d10 production baseline v2 (retry)

Completed
UUID:4c709e8f-0122-414f-be29-47597ebe160f
Tao Linvia🐎 pentiumCreated 213 days agoUpdated 212 days ago

COMPLETED ✓ — Production Baseline Validated

Final Results

MetricValue
val_bpb0.897
Training time44.0 min
Wall time55.1 min
MFU54.9%
Cost$3.95
Peak mem21.2 GB

val_bpb curve

Stepval_bpb
03.170
3001.090
6001.005
9000.955
12000.919
15000.897

vs Previous Baseline

  • MFU: 52.3% → 54.9% (+5%)
  • Wall time: 60 min → 55 min (5 min saved)
  • Cost: $4.33 → $3.95
  • val_bpb: 0.898 → 0.897 (identical)

This is the recommended production config for future experiments.

Dryrun Job 
Funding 
Funded Job 
Done 

Finance

Fundings at snapshot$17.66

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

Configuration

Repository
https://github.com/mlpatron/mlpatron-nanochat
Version
experiment/baseline-search → e39ceb9a3a373755eb29e04fe8c88563830ba089
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·tok_max_chars=1000000000·num_iterations=1500(dryrun: 20)·num_data_shards=14·device_batch_size=32