d12 SFT (from pretrain checkpoint, with artifact upload)

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. Upload SFT checkpoint to MLflow artifacts (for downstream RL)
5. num_iterations=9999 as cap; epoch-based stopping triggers first (~600 steps)

Expected: ~20 min on 1×H100, ~$2-3

Status: dryrun_failed — IndentationError in inline Python. YAML >- folding adds leading spaces to continuation lines. Fixed by collapsing to single line.