d10-1500iter eval_every=300 (optimized baseline)

Completed
UUID:9cb7805e-94ec-41db-8929-42a4303ef82a
Tao Linvia🐎 pentiumCreated 213 days agoUpdated 213 days ago

Results: COMPLETED ✓ — Wall time validated at 60 min!

Key Metrics

  • Final val_bpb: 0.898 (identical to eval@100 run)
  • Training time: 46.1 min
  • Total wall time: 60.4 min (vs 71 min with eval@100)
  • Actual cost: $4.33
  • MFU: 52.3%
  • Peak mem: 11.6 GB

Validation BPB Curve (eval_every=300)

Stepval_bpb
03.172
300~1.05
600~0.96
900~0.95
12000.920
15000.898

Conclusion

This is the recommended baseline configuration for ~1hr A100 runs:

  • depth=10, num_iterations=1500, eval_every=300
  • Matches Chinchilla-optimal within 7%
  • Fits exactly in 1 hour wall time
  • Cost: ~$4-5 per run
Dryrun Job 
Funding 
Funded Job 
Done 

Finance

Fundings at snapshot$17.40

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=300·max_seq_len=2048·num_iterations=1500(dryrun: 20)·num_data_shards=12·device_batch_size=16