scientific-summarizer-v1

Starter model repository for scientific summarization experiments.

Status

Fine-tuned checkpoint uploaded: v0.2 (trained on sft_review_queue_500.jsonl, 1 epoch, batch size 2). Pilot checkpoint from 10-row run is also available in repo history.

Base model plan

Recommended first baseline: facebook/bart-large-cnn

Dataset

  • YanJo199/scientific-papers-sft-v1
  • Start with pilot/sft_review_queue_10.jsonl for smoke tests
  • Scale to sft_review_queue_500.jsonl after validation

Training objective

Supervised fine-tuning for the summary task.

Evaluation

Latest evaluation (February 25, 2026):

  • Validation file: data/sft_valid_v6_summary.jsonl
  • Samples: 100 (summary task)
  • ROUGE-1: 0.4309
  • ROUGE-2: 0.3134
  • ROUGE-L: 0.3586
  • ROUGE-Lsum: 0.3584

Next release target

Run full validation (all rows) and publish a tuned v0.3 checkpoint with updated training settings.

Downloads last month
-
Safetensors
Model size
0.4B params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support