build-small-hackathon/pozify-coach-summary-nemotron
LoRA adapter for Pozify grounded coach-summary generation.
Base Model
nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16
Training Data
- Train file:
/Users/h/Documents/Pozify/data/sft/coach_summary_train.jsonl
- Eval file:
/Users/h/Documents/Pozify/data/sft/coach_summary_eval.jsonl
- Style file:
/data/sft/public_fitness_style.jsonl
Objective
Generate grounded coach_summary.json outputs from Pozify structured evidence and knowledge cards.
Training Summary
- Train rows:
68
- Eval rows:
10
- Style rows mixed in:
14
- Output dir:
/models/adapter
Evaluation
- JSON validity rate:
0.0
- Verifier pass rate:
0.0
- Section completeness rate:
0.0
Packaging
- Published as a merged, inference-ready Transformers checkpoint.
