Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
base_model: facebook/bart-large-cnn
|
| 4 |
+
tags: [summarization, scientific-papers, highlights, scihigh-2026, fire-2026]
|
| 5 |
+
---
|
| 6 |
+
|
| 7 |
+
# SciHigh-2026 Subtask 1 — bart-large-cnn titled (run 1)
|
| 8 |
+
|
| 9 |
+
Full fine-tune of bart-large-cnn.
|
| 10 |
+
|
| 11 |
+
Fine-tuned for the FIRE 2026 SciHigh shared task, Subtask 1: generating
|
| 12 |
+
research highlights from scientific paper abstracts (MixSub corpus).
|
| 13 |
+
|
| 14 |
+
- Input format: `<paper title> | <abstract>` (corpus punctuation-stripped
|
| 15 |
+
style; abstracts repaired via DOI-verified Semantic Scholar recovery)
|
| 16 |
+
- Output: 3-5 highlight sentences, ~55 words, greedy decoding
|
| 17 |
+
- Training: one epoch on 15,960 pairs, Kaggle dual T4
|
| 18 |
+
- Validation ROUGE-L F1 (plain LCS, no stemming): **26.37**
|
| 19 |
+
|
| 20 |
+
Pipeline and evaluation protocol: see the `final/` directory of the
|
| 21 |
+
project repository. Trained 2026-08-08.
|