EasyDeepLearn

What should you track during LLM fine-tuning?

medium

Answer

  • (1) Loss per step (train + val).
  • (2) Perplexity on held-out.
  • (3) Task-specific metrics: BLEU / ROUGE / accuracy / rubric score.
  • (4) Sample outputs at regular intervals — read them.
  • (5) LR schedule + gradient norm + weight norm.
  • (6) GPU utilization + throughput (tokens/sec).
  • (7) OOM / crash logs.
  • (8) Data mixture per step (for multi-source).
  • (9) Reference model divergence (KL) if RLHF/DPO.
  • (10) Downstream benchmark (MMLU / HellaSwag / your-eval) at checkpoints.
  • Track via W&B / MLflow.
Check yourself — multiple choice
  • Just loss
  • Loss/perplexity + task metrics + sample outputs + LR/grad/weight norm + GPU/throughput + OOM logs + data mixture + KL ref divergence + downstream bench at ckpts; W&B/MLflow
  • Random
  • Not real

Fine-tune tracking: loss + task + samples + norms + GPU + KL + downstream.

#llmops#reproducibility

Practise MLOps & Data Quality

215 interview questions in this topic.

Related questions