d12 SFT (from pretrain checkpoint)

Cancelled
UUID:b9037d59-96bd-4c04-9140-47bb1994b29c
Tao Linvia🐎 pentiumCreated 209 days agoUpdated 209 days ago

SFT on d12 pretrain checkpoint

Pretrain source: d12 H100 SSSL final (val_bpb=0.854, step 2205)

Plan:

  1. Download pretrain checkpoint from MLflow artifacts (~2GB)
  2. Train tokenizer (deterministic, ~80s)
  3. Run SFT with default mixture: SmolTalk 460K + MMLU×3 + GSM8K×4 + Identity×2 + Spelling 280K = ~1.07M rows
  4. num_iterations=9999 as cap; epoch-based stopping triggers first (~600 steps)

Expected: ~20 min on 1×H100, ~$2-3

Status: cancelled — missing --upload-checkpoint flag. Recreated with upload enabled.

Collapsedv2 · Edited 209 days ago
—
Dryrun Job 
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
master → f988ab981d4d1b9072c6f5423f1e29425f715903
Image
ghcr.io/mlpatron/mlpatron-nanochat:latest
→ ghcr.io/mlpatron/mlpatron-nanochat@sha256:a142e86efdcd41e0355e70d46c341cbf9a86b2da9448459009d7b5e8209ffa61
Entrypoint
h100_d12_sft
Resources
NVIDIA H100 80GB × 1·a3-highgpu-1g (26 vCPU / 234 GiB)
Parameters
num_iterations=9999(dryrun: 20)·pretrain_mlflow_run_id=25207c33abca4a8197870adec4ead007