89-4d64-8e4c-b77d467f311c
Tao Linvia🐎 pentiumCreated 213 days agoUpdated 213 days ago

Results: COMPLETED ✓

Validation BPB Curve

Stepval_bpb
03.172
1001.400
2001.146
300~1.05
400~1.00
500~0.97
600~0.96
700~0.95
8000.970
900~0.95
1000~0.94
11000.930
1200~0.92
1300~0.91
1400~0.90
15000.898

Key Metrics

  • Final val_bpb: 0.898 (min=0.898 at step 1500, still decreasing)
  • Training time: 46.0 min
  • Total wall time: 70.6 min (incl. ~5min preprocessing, ~19min eval)
  • Actual cost: $5.07 (estimated was $16.43)
  • MFU: 52.7%
  • Peak mem: 11.6 GB
  • Step time: ~1.85s

Observations

  • val_bpb still decreasing at step 1500 → could benefit from more iterations
  • Eval overhead significant: ~19 min for 16 eval points
  • If eval_every=200 instead of 100, wall time would be ~60 min
  • d10-1800iter with eval_every=200 would fill ~1hr perfectly
  • Far better than d6-500iter (0.898 vs 1.096)
Dryrun Job 
Funding 
Funded Job 
Done 

Finance

Fundings at snapshot$16.80

Individual funding and transaction records are not included in the public archive.

Configuration

Repository
https://github.com/mlpatron/mlpatron-nanochat
Version
experiment/baseline-search → d7426893a51c9b71bf739e3ef94c13b456b43731
Image
ghcr.io/nblintao/mlpatron-nanochat:latest
→ ghcr.io/nblintao/mlpatron-nanochat@sha256:28c5544c206237e407241b0a72a8707e63e52750f487b751675b51ff1bce1520
Entrypoint
a100_preset
Resources
NVIDIA A100 40GB × 1·a2-highgpu-1g (12 vCPU / 85 GiB)
Parameters
depth=10·eval_every=100·max_seq_len=2048·num_iterations=1500(dryrun: 20)·num_data_shards=12·device_batch_size=16