Datasets:

juppy44
/

verifier-rm-data

schema_version string	example_id string	group_id string	source_dataset string	source_config string	source_row string	domain string	language string	custom_instruction string	current_completion string	text_to_score string	reference_solution string	privileged_context string	evidence_mode string	candidate_span string	target_label string	target_binary int64	supervision_type string	label_provenance string	label_confidence float64	pairwise_margin float64	metadata_json string
verifier-rm-1.0	098d3abe3afc31f29427	math:586fd646-76d6-5070-8c81-9993ab9d8559	open-r1/OpenR1-Math-220k	default	0:0	math	en	Reward mathematical correctness and self-contained, logically valid reasoning. Accept equivalent derivations and equivalent final answers. The privileged context is an authoritative hint; use it to check the candidate.	## Task B-1.3. A ship traveling along a river has covered $24 \mathrm{~km}$ upstream and $28 \mathrm{~km}$ downstream. For this journey, it took half an hour less than for traveling $30 \mathrm{~km}$ upstream and $21 \mathrm{~km}$ downstream, or half an hour more than for traveling $15 \mathrm{~km}$ upstream and $42 \...	<think> Okay, so I need to find the speed of the ship in still water and the speed of the river. Let me start by recalling that when a ship is moving upstream, its effective speed is the speed of the ship minus the speed of the river. Conversely, when moving downstream, its effective speed is the ship's speed plus the ...		Oracle final answer: v_{R}=4\mathrm{~}/\mathrm{},v_{B}=10\mathrm{~}/\mathrm{}	privileged	whole_completion	YES	1	absolute_verifiable	openr1_math_verify	1	null	{"answer": "v_{R}=4\\mathrm{~}/\\mathrm{},v_{B}=10\\mathrm{~}/\\mathrm{}", "generation_index": 0, "is_reasoning_complete": true, "problem_type": "Algebra", "question_type": "math-word-problem", "source": "olympiads", "uuid": "586fd646-76d6-5070-8c81-9993ab9d8559"}
verifier-rm-1.0	dba1dcfb15e366535acf	math:586fd646-76d6-5070-8c81-9993ab9d8559	open-r1/OpenR1-Math-220k	default	0:1	math	en	Reward mathematical correctness and self-contained, logically valid reasoning. Accept equivalent derivations and equivalent final answers. Judge from the problem and candidate only.	## Task B-1.3. A ship traveling along a river has covered $24 \mathrm{~km}$ upstream and $28 \mathrm{~km}$ downstream. For this journey, it took half an hour less than for traveling $30 \mathrm{~km}$ upstream and $21 \mathrm{~km}$ downstream, or half an hour more than for traveling $15 \mathrm{~km}$ upstream and $42 \...	<think> Okay, so I need to solve this problem about a ship traveling upstream and downstream. The question is asking for the speed of the ship in still water and the speed of the river. Let me think through this step by step. First, let me parse the problem again. The ship covers different distances upstream and downs...			none	whole_completion	NO	0	absolute_verifiable	openr1_math_verify	1	null	{"answer": "v_{R}=4\\mathrm{~}/\\mathrm{},v_{B}=10\\mathrm{~}/\\mathrm{}", "generation_index": 1, "is_reasoning_complete": true, "problem_type": "Algebra", "question_type": "math-word-problem", "source": "olympiads", "uuid": "586fd646-76d6-5070-8c81-9993ab9d8559"}
verifier-rm-1.0	156ac7cacade21bbd6aa	math:1cd82e3d-2e7a-5eca-9571-62f5669824c4	open-r1/OpenR1-Math-220k	default	2:0	math	en	Reward mathematical correctness and self-contained, logically valid reasoning. Accept equivalent derivations and equivalent final answers. The reference solution is authoritative evidence. The privileged context is an authoritative hint; use it to check the candidate.	Prove that number $1$ can be represented as a sum of a finite number $n$ of real numbers, less than $1,$ not necessarily distinct, which contain in their decimal representation only the digits $0$ and/or $7.$ Which is the least possible number $n$?	<think> Okay, so the problem is to prove that the number 1 can be represented as a sum of a finite number n of real numbers, each less than 1, where each number in the sum has a decimal representation consisting only of the digits 0 and/or 7. And then we need to find the least possible n. First, let me make sure I und...	1. Restate the problem in a more manageable form: We need to prove that the number $1$ can be represented as a sum of a finite number $n$ of real numbers, each less than $1$, and each containing only the digits $0$ and $7$ in their decimal representation. We also need to find the least possible value o...	Oracle final answer: 8	both	whole_completion	YES	1	absolute_verifiable	openr1_math_verify	1	null	{"answer": "8", "generation_index": 0, "is_reasoning_complete": true, "problem_type": "Number Theory", "question_type": "math-word-problem", "source": "aops_forum", "uuid": "1cd82e3d-2e7a-5eca-9571-62f5669824c4"}
verifier-rm-1.0	2bd556e23aa0ca3b1076	math:1cd82e3d-2e7a-5eca-9571-62f5669824c4	open-r1/OpenR1-Math-220k	default	2:1	math	en	Reward mathematical correctness and self-contained, logically valid reasoning. Accept equivalent derivations and equivalent final answers. The privileged context is an authoritative hint; use it to check the candidate.	Prove that number $1$ can be represented as a sum of a finite number $n$ of real numbers, less than $1,$ not necessarily distinct, which contain in their decimal representation only the digits $0$ and/or $7.$ Which is the least possible number $n$?	<think> Okay, so I need to prove that the number 1 can be represented as a sum of a finite number n of real numbers, each less than 1, and each having only digits 0 and/or 7 in their decimal representations. Then, I need to find the least possible n. Hmm, let's start by understanding the problem better. First, each of...		Oracle final answer: 8	privileged	whole_completion	NO	0	absolute_verifiable	openr1_math_verify	1	null	{"answer": "8", "generation_index": 1, "is_reasoning_complete": false, "problem_type": "Number Theory", "question_type": "math-word-problem", "source": "aops_forum", "uuid": "1cd82e3d-2e7a-5eca-9571-62f5669824c4"}
verifier-rm-1.0	874b5a915da88cd078c1	math:ea24dff6-9e4d-5c55-9ec0-c4d55424997a	open-r1/OpenR1-Math-220k	default	3:0	math	en	Reward mathematical correctness and self-contained, logically valid reasoning. Accept equivalent derivations and equivalent final answers. Judge from the problem and candidate only.	4. Given the three sides of an obtuse triangle are 3, $4, x$, then the range of values for $x$ is ( ). (A) $1<x<7$. (B) $5 \ll x<7$. (C) $1<x<\sqrt{7}$. (D) $5<x<7$ or $1<x<\sqrt{7}$.	<think> Okay, so I have this problem here where I need to find the range of values for x, given that the three sides of an obtuse triangle are 3, 4, and x. The answer choices are A through D. Let me think through this step by step. First, I remember that for any triangle, the sum of the lengths of any two sides must b...			none	whole_completion	YES	1	absolute_verifiable	openr1_math_verify	1	null	{"answer": "D", "generation_index": 0, "is_reasoning_complete": true, "problem_type": "Geometry", "question_type": "MCQ", "source": "cn_contest", "uuid": "ea24dff6-9e4d-5c55-9ec0-c4d55424997a"}
verifier-rm-1.0	b0b6fb767e0ac360ecbd	math:ea24dff6-9e4d-5c55-9ec0-c4d55424997a	open-r1/OpenR1-Math-220k	default	3:1	math	en	Reward mathematical correctness and self-contained, logically valid reasoning. Accept equivalent derivations and equivalent final answers. The reference solution is authoritative evidence. The privileged context is an authoritative hint; use it to check the candidate.	4. Given the three sides of an obtuse triangle are 3, $4, x$, then the range of values for $x$ is ( ). (A) $1<x<7$. (B) $5 \ll x<7$. (C) $1<x<\sqrt{7}$. (D) $5<x<7$ or $1<x<\sqrt{7}$.	<think> Okay, so I need to figure out the range of values for x such that the sides 3, 4, and x form an obtuse triangle. Hmm, let's start by recalling some basics about triangles and obtuse triangles specifically. First off, the triangle inequality theorem says that the sum of any two sides must be greater than the th...	$D$	Oracle final answer: D	both	whole_completion	YES	1	absolute_verifiable	openr1_math_verify	1	null	{"answer": "D", "generation_index": 1, "is_reasoning_complete": true, "problem_type": "Geometry", "question_type": "MCQ", "source": "cn_contest", "uuid": "ea24dff6-9e4d-5c55-9ec0-c4d55424997a"}
verifier-rm-1.0	69e7fb823e5cbdbf4175	math:a583d65c-9b4c-56ce-8263-47edbc22baa3	open-r1/OpenR1-Math-220k	default	4:0	math	en	Reward mathematical correctness and self-contained, logically valid reasoning. Accept equivalent derivations and equivalent final answers. Judge from the problem and candidate only.	1. Solve the equation: $\frac{8 x+13}{3}=\frac{19-12 x}{2}-\left(16 x-\frac{7-4 x}{6}\right)$.	<think> Okay, so I need to solve this equation: (8x + 13)/3 = (19 - 12x)/2 - (16x - (7 - 4x)/6). Hmm, that looks a bit complicated with all the fractions and parentheses. Let me take it step by step. First, maybe I should simplify both sides of the equation as much as possible by getting rid of the denominators. To do ...			none	whole_completion	YES	1	absolute_verifiable	openr1_math_verify	1	null	{"answer": "\\frac{1}{4}", "generation_index": 0, "is_reasoning_complete": true, "problem_type": "Algebra", "question_type": "math-word-problem", "source": "olympiads", "uuid": "a583d65c-9b4c-56ce-8263-47edbc22baa3"}
verifier-rm-1.0	d984edc971d159f7ac0d	math:a583d65c-9b4c-56ce-8263-47edbc22baa3	open-r1/OpenR1-Math-220k	default	4:1	math	en	Reward mathematical correctness and self-contained, logically valid reasoning. Accept equivalent derivations and equivalent final answers. Judge from the problem and candidate only.	1. Solve the equation: $\frac{8 x+13}{3}=\frac{19-12 x}{2}-\left(16 x-\frac{7-4 x}{6}\right)$.	<think> Okay, so I need to solve this equation: (8x + 13)/3 = (19 - 12x)/2 - (16x - (7 - 4x)/6). Hmm, that looks a bit complicated with all the fractions and parentheses. Let me take it step by step. First, maybe I should simplify both sides of the equation as much as possible. Let me start by looking at the right-han...			none	whole_completion	YES	1	absolute_verifiable	openr1_math_verify	1	null	{"answer": "\\frac{1}{4}", "generation_index": 1, "is_reasoning_complete": true, "problem_type": "Algebra", "question_type": "math-word-problem", "source": "olympiads", "uuid": "a583d65c-9b4c-56ce-8263-47edbc22baa3"}
verifier-rm-1.0	6f28e935735fe71b2a98	math:79e837be-2801-5a6a-bc7a-0acee3cdd5fd	open-r1/OpenR1-Math-220k	default	5:0	math	en	Reward mathematical correctness and self-contained, logically valid reasoning. Accept equivalent derivations and equivalent final answers. Judge from the problem and candidate only.	A right-angled triangle has side lengths that are integers. What could be the last digit of the area's measure, if the length of the hypotenuse is not divisible by 5?	<think> Okay, so I need to figure out the possible last digits of the area of a right-angled triangle with integer side lengths, where the hypotenuse isn't divisible by 5. Let me start by recalling some basics about Pythagorean triples and areas. First, the area of a right-angled triangle is (1/2)baseheight. Since t...			none	whole_completion	YES	1	absolute_verifiable	openr1_math_verify	1	null	{"answer": "0", "generation_index": 0, "is_reasoning_complete": true, "problem_type": "Geometry", "question_type": "math-word-problem", "source": "olympiads", "uuid": "79e837be-2801-5a6a-bc7a-0acee3cdd5fd"}
verifier-rm-1.0	4bfff77beeed17b65799	math:79e837be-2801-5a6a-bc7a-0acee3cdd5fd	open-r1/OpenR1-Math-220k	default	5:1	math	en	"Reward mathematical correctness and self-contained, logically valid reasoning. Accept equivalent de(...TRUNCATED)	"A right-angled triangle has side lengths that are integers. What could be the last digit of the are(...TRUNCATED)	"<think>\nOkay, so I need to figure out the possible last digits of the area of a right-angled trian(...TRUNCATED)	"Let the lengths of the legs be $a$ and $b$, and the length of the hypotenuse be $c$. According to t(...TRUNCATED)		reference	whole_completion	YES	1	absolute_verifiable	openr1_math_verify	1	null	"{\"answer\": \"0\", \"generation_index\": 1, \"is_reasoning_complete\": true, \"problem_type\": \"G(...TRUNCATED)

End of preview. Expand in Data Studio

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Instruction-conditioned verifier reward-model dataset

This folder contains a bounded, reproducible mixture for a Qwen3 4B LoRA reward model. Each row is a single candidate to score. The model is trained to emit YES for a candidate that satisfies the row's custom_instruction and NO otherwise; downstream users can use the Yes/No logit margin as the continuous reward.

The dataset is intended to remain private. The model weights can be released separately. The build script streams source datasets and stores source IDs, provenance, evidence regime, and supervision type in every row.

Build locally

The current environment already has datasets, huggingface_hub, and pyarrow. Otherwise:

python -m pip install -r requirements.txt
python build_dataset.py

The default build is intentionally bounded at roughly 54k candidate rows (about 160 MB of Parquet). Adjust quotas if desired:

python build_dataset.py --math-examples 20000 --shp-pairs 3000 --max-total 70000

Output files are written under data/ as Parquet shards. Build provenance and counts are written to reports/build_manifest.json and reports/dataset_stats.json.

Validate the generated shards and group-level split isolation:

python validate_dataset.py

Upload as a private Hub dataset

Create a user token with write access, then set it only in the current PowerShell process:

$env:HF_TOKEN = "hf_..."
python push_to_hub.py --repo-id YOUR_USERNAME/verifier-rm-data

The uploader creates the dataset repository as private by default and uploads only the data, README, and JSON reports. It never writes the token to disk.

Normalized schema

The core fields are:

custom_instruction: the criterion that defines what YES means.
current_completion: the user prompt or preceding conversation.
text_to_score: the candidate solution/response/trace.
reference_solution: optional authoritative solution.
privileged_context: optional authoritative hint for v1.
evidence_mode: none, reference, privileged, or both.
target_label / target_binary: YES/NO supervision target.
supervision_type: absolute_verifiable or pairwise_preference.
label_provenance and label_confidence: source and trust metadata.
group_id: keeps candidates from one underlying problem in the same split.

Splits are assigned by group_id, not by row, so paired candidates and math generations cannot leak across train/validation/test.

Included source families

open-r1/OpenR1-Math-220k: math reasoning traces with Math-Verify correctness labels. The builder creates several evidence views while retaining the verified label.
aladinDJ/ultramix-DPO-annotated: quality-filtered, task-balanced preference pairs with math, reasoning, coding, and general tasks.
nvidia/HelpSteer3 (preference): human-annotated multi-turn preferences across general, STEM, code, and multilingual domains.
coseal/CodeUltraFeedback_binarized: coding preference pairs with ratings and model provenance.
Skywork/Skywork-Reward-Preference-80K-v0.2: general reward-preference pairs.
R-I-S-E/RISE-Judge-SFT-20K: position-verified judge comparisons derived from math and preference sources, normalized into candidate-level YES/NO rows.
iknow-lab/JudgeBias-DPO-RefFree (validation_1000.parquet only): a bounded anti-shortcut/materials-science slice designed to preserve semantic equivalence while detecting injected errors. The full upstream repository has inconsistent parquet schemas, so the builder reads only this stable small file directly.
Anthropic/hh-rlhf: multi-turn helpfulness/harmlessness comparisons with a transcript-aware context split.
stanfordnlp/SHP: naturally occurring Reddit preferences spanning 18 question domains; the builder keeps stronger score-ratio pairs and records the subreddit. Treat this as private research data because the upstream card describes Reddit terms rather than a simple redistribution license.
AIPlans/PKU-SafeRLHF-RLHF: filtered safety-aware comparisons with a confidence margin (CC-BY-NC-4.0 inherited from PKU SafeRLHF).
SUSTech-NLP/MixReward: Apache-2.0 multilingual preference pairs across 103 languages and several domains.
project-themis/Themis-CodePreference: Apache-2.0 code preferences covering functional correctness, runtime/memory efficiency, security, and maintainability across eight languages.

Preference pairs remain marked as pairwise rather than being presented as absolute truth. That distinction matters because a chosen response can be better than a rejected response while still being imperfect.

Deliberate exclusions

PRM800K is not used as the main math source because this first release emphasizes richer full-solution and preference variation. Evaluation-only sets and sources with inconsistent Hub schemas are not forced into the build; the manifest records any source that fails to stream.

Check each upstream dataset's current license and terms before redistributing derived data. This local/private build does not grant redistribution rights.

Downloads last month: 34