Nemotron-RL-Multichallenge-v1

Nemotron-RL-Multichallenge-v1** License: cc-by-4.0 Task Categories: reinforcement-learning, text-generation Tags: instruction-following, multi-turn, rlvr, nemo-gym Configs: advanced train split at data/advanced.jsonl; vanilla train split at data/vanilla.jsonl Domain: instruction following, multi-turn conversation Modality: text Capability Breakdown: Multi-turn 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 difficult multi-turn instruction-following tasks. Reinforcement learning with verifiable rewards (RLVR) experiments using rubric-based verification in the VerifIF Gym environment. Training and evaluating inference memory, instruction retention, version editing, and self-coherence. Measuring model robustness when constraints appear across multiple turns and may require edits to earlier answers. Building NeMo Gym-compatible multi-turn conversation environments. Dataset Characterization Dataset Composition and Generation Problem Sources The dataset is manually collected and synthetically augmented, and uses no seed data. Tasks are designed as model-breaking multi-turn conversations in vanilla and advanced variants. Advanced records include one system message; vanilla records do not include a system message. Curation and Filtering The retained tasks are filtered using GPT-5 and human evaluation. Acceptance criteria require at most one of four generated model responses to pass all checks, and the response set must include meaningful pass/fail variance. Dataset Fields The Ultra-format JSONL files contain 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. `ins