PhoLaya-base — Vietnamese typed-decision mixture

Typed-decision training data for PhoLaya, a Vietnamese choice / score / noul model fine-tuned from convaiinnovations/laya-multilingual (mmBERT-base + Laya decision head).

Not affiliated with VinAI (PhoBERT) or Convai Innovations (Laya).

Each row is one question about a state, in the Laya schema:

Column	Contents
state	JSON string; Vietnamese keys (binh_luan, doan_van, cau_noi, tieu_de, ...)
questions	JSON string {qid: {type, instructions, criteria}}; ASCII snake_case option keys, Vietnamese descriptions
gold	JSON string {qid: {label, probabilities}}

type is choice / score / noul. probabilities is one-hot for recast sources (soft targets are a v2 item). The corpus is Vietnamese-only — there is no English replay stream.

Sources

Source	Unique questions	License	Recast as
ViGLUE (tmnam20/ViGLUE)	1,010,702	see source	9 translated GLUE tasks (sentiment / acceptable / paraphrase / entailment / NLI / QA)
UIT-ViSFD (visolex/ViSFD)	122,342	research (UIT)	per-aspect 4-way choice + complaint noul
ViNLI (presencesw/vinli_4_label)	60,726	research (UIT)	relation choice(4) + entailment noul
VSFC (uitnlp/vietnamese_students_feedback)	32,286	research (UIT)	sentiment choice(3) + topic choice(4)
MASSIVE vi-VN (AmazonScience/massive v1.1)	31,955	CC BY 4.0	scenario choice(18) + intent choice(60)
ViANLI (uitnlp/ViANLI)	20,022	research (UIT)	relation choice(3) + entailment noul
UIT-ViCTSD (visolex/ViCTSD)	19,991	research (UIT)	constructive choice(2) + toxic choice(2)
UIT-VSMEC (tridm/UIT-VSMEC)	6,922	research (UIT)	emotion choice(7)
UTS2017_Bank (undertheseanlp/UTS2017_Bank)	4,604	research	overall + per-aspect sentiment choice(3)

Counts are unique questions after dedup (sum over splits). Total ~1.31M unique questions across the three splits.

Deduplication

Rows are deduplicated by a canonical key over (normalized state, question type, criteria, instruction, gold label):

• the state is diacritic-stripped and lowercased, and criteria are compared as an unordered option set — so diacritic and option-order differences never produce two rows;
• the instruction is part of the key, because it carries the question's identity (e.g. ViSFD's per-aspect wording) — so distinct questions are not merged;
• dedup runs within each source, then again across the concatenated mixture, and any dev/test question whose key appears in train is dropped (leakage/decontamination).

The mixture is not materialized with augmentation. There are no paraphrase, option-shuffle, or diacritic-stripped duplicate rows on disk; diacritic stripping, option permutation, and instruction sampling belong at TypedDecisionDataset load time.

Splits

Split	Unique cases	Size
train	1,166,001	744 MB
validation	92,446	64 MB
test	49,580	43 MB

Train composition by source: ViGLUE 947,553; ViSFD 85,646; ViNLI 48,726; VSFC 22,790; MASSIVE 22,069; ViANLI 16,024; ViCTSD 13,991; VSMEC 5,544; UTS 3,658.

Intended use

Training and evaluating Vietnamese typed-decision models and studying calibration / token-budget ablations. Not for production decisions.

Limitations

• Recast labels are one-hot; score-type data is deferred to v2.
• ViGLUE tasks are machine translations of GLUE; labels inherit GLUE's annotation noise and translation artifacts.
• ViGLUE's mnli/qqp have no test split (train/validation only).
• MASSIVE intent descriptions are composed from label keys; a native translation pass is pending. MASSIVE ships the full intent set (K=60).
• UTS2017_Bank has no dev split.
• Dedup is exact on the canonical key; semantically equivalent questions with different wording across sources are kept separate.

License & attribution

license: other (research-use-only): the mixture contains derivatives of the UIT corpora (ViSFD, VSFC, ViNLI, ViANLI, ViCTSD, UIT-VSMEC) and UTS2017_Bank, which are research-licensed. Do not use commercially without contacting the dataset authors. MASSIVE vi-VN is CC BY 4.0; ViGLUE is a translation of GLUE — see its source card. Attribute each source per its own license.

Reproduce with scripts/data/download.py + scripts/data/prepare.py; raw and processed data (including the canonical per-source shards under sources/) are regenerated from scripts and never committed to the code repo.