d6-500iter calibration

Dryrun Job Failed
UUID:002185a0-3f67-4c53-a0e0-90d6d6487a1f
Tao Linvia🐎 pentiumCreated 213 days agoUpdated 213 days ago

Purpose

Calibration run to measure per-step throughput on A100 40GB with d6 config.

Result: DRYRUN FAILED

Bug in MLflow logging code: used total_mem instead of total_memory for torch.cuda.get_device_properties(). Fixed in commit d742689.

Useful observations from partial run

  • A100-SXM4-40GB, Peak BF16 FLOPS: 3.12e+14
  • Dataset download: ~2 seconds (9 shards)
  • Tokenizer training: ~11 seconds
  • Code fetch + env setup: ~4 minutes
  • Total preprocessing before training: ~30 seconds (excluding env setup)
Dryrun JobFailed
2
Funding 
3
Funded Job 
4
Done 

Finance

Fundings at snapshot$0.00

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

Configuration

Repository
https://github.com/mlpatron/mlpatron-nanochat
Version
experiment/baseline-search → 37c37624a543dddf21010299d8685ee97c6f4c2c
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=6·eval_every=50·max_seq_len=2048·num_iterations=500(dryrun: 10)·num_data_shards=8·device_batch_size=16