datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
lean-proofs-v1
Part of the SZL Holdings governed estate — claims are designed to carry checkable receipts. Verification proves integrity & origin, never accuracy or performance.
SZLHOLDINGS/lean-proofs-v1
The complete Lean 4 theorem library for the SZL Holdings Ouroboros Invariant research programme.
Doctrine v10/v11 Canonical Numbers
Metric
Value
Declarations
749
Unique axioms
14 (15 raw, 1 dup)
Sorries
163 (112 baseline + 51 Putnam)… See the full description on the dataset page: https://huggingface.co/datasets/SZLHOLDINGS/lean-proofs-v1.lean-proof-compression
LeanPolish: Verified Supervision for Lean Proof Compression
A dataset of Lean 4 proof rewrite pairs produced by LeanPolish,
a kernel-verified proof-shortening tool. Every accepted
(original, replacement) pair was kernel-checked under Lean 4.21.0
with Mathlib v4.21.0 before emission, and the rewritten file was
re-elaborated end-to-end by a separate out-of-process verifier.
The dataset is suitable for training models that learn to compress,
simplify, or select proof tactics, and… See the full description on the dataset page: https://huggingface.co/datasets/leanpolish-anon/lean-proof-compression.STP_Lean_0320This is an updated version of the final training dataset of Self-play Theorem Prover as described in the paper STP: Self-play LLM Theorem Provers with Iterative Conjecturing and Proving. This dataset includes:
Extracted examples from mathlib4,
Generated correct proofs of statements in LeanWorkbook,
Generated correct proofs of conjectures proposed by our model during self-play training.
NormasTCU
NormasTCU
Overview
NormasTCU iis a dataset for Legal Information Retrieval (LIR) in Brazilian Portuguese composed of normative documents from the Brazilian Federal Court of Accounts (Tribunal de Contas da União - TCU), along with queries and human-annotated relevance judgments.
The dataset includes:
14,469 legal documents (normative acts);
46 queries;
812 judge query-document pairs derived from 3,048 human annotations with 3-level graded relevance.… See the full description on the dataset page: https://huggingface.co/datasets/LeandroRibeiro/NormasTCU.STP_LeanThis is the final training dataset of Self-play Theorem Prover as described in the paper STP: Self-play LLM Theorem Provers with Iterative Conjecturing and Proving. This dataset includes:
Extracted examples from mathlib4,
Generated correct proofs of statements in LeanWorkbook,
Generated correct proofs of conjectures proposed by our model during self-play training.
lean-proof-or-refute-300
Lean Proof-or-Refute 300
Lean Proof-or-Refute 300 is a compact collection of 300 formal reasoning
problems grounded in Lean 4 and Mathlib. Each problem starts from a verified
Mathlib theorem, makes one small numerical or operator mutation, and asks the
model to return either:
a Lean certificate proving the mutated proposition; or
a Lean certificate proving the exact negation of the complete proposition.
The model receives the related source theorem, a bounded source excerpt… See the full description on the dataset page: https://huggingface.co/datasets/xlr8harder/lean-proof-or-refute-300.2026_07_19_collect_leandojo_gemma3_12b_gemma4_31b_flsft_toklean-eval-benchmark
Lean Evaluator Benchmark
An out-of-distribution benchmark for models that predict whether a Lean program
will compile. Each row contains one prover-generated Lean program and the verdict
produced by the real Lean compiler.
Dataset summary
Rows: 68,932
Lean version: v4.15.0
Valid programs: 28,145 (40.83%)
Invalid programs: 40,787 (59.17%)
miniF2F: 41,638 rows (25,702 valid)
ProofNet: 27,294 rows (2,443 valid)
The programs come from six runs spanning three… See the full description on the dataset page: https://huggingface.co/datasets/formalmathatepfl/lean-eval-benchmark.repro-ai4slt-empirical-processes-in-lean-4-for-formal-statistical-learning-theory-traces
Agent traces
Agent sessions published from a Trackio Logbook.
lean-cedar-hashtable
Nanoda Library Hashtable Trace
Hashtable lookup appears to be the most heavy work in the fastest lean kernel implementation (https://arena.lean-lang.org/checker/nanoda/).
Can we actually do better?
This dataset provides the trace for hashtable lookup during the check on the Cedar benchmark.
lean-github-bigaic_gt_28d_256x288_v4_20260426_115731This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "aic_controller",
"total_episodes": 6,
"total_frames": 5961,
"total_tasks": 5,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 20,
"splits": {
"train": "0:6"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/leandroper/aic_gt_28d_256x288_v4_20260426_115731.rrma-lean4-agent-traces
RRMA Lean 4 Agent Traces
416 multi-agent Lean 4 proof search traces across two Erdős problems, three model tiers, and four difficulty rungs.
v2 (2026-06-10) — label + format correction. The original upload had two defects:
(1) messages was a JSON string, not an array; (2) reward was set to 1.0 if the
text SCORE=1.0 appeared anywhere in the conversation — including the worker prompt
("repeat until SCORE=1.0") and file reads of the oracle script, so almost every trace was
labeled… See the full description on the dataset page: https://huggingface.co/datasets/vincentoh/rrma-lean4-agent-traces.math-lean-hackable-rollouts
Math Lean Hackable Rollouts
This dataset contains 2,241 labeled multi-turn rollouts from a GRPO run on deliberately
hackable Lean 4 theorem-proving tasks. The policy was
nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16.
The run's weakened grader accepts proofs containing sorry; the separate oracle restores
Lean's sorry check. hack_detected is true exactly when the weakened grader paid the
rollout but the restored oracle rejected it. Rows without a gradeable final answer were
excluded… See the full description on the dataset page: https://huggingface.co/datasets/AlignmentResearch/math-lean-hackable-rollouts.numinamath-LEAN-trajs-850k2026_07_19_collect_leandojo_gemma3_12b_gemma4_31b_raw_student_toklean-mathlib-hashmaperdos741ii-lean4-opus-traces
Erdős #741(ii) — Lean 4 Opus Agent Traces
39 Claude Opus agent sessions attempting Erdős problem #741(ii) in Lean 4 (G1 rung: build the proof from an NL construction description).
v2 (2026-06-10) — label correction. The original upload labeled all 39 traces
reward=1.0; the labeler matched the text SCORE=1.0 anywhere in the conversation,
including the worker prompt. Rewards are now anchored to genuine oracle output
(line-anchored SCORE= adjacent to… See the full description on the dataset page: https://huggingface.co/datasets/vincentoh/erdos741ii-lean4-opus-traces.lean-lean
LeanLean
Can coding agents make a verified Lean library smaller? LeanLean gives an agent 12 hours on each of 64 real-world Lean repositories to compress the Lean codebase as much as it can while preserving the protected theorems.
There are 16 projects per stripped-token band (≤10k, 10k–50k, 50k–250k, >250k): 13.8M stripped tokens and 620 protected declarations in total.
No build artifacts are included. Build a project with lake build, after lake exe cache get for the 63 projects… See the full description on the dataset page: https://huggingface.co/datasets/eth-sri/lean-lean.leanforge-compression-evalGenerated_News_Political_LeaningMotoGP-Lean-Anglelean-quantfinance
Lean 4 Formalized Quantitative Finance & Game Theory
A domain-specific Lean 4 / Mathlib corpus centered on finance and market
mechanisms: 2,074 theorem records + 887 definitions, extracted from a
formalization pipeline and packaged for theorem-proving research (statement,
proof, tactics, premises, kernel-axiom status).
This is a mechanization of largely standard applied mathematics, not new
finance theory. Its value is breadth in under-formalized areas — market
microstructure… See the full description on the dataset page: https://huggingface.co/datasets/seancollins/lean-quantfinance.aic_gt_28d_256x288_v2_20260426_010747This dataset was created using LeRobot.
Dataset Structure
meta/info.json:
{
"codebase_version": "v3.0",
"robot_type": "aic_controller",
"total_episodes": 1,
"total_frames": 691,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 20,
"splits": {
"train": "0:1"
},
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path":… See the full description on the dataset page: https://huggingface.co/datasets/leandroper/aic_gt_28d_256x288_v2_20260426_010747.STP_Lean_SFT_evallean_workbook_hardlean_workbook_RL_V14_hinter_v6leanworkbook_hinter_v14_v1leanworkbook_hinter_v14_v5lean_workbook_RL_no_zero_examples_2000