datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
ntp-mathlib
miniCTX: Neural Theorem Proving with (Long-)Contexts
Lean 4 tactic prediction examples extracted from Mathlib.
These examples have not been formatted for instruction tuning (including data splits).
Please see l3lab/ntp-mathlib-instruct-* for datasets with instruction tuning examples.
Version
Generated using ntptoolkit's ntp-training-data.
It used the following config for ntp-training-data:
{
"repo": "https://github.com/leanprover-community/mathlib4",
"commit":… See the full description on the dataset page: https://huggingface.co/datasets/l3lab/ntp-mathlib.ntp-mathlib-instruct-context
miniCTX: Neural Theorem Proving with (Long-)Contexts
Lean 4 tactic prediction examples extracted from Mathlib.
Examples contain:
prompt:
instruction, preceding file content, proof state
instruction, proof state
completion: tactic
The file content has been truncated to 1024 tokens.
Version
Generated using ntptoolkit's ntp-training-data and instruction_tuning.py.
It used the following config for ntp-training-data:
{
"repo":… See the full description on the dataset page: https://huggingface.co/datasets/l3lab/ntp-mathlib-instruct-context.ntp-mathlib-instruct-st
miniCTX: Neural Theorem Proving with (Long-)Contexts
Lean 4 tactic prediction examples extracted from Mathlib.
Examples contain:
prompt:
instruction, proof state
completion: tactic
Version
Generated using ntptoolkit's ntp-training-data and instruction_tuning.py.
It used the following config for ntp-training-data:
{
"repo": "https://github.com/leanprover-community/mathlib4",
"commit": "cf8e23a62939ed7cc530fbb68e83539730f32f86",
"lean":… See the full description on the dataset page: https://huggingface.co/datasets/l3lab/ntp-mathlib-instruct-st.ntp-mathlib-instruct-context-fullproof
miniCTX: Neural Theorem Proving with (Long-)Contexts
Lean 4 full proof generation examples extracted from Mathlib.
Examples contain:
prompt:
instruction, preceding file content
completion: proof
The file content has been truncated to 1024 tokens.
Version
Generated using ntptoolkit's ntp-training-data and instruction_tuning.py.
It used the following config for ntp-training-data:
{
"repo": "https://github.com/leanprover-community/mathlib4",
"commit":… See the full description on the dataset page: https://huggingface.co/datasets/l3lab/ntp-mathlib-instruct-context-fullproof.ViBidLQA_v1
ViBidLQA: Vietnamese Bidding Legal Question Answering Dataset
Summary
ViBidLQA is a synthesized Vietnamese legal question-answering dataset built from the Vietnamese Bidding Law. It was created to address the scarcity of large-scale annotated datasets for legal AI in Vietnamese — a low-resource language setting. The dataset contains 3,013 QA pairs generated automatically using a large language model (Gemini) and verified by two domain experts. You may also like the new… See the full description on the dataset page: https://huggingface.co/datasets/ntphuc149/ViBidLQA_v1.ViBidLQA
📘 ViBidLQA — Vietnamese Bidding Law QA Dataset
ViBidLQA is a high-quality Vietnamese legal QA dataset specifically curated from the Vietnamese Bidding Law (No. 22/2023/QH15) and its Implementation Decree (No. 24/2024/NĐ-CP). This dataset was developed to support both extractive and abstractive QA tasks in the legal domain, and serves as a robust benchmark for evaluating Vietnamese Legal QA systems.
🔍 Motivation
Although existing datasets like ALQAC have been… See the full description on the dataset page: https://huggingface.co/datasets/ntphuc149/ViBidLQA.vit5-tst-data-sino-vietnamesedclm-fineweb-ntp-eval-3k
DCLM and FineWeb-Edu NTP Evaluation Retained 3K
This repository contains two 3,000-document subsets retained by the next-token
prediction (NTP) ordering analysis:
dclm: 3,000 unique documents selected from
dclm_new_eval50k.jsonl, originally sampled from
mlfoundations/dclm-baseline-1.0.
fineweb_edu: 3,000 unique documents selected from
fineweb_edu_new_eval50k.jsonl, originally sampled from
HuggingFaceFW/fineweb-edu (sample-10BT).
Load either configuration with:
from datasets… See the full description on the dataset page: https://huggingface.co/datasets/ardauzunoglu/dclm-fineweb-ntp-eval-3k.ViLegalMCQ
ViLegalMCQ
ViLegalMCQ is a Vietnamese legal Multiple Choice Question Answering (MCQ) dataset released alongside the ViLegalLM suite. It is synthetically generated from the ALQAC legal corpus using Qwen3-8B with human filtering, providing training data for context-based legal MCQ tasks.
Paper: ViLegalLM: Language Models for Vietnamese Legal Text — Read paper
Resources: GitHub | ViLegalBERT | ViLegalQwen2.5-1.5B-Base | ViLegalQwen3-1.7B-Base
Dataset Summary… See the full description on the dataset page: https://huggingface.co/datasets/ntphuc149/ViLegalMCQ.ViSpanExtractQA
📘 ViSpanExtractQA — Vietnamese Span-based QA Benchmark
ViSpanExtractQA is a consolidated Vietnamese QA dataset designed for span-based extractive question answering tasks. It aggregates and harmonizes multiple high-quality resources to create a diverse, multilingual, and robust benchmark for Vietnamese QA systems.
🔍 Motivation
While several QA datasets exist in Vietnamese, most are limited in size, scope, or consistency. ViSpanExtractQA aims to bridge this gap by… See the full description on the dataset page: https://huggingface.co/datasets/ntphuc149/ViSpanExtractQA.vit5-tst-data-coarseqwen3-4b-s70pct-ntpkd-only-ot3-rollouts-longqwen3-4b-s70pct-ntpkdopkd-ot3-rollouts-longntp_temples1ViLegalText
ViLegalTexts: A 16GB Vietnamese Legal Pre-training Corpus
This repository contains the pre-training corpus used to train ViLegalLM. The corpus was crawled from four publicly available Vietnamese legal repositories. For full details, please refer to the paper: Read paper
📋 Overview
Property
Value
Language
Vietnamese
Domain
Legal
Corpus size
16 GB
Format
.zip containing multiple .txt files
Sources
4 public Vietnamese legal repositories… See the full description on the dataset page: https://huggingface.co/datasets/ntphuc149/ViLegalText.bimnext_math_vi_1vit5-tst-data-formalqwen3-1.7b-s70pct-ntpkd-only-ot3-rollouts-longViLegalTF
ViLegalTF
ViLegalTF is a Vietnamese legal True/False Question Answering (TF) dataset released alongside the ViLegalLM suite. It is synthetically generated from the ALQAC legal corpus using Qwen3-8B with human filtering, providing training data for context-based legal true/false judgment tasks.
Paper: ViLegalLM: Language Models for Vietnamese Legal Text — Read paper
Resources: GitHub | ViLegalBERT | ViLegalQwen2.5-1.5B-Base | ViLegalQwen3-1.7B-Base
Dataset Summary… See the full description on the dataset page: https://huggingface.co/datasets/ntphuc149/ViLegalTF.vit5-tst-data-casualqwen3-1.7b-s70pct-ntpkdopkd-ot3-rollouts-long