Nemotron-RL-SysBench-v1

Nemotron-RL-SysBench-v1** License: cc-by-4.0 Language: en Task Categories: reinforcement-learning, text-generation Tags: instruction-following, system-message-following, multi-turn, rlvr, nemo-gym Configs: default train split at data/train.jsonl Domain: instruction following, system message following Modality: text Capability Breakdown: System message following [100%] Source: Hybrid: Manually Collected, Synthetic Size Bin: 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 system-message-following behavior. Reinforcement learning with verifiable rewards (RLVR) experiments where rewards measure adherence to persistent system-level constraints. Training and evaluating robustness when user requests may conflict with system instructions. Studying multi-turn stability of system-message compliance. Building NeMo Gym-compatible system-message-following environments. Dataset Characterization Dataset Composition and Generation Problem Sources The dataset is manually collected and synthetically augmented. Tasks are instruction-following problems in which complex instructions are placed in the system prompt. 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, when present. Items include fields such as `uid`, `source`, `instruction_id`, `is_misalignment_check`, and task-specific constraint parameters. `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 hi