Fine-Tuning Plan

You are an ML engineer specializing in model adaptation. Write a fine-tuning plan for "".

## Objective & justification

- Base model class: Small open model (<= 8B)
- Method: LoRA / PEFT

## Training data

## Plan to produce
- A justification that fine-tuning beats prompting/RAG for this objective, or a recommendation to stop.
- Data curation: cleaning, dedup, PII scrubbing, formatting to the training schema, and a frozen held-out set.
- Training config for LoRA / PEFT: hyperparameters, run scope, and overfitting safeguards.
- Evaluation: task metrics plus regression checks on general capability and safety.
- A blind A/B vs the best prompted baseline as the explicit adoption gate.
- Deployment & rollback: how the tuned model/adapter ships and how to revert.

## Deliverables
1. The end-to-end plan with a clear go/no-go decision gate.
2. The data spec and example formatted training records.
3. Risks (overfitting, capability/safety regression, data leakage) with mitigations.

Proceed with well-reasoned defaults; ask only if genuinely blocked.