Dataset
Litmus-Bench v0.2 is an open dataset for training and evaluating chemical reasoning in language models. It includes 47,118 training questions and 2,056 test questions, each in short-answer format and was created from the ChEMBL, ExpansionRx, and Biogen datasets, with some questions utilizing RDKit descriptors. All questions require short answers The dataset is for RL training.
This dataset is released as part of NVIDIA NeMo Gym, an open-source library within the NVIDIA NeMo framework for evaluating and improving models and agents using environments. NeMo Gym provides infrastructure to develop environments, run reproducible agentic evaluations, and collect verifiable rollouts at scale for post-training, including Reinforcement Learning with Verifiable Rewards (RLVR). Developers can use these environments and rollouts with their preferred post-training framework and libraries, including NVIDIA NeMo RL, TRL, Unsloth, veRL, etc.
NeMo Gym serves as a unified hub for evaluation and training tasks, environments, verifiers, and reward signals for improving models and agents. Its environments provide the simulation “worlds” in which agents execute tasks, use tools, maintain state, and have their outcomes verified.
This dataset is part of the Hugging Face Org.
This dataset is ready for commercial or non-commercial uses.
NVIDIA Corporation
Sept 2, 2026
v0.2
Previous Version(s): Litmus Bench v0.1.
Governing terms: questions containing data from the ChEMBL database are licensed under CC BY SA 4.0. All other questions are licensed under OpenMDW 1.1.
To be used with NeMo-Gym for post-training LLMs.
Data Collection Method
Labeling Method
Text Only, Compatible with NeMo-Gym
| Metric | Value |
|---|---|
| Record Count: | 47,118 (train), 2,056 (test) |
| Feature Count: | 126 different properties and 12 distinct assay endpoints across the entire dataset. All questions are short answer. |
| Total Data Storage: | 64 MB |
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 model quality, risk, security vulnerabilities or NVIDIA AI Concerns here.
