Datasets:
docs: add arxiv: tags for paper bibliography anchors
Browse filesAdds 8 arxiv: tags to the YAML frontmatter so the paper's load-bearing references (base model, source corpus, predecessor work, and benchmark origins) appear under the HF Papers discovery surface. Tag IDs verified against arxiv.org titles.
Tags:
- arxiv:2408.10914 Aryabumi 2024 - To Code or Not to Code
- arxiv:2603.11510 Salamanca 2026 - Tiny Aya (base model)
- arxiv:2211.15533 Kocetkov 2022 - The Stack (source corpus)
- arxiv:2510.09591 Bazaz & Beg 2025 - Multilingual Python
- arxiv:1809.05053 Conneau 2018 - XNLI
- arxiv:2308.16884 Bandarkar 2024 - Belebele
- arxiv:2106.06937 Lin 2021 - X-CSQA
- arxiv:2210.03057 Shi 2022 - MGSM (Phase 2 legacy)
SIB-200 (arxiv:2309.07445) intentionally held for a follow-up.
Tags only — no other YAML or body changes.
## Summary
Adds 8 `arxiv:` tags to YAML frontmatter. Tags anchor the repo on the HF Papers discovery surface for each load-bearing paper in the submitted paper's bibliography.
## Tag selection rationale
Tier A — load-bearing methodological foundations:
- `arxiv:2408.10914` Aryabumi 2024 — direct precedent (English code improves reasoning); our work extends to multilingual code
- `arxiv:2603.11510` Salamanca 2026 — Tiny Aya, the base model
- `arxiv:2211.15533` Kocetkov 2022 — The Stack, the source corpus
- `arxiv:2510.09591` Bazaz & Beg 2025 — Multilingual Python, predecessor to Legesher
Tier B — Phase 3 benchmark anchors:
- `arxiv:1809.05053` Conneau 2018 — XNLI
- `arxiv:2308.16884` Bandarkar 2024 — Belebele
- `arxiv:2106.06937` Lin 2021 — X-CSQA
Tier B legacy — Phase 2:
- `arxiv:2210.03057` Shi 2022 — MGSM (used in Phase 2 only; retained because Phase 2 artifacts still live in this repo)
## What's NOT added
- SIB-200 (`arxiv:2309.07445`) — verified clean, deliberately held for a follow-up batch
- Background / methodology survey citations (mT5, MindMerger, Ghosh survey, etc.) — would dilute the discovery surface; remain in the paper's bibliography
- Software-only references (Wenyan, Qi, Latino, Legesher itself, Edgar & Buhai 2026) — no arxiv preprints exist
## Verification
All 8 IDs were fetched from arxiv.org and title-matched against the paper bibliography. No tag refers to a paper not cited in the work.
|
@@ -14,6 +14,14 @@ tags:
|
|
| 14 |
- transpilation
|
| 15 |
- tiny-aya-expedition
|
| 16 |
- language-decoded
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
pretty_name: Language Decoded Data
|
| 18 |
size_categories:
|
| 19 |
- 100K<n<1M
|
|
|
|
| 14 |
- transpilation
|
| 15 |
- tiny-aya-expedition
|
| 16 |
- language-decoded
|
| 17 |
+
- arxiv:2408.10914
|
| 18 |
+
- arxiv:2603.11510
|
| 19 |
+
- arxiv:2211.15533
|
| 20 |
+
- arxiv:2510.09591
|
| 21 |
+
- arxiv:1809.05053
|
| 22 |
+
- arxiv:2308.16884
|
| 23 |
+
- arxiv:2106.06937
|
| 24 |
+
- arxiv:2210.03057
|
| 25 |
pretty_name: Language Decoded Data
|
| 26 |
size_categories:
|
| 27 |
- 100K<n<1M
|