ANY PROBLEMS YOU RUN INTO ARE DUE TO MEMORY ADDRESS AND OFFSETS BEING INCORRECT. I WILL BE LOOKING INTO A FIX BUT IT WONT BE FOR A LITTLE WHILE. IF YOU WANT A WORKING CHEAT JUST FIND THE POINTERS FROM ...
OpenRLHF is a high-performance RLHF framework built on Ray, DeepSpeed and HF Transformers: data = { "prompt": xxx, "query": xxx, "label": json.dumps({ 'uuid': uuid ...