CoolFace
Datasetpublic

Siye01/LLM-Fusion-Train

Multi-Domain RLVR Training Data Per-domain reinforcement-learning-with-verifiable-rewards (RLVR) training sets for five domains, plus the mixed-domain blend used as a joint-training baseline. Every subset uses the verl RLHF parquet schema: data_source, prompt, ability, reward_model, extra_info. Subsets Subset Rows Size Math 38,131 11.6 MiB Science 50,000 45.6 MiB Code 19,169 1,432.5 MiB IF 16,575 9.3 MiB Agent 10,229 0.2 MiB Mix 87,699 1,505.1… See the full description on the dataset page: https://huggingface.co/datasets/Siye01/LLM-Fusion-Train.

sourceHugging Faceapache-2.0updated 27d agoView on Hugging Face
0likes146downloads
Dataset Card

Multi-Domain RLVR Training Data

Per-domain reinforcement-learning-with-verifiable-rewards (RLVR) training sets for five domains, plus the mixed-domain blend used as a joint-training baseline. Every subset uses the verl RLHF parquet schema: data_source, prompt, ability, reward_model, extra_info.

Subsets

SubsetRowsSize
Math38,13111.6 MiB
Science50,00045.6 MiB
Code19,1691,432.5 MiB
IF16,5759.3 MiB
Agent10,2290.2 MiB
Mix87,6991,505.1 MiB

Mix is the exact blend used in our experiments: it subsamples Math and Science while using Code, IF, and Agent in full. It is released verbatim rather than as a sampling script, since re-running the sampler would not reproduce the same rows.

Provenance and licensing

Each subset is a reformatting of a public upstream dataset into verl's RLHF schema. Attribution and terms follow the upstream sources:

SubsetUpstreamLicense
MathPOLARIS-Project/Polaris-Dataset-53KApache-2.0
Sciencenvidia/OpenScienceReasoning-2CC-BY-4.0
Codenvidia/Nemotron-RL-coding-competitive_codingApache-2.0
IFnvidia/Nemotron-RL-instruction_followingApache-2.0
Agentnvidia/Nemotron-RL-agent-workplace_assistantApache-2.0
Mixall five aboveas the component subsets

Code, IF, and Agent come from NVIDIA's Nemotron RL collection, released through NeMo Gym.

Usage

python
from datasets import load_dataset

math_train = load_dataset("<repo-id>", "Math", split="train")