kobzaond/RLVRAMBench

RLVRAMBench Which language-model training configurations can I use with the memory I have, and how much testing does that decision require? RLVRAMBench is a measurement dataset with open evaluation tasks for a specific language-model training system. It measures memory feasibility when response generation and reinforcement-learning updates share the same graphics processors. It provides measured outcomes, fixed prediction tasks, a budgeted decision replay, reference methods, and… See the full description on the dataset page: https://huggingface.co/datasets/kobzaond/RLVRAMBench.