Datasets:

nvidia
/

Nemotron-RL-CFBench-v1

license stringclasses 1 value	id int64 65.6k 71.8k	agent_ref dict	instructions listlengths 3 16	llm_judge listlengths 1 7	responses_create_params dict	messages listlengths 2 20	tools listlengths 0 0	uuid stringlengths 20 20	used_in listlengths 1 1
CC BY 4.0	69,302	{ "type": "responses_api_agents", "name": "turing_vif_simple_agent" }	[ { "instruction_id": "startend:end_checker", "source": "user", "is_misalignment_check": false, "end_phrase": "注意を要する。", "uid": 2 }, { "instruction_id": "detectable_format:number_bullet_lists", "source": "user", "is_misalignment_check": false, "relation": "equal to", "num_b...	[ { "uid": 1, "content": "Are these analysis results listed in order of severity?", "source": "user", "is_misalignment_check": false } ]	{ "input": [ { "role": "system", "content": "あなたは、政策・法律分野におけるデータ分析を専門とする「政策データ分析スペシャリスト」です。スポーツ・レクリエーション活動に関するデータを、政策的・法的観点から分析・評価し、実用的な洞察を提供することがあなたの役割です。成功とは、与えられたデータと文脈から、法的リスク、政策的含意、改善提案を明確に示す分析結果を提供することです。\n \n 条件分岐に基づく処理ルール\n * ユーザータイプ: 一般市民、事業者、行政職員、研究者などが想定されます。説明の詳細さと専門用語の使用は、想定されるユ...	[ { "role": "system", "content": "あなたは、政策・法律分野におけるデータ分析を専門とする「政策データ分析スペシャリスト」です。スポーツ・レクリエーション活動に関するデータを、政策的・法的観点から分析・評価し、実用的な洞察を提供することがあなたの役割です。成功とは、与えられたデータと文脈から、法的リスク、政策的含意、改善提案を明確に示す分析結果を提供することです。\n \n 条件分岐に基づく処理ルール\n * ユーザータイプ: 一般市民、事業者、行政職員、研究者などが想定されます。説明の詳細さと専門用語の使用は、想定されるユーザーに応じて調整してください（例: ...	[]	cfbench-69302-000001	[ "ultra_v3" ]
CC BY 4.0	70,974	{ "type": "responses_api_agents", "name": "turing_vif_simple_agent" }	[ { "instruction_id": "detectable_format:table", "source": "user", "is_misalignment_check": false, "min_rows": 5, "min_cols": 5, "uid": 2 }, { "instruction_id": "change_case:all_caps_target", "source": "user", "is_misalignment_check": false, "target_string": "terreno", ...	[ { "uid": 1, "content": "Include a VBA code to create a Gantt chart.", "source": "user", "is_misalignment_check": false } ]	{ "input": [ { "role": "system", "content": "" }, { "role": "user", "content": "Hola, necesito una línea de tiempo básica para la expansión de las granjas solares en las regiones A, B y C. Tengo 5 equipos (cada uno con 4 ingenieros y 3 técnicos) y quiero empezar en enero. La región...	[ { "role": "system", "content": "" }, { "role": "user", "content": "Hola, necesito una línea de tiempo básica para la expansión de las granjas solares en las regiones A, B y C. Tengo 5 equipos (cada uno con 4 ingenieros y 3 técnicos) y quiero empezar en enero. La región B no puede operar de novie...	[]	cfbench-70974-000002	[ "ultra_v3" ]
CC BY 4.0	69,402	{ "type": "responses_api_agents", "name": "turing_vif_simple_agent" }	[ { "instruction_id": "keywords:forbidden_words", "source": "user", "is_misalignment_check": false, "forbidden_words": [ "very", "significant", "important", "major", "substantial", "profound", "notable", "pronounced", "marked", "substantive",...	[ { "uid": 1, "content": "Are all (10) json objects in a JSON list named 'studies' with each object having seven keys and the relevance score being a number between 1-5?", "source": "user", "is_misalignment_check": false } ]	{ "input": [ { "role": "system", "content": "" }, { "role": "user", "content": "Hello, I am a doctoral candidate in Marine Ecology at the Scripps Institution of Oceanography. I’m currently in the data compilation phase of my dissertation, which is a systematic literature review on ...	[ { "role": "system", "content": "" }, { "role": "user", "content": "Hello, I am a doctoral candidate in Marine Ecology at the Scripps Institution of Oceanography. I’m currently in the data compilation phase of my dissertation, which is a systematic literature review on the impacts of climate chan...	[]	cfbench-69402-000003	[ "ultra_v3" ]
CC BY 4.0	69,599	{ "type": "responses_api_agents", "name": "turing_vif_simple_agent" }	[ { "instruction_id": "stylistic:tone_formality", "source": "system", "is_misalignment_check": false, "tone_level": "formal", "uid": 2 }, { "instruction_id": "stylistic:voice", "source": "user", "is_misalignment_check": false, "voice_type": "passive", "uid": 3 }, { ...	[ { "uid": 1, "content": "Is the table coherently integrated into the context?", "source": "user", "is_misalignment_check": false } ]	{ "input": [ { "role": "system", "content": "Instrucciones del Sistema para el Asistente: Servicio al Cliente y Análisis de Datos Financieros\n \n Rol: Eres un representante especializado en servicio al cliente y análisis de datos para una institución financiera. Tu función es asistir a los cl...	[ { "role": "system", "content": "Instrucciones del Sistema para el Asistente: Servicio al Cliente y Análisis de Datos Financieros\n \n Rol: Eres un representante especializado en servicio al cliente y análisis de datos para una institución financiera. Tu función es asistir a los clientes con consulta...	[]	cfbench-69599-000004	[ "ultra_v3" ]
CC BY 4.0	68,755	{ "type": "responses_api_agents", "name": "turing_vif_simple_agent" }	[ { "instruction_id": "keywords:existence", "source": "user", "is_misalignment_check": false, "keywords": [ "data", "targa", "violazione" ], "uid": 2 }, { "instruction_id": "keywords:frequency", "source": "user", "is_misalignment_check": false, "keyword": ...	[ { "uid": 1, "content": "Does it confirm that the ‘tipo di violazione’ is correctly normalized, anomalies are explicitly flagged, and a clear final summary focuses on the accuracy of the violazione?", "source": "user", "is_misalignment_check": false } ]	{ "input": [ { "role": "system", "content": "Istruzione di Sistema per Assistente AI\n \n Sei un Analista di Dati Normativo, specializzato nell'interpretazione e nell'applicazione di linee guida e normative. Il tuo obiettivo è **estrarre, normalizzare e strutturare in modo coerente informazion...	[ { "role": "system", "content": "Istruzione di Sistema per Assistente AI\n \n Sei un Analista di Dati Normativo, specializzato nell'interpretazione e nell'applicazione di linee guida e normative. Il tuo obiettivo è **estrarre, normalizzare e strutturare in modo coerente informazioni precise da input ...	[]	cfbench-68755-000005	[ "ultra_v3" ]
CC BY 4.0	70,510	{ "type": "responses_api_agents", "name": "turing_vif_simple_agent" }	[ { "instruction_id": "stylistic:rhythm_pattern", "source": "user", "is_misalignment_check": false, "rhythm_type": "short", "uid": 2 }, { "instruction_id": "keywords:frequency", "source": "user", "is_misalignment_check": false, "keyword": "प्रक्रिया ", "relation": "equal to...	[ { "uid": 1, "content": "Does the response clearly evaluate the CCTV plan’s budget constraints, process gaps, accountability issues, and real impact versus cost?", "source": "user", "is_misalignment_check": false } ]	{ "input": [ { "role": "system", "content": "" }, { "role": "user", "content": "मोहल्ले में नई सीसीटीवी कैमरा लगाने की municipal corporation की योजना का विश्लेषण करो। डेटा है: बजट 20 लाख, 10 कैमरे, अपराध दर वाले 5 हॉटस्पॉट। योजना का उद्देश्य अपराध रोकना और नागरिकों को आश्वस्त करना ...	[ { "role": "system", "content": "" }, { "role": "user", "content": "मोहल्ले में नई सीसीटीवी कैमरा लगाने की municipal corporation की योजना का विश्लेषण करो। डेटा है: बजट 20 लाख, 10 कैमरे, अपराध दर वाले 5 हॉटस्पॉट। योजना का उद्देश्य अपराध रोकना और नागरिकों को आश्वस्त करना है। बताओ क्या यह योजना प्रभ...	[]	cfbench-70510-000006	[ "ultra_v3" ]
CC BY 4.0	68,731	{ "type": "responses_api_agents", "name": "turing_vif_simple_agent" }	[{"instruction_id":"detectable_format:sentence_count","source":"user","is_misalignment_check":false,(...TRUNCATED)	[{"uid":1,"content":"Does the answer include a table with 4 rows and 4 columns that lists each oppon(...TRUNCATED)	{"input":[{"role":"system","content":"Rôle: Vous êtes un analyste de données spécialisé dan(...TRUNCATED)	[{"role":"system","content":"Rôle: Vous êtes un analyste de données spécialisé dans les spo(...TRUNCATED)	[]	cfbench-68731-000007	[ "ultra_v3" ]
CC BY 4.0	69,762	{ "type": "responses_api_agents", "name": "turing_vif_simple_agent" }	[{"instruction_id":"punctuation:no_period","source":"user","is_misalignment_check":false,"uid":2},{"(...TRUNCATED)	[{"uid":1,"content":"Did the model comprehend the underlying purpose of creating comparable, action-(...TRUNCATED)	{"input":[{"role":"system","content":"Rolle: Der Assistent agiert als ein hybrides Werkzeug, das die(...TRUNCATED)	[{"role":"system","content":"Rolle: Der Assistent agiert als ein hybrides Werkzeug, das die Funktion(...TRUNCATED)	[]	cfbench-69762-000008	[ "ultra_v3" ]
CC BY 4.0	69,881	{ "type": "responses_api_agents", "name": "turing_vif_simple_agent" }	[{"instruction_id":"punctuation:no_comma","source":"user","is_misalignment_check":false,"uid":2},{"i(...TRUNCATED)	[{"uid":1,"content":"Is the word 'ensino' repeated without being used in the same syntactic structur(...TRUNCATED)	{"input":[{"role":"system","content":"Papel: Você é um Assistente de Pesquisa Acadêmica espec(...TRUNCATED)	[{"role":"system","content":"Papel: Você é um Assistente de Pesquisa Acadêmica especializado (...TRUNCATED)	[]	cfbench-69881-000009	[ "ultra_v3" ]
CC BY 4.0	69,472	{ "type": "responses_api_agents", "name": "turing_vif_simple_agent" }	[{"instruction_id":"detectable_format:sentence_count","source":"user","is_misalignment_check":false,(...TRUNCATED)	[{"uid":1,"content":"Is the Installed Solar Capacity given in GW?","source":"user","is_misalignment_(...TRUNCATED)	{"input":[{"role":"system","content":""},{"role":"user","content":"I am working as a senior research(...TRUNCATED)	[{"role":"system","content":""},{"role":"user","content":"I am working as a senior research analyst (...TRUNCATED)	[]	cfbench-69472-000010	[ "ultra_v3" ]

End of preview. Expand in Data Studio

Nemotron-RL-CFBench-v1

License: cc-by-4.0
Language: en, ar, hi, zh, ja, ko
Task Categories: reinforcement-learning, text-generation
Tags: instruction-following, constraint-following, rlvr, nemo-gym
Configs: default train split at data/train.jsonl
Domain: instruction following, constraint following
Modality: text
Capability Breakdown: Constraint following [100%]
Source: Hybrid: Manually Collected, Synthetic
Size Bin: <10K
Associated Model Release: Nemotron Ultra

Dataset Description:

Nemotron-RL-CFBench-v1 is an RL dataset for instruction-following problems where the focus is on whether an LLM can satisfy explicit constraints. The dataset is manually collected and synthetically augmented, and formatted for the VerifIF Gym environment.

The seed data comes from manually collected instruction-following sources. NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 and Qwen/Qwen3-235B-A22B-Thinking-2507 are used as SDG models, and GPT-5 is used for filtering.

The dataset uses the VerifIF Gym schema with agent_ref, id, instructions, llm_judge, and responses_create_params. Each record contains one system message and at least one user message in responses_create_params.input; many records also include prior assistant messages. The file also contains structured instruction metadata and judge checks that describe the constraints to verify.

This dataset is ready for commercial or non-commercial uses.

Dataset Owner(s):

NVIDIA Corporation

Dataset Creation Date:

Created on: 04/28/2026 Last Modified on: 05/21/2026

Version:

Nemotron-RL-CFBench-v1

Previous Version(s): N/A

License/Terms of Use:

This dataset is licensed under Creative Commons Attribution 4.0 International (CC BY 4.0).

Intended Usage:

This dataset is intended for:

Reinforcement learning of LLMs on complex constraint-following prompts.
Reinforcement learning with verifiable rewards (RLVR) experiments where rewards measure satisfaction of granular instruction criteria.
Training and evaluating robustness to multiple simultaneous user constraints.
Studying model behavior on prompts with formatting, keyword, scenario, and other constraint types.
Building NeMo Gym-compatible constraint-following environments.

Dataset Characterization

Dataset Composition and Generation

Problem Sources

The dataset is manually collected and synthetically augmented. Tasks are instruction-following problems focused on constraint satisfaction.

Curation and Filtering

RL problems are curated and filtered with GPT-5.

Dataset Fields

The Ultra-format JSONL file contains the following top-level fields:

agent_ref: Agent metadata for the VerifIF Gym environment. Records use responses_api_agents/verifif_simple_agent.
id: Numeric example identifier.
instructions: Structured instruction metadata. Items include fields such as uid, source, instruction_id, is_misalignment_check, and task-specific constraint parameters such as keywords.
llm_judge: Judge checks. Items include uid, source, content, and is_misalignment_check.
responses_create_params: Responses API-style input payload containing system/user messages with optional assistant history.

Data Collection Method

Hybrid: Manually Collected, Synthetic

Labeling Method

Hybrid: Manually-Labelled, Automated. GPT-5 is used for filtering.

Dataset Format

Language: English (en), Arabic (ar), Hindi (hi), Chinese (zh), Japanese (ja), Korean (ko) Modality: Text Format: JSONL Structure: VerifIF Gym records with agent metadata, Responses API-style system/user messages with optional assistant history, structured instruction metadata, and LLM-judge checks.

Dataset Quantification

Subset	Samples	File Size	Notes
train	1,121	25MB	Input length ranges from 2 to 20 messages; instruction checks range from 3 to 16 per record; LLM-judge checks range from 1 to 7 per record

Reference(s):

N/A

Ethical Considerations:

NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. Developers should work with their internal developer teams to ensure this dataset meets requirements for the relevant industry and use case and addresses unforeseen product misuse. Please report quality, risk, security vulnerabilities or NVIDIA AI Concerns here.

Downloads last month: 27

Collections including nvidia/Nemotron-RL-CFBench-v1