Nemotron-RL-InverseIFEval-v1

Nemotron-RL-InverseIFEval-v1** License: cc-by-4.0 Task Categories: reinforcement-learning, text-generation Tags: instruction-following, adversarial, rlvr, nemo-gym Configs: default train split at data/train.jsonl Domain: instruction following, adversarial instruction following Modality: text Capability Breakdown: Counter-conventional instruction following [100%] Source: Hybrid: Manually Collected, Synthetic Size Bin: Previous Version(s): Previous Super-format submission. 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 adversarial instruction-following prompts. Reinforcement learning with verifiable rewards (RLVR) experiments where rewards measure strict compliance with unusual or counter-conventional instructions. Improving robustness to default coding, formatting, helpfulness, or correction behaviors that conflict with a user request. Evaluating GPT-5/human-validated rubrics for difficult instruction-following cases. Building NeMo Gym-compatible adversarial instruction-following environments. Dataset Characterization Dataset Composition and Generation Problem Sources The dataset is manually collected and synthetically augmented, and uses no seed data. Examples are adversarial prompts corresponding to anti-convention instruction-following patterns. Curation and Filtering The retained tasks are filtered with GPT-5 and blind human evaluation. Acceptance requires at most one passing response out of four candidate model responses, at least one pass and one fail across candidates, and at least 85% human/GPT-5 agreement. 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`: Instruction metadata array. This array is empty in the Inverse