---
license: cc-by-4.0
tags:
  - instruction-following
  - sft
  - data-mentions
  - extraction
configs:
  - config_name: default
    data_files:
      - split: train
        path: train.jsonl
      - split: val
        path: val.jsonl
      - split: holdout
        path: holdout.jsonl
---

# Data-mention extraction SFT

Single-task ChatML `messages` dataset for data-mention extraction, built
from the GLiNER2 labels in `rafmacalaba/data-use-mentions`. Static
instructions live in the `system` message; dynamic text in the `user`
message; the assistant emits compact JSON
`{"data_mentions":[{"data_mention":"<span>","specificity_type":"named|descriptive|vague"}]}` (or
`{"data_mentions":[]}` when none qualify).

Each row also carries `corpus` (`prwp` or `fcv`) and `origin` (e.g.
`general_prwp`, `fcv_pads_east_asia`, `jdc_operational`, `refugee_pads`,
`reliefweb`) for corpus/origin-filtered training. Filter with
`finetune_lfm2_data_mention.py --corpus prwp` (or `--origin ...`).

`{"train": 59970, "val": 12325, "holdout": 12531}`