List your company
ResearchFormerly WizWand

Math Reasoning on MATH (held-out)

26.04PerformanceENVISIONS
0.33127.005613.6820.3544Jun 17, 2024
Updated 8mo ago

Evaluation results

MethodLinks
ENVISIONSBackbone=LLaMA2-Chat (13B), Approach=Env-guided Self-Training2024.0626.0463.56
GPT-4 + LLaMA2-ChatBackbone=LLaMA2-Chat (13B), Approach=Distill-then-Finetune2024.0623.6456.49
Claude-2 + LLaMA2-ChatBackbone=LLaMA2-Chat (13B), Approach=Distill-then-Finetune2024.0623.4757.17
Iterative SFT+DPOBackbone=LLaMA2-Chat (13B), Approach=Reinforced Self-Training2024.0622.3259.77
STaR + Env.Backbone=LLaMA2-Chat (13B), Approach=Env-guided Self-Training2024.0620.5760.6
ENVISIONSBackbone=LLaMA2-Chat (7B), Approach=Env-guided Self-Training2024.061958.69
GPT-4 + LLaMA2-ChatBackbone=LLaMA2-Chat (7B), Approach=Distill-then-Finetune2024.0618.8452.46
Claude-2 + LLaMA2-ChatBackbone=LLaMA2-Chat (7B), Approach=Distill-then-Finetune2024.0618.1753.03
STaR + Env.Backbone=LLaMA2-Chat (7B), Approach=Env-guided Self-Training2024.0615.9755.91
Iterative SFT+DPOBackbone=LLaMA2-Chat (7B), Approach=Reinforced Self-Training2024.0614.7554.73
Self-RewardingBackbone=LLaMA2-Chat (13B), Approach=Reinforced Self-Training2024.0613.9752.5
Self-RewardingBackbone=LLaMA2-Chat (7B), Approach=Reinforced Self-Training2024.0610.744.22
LLaMA2-Chat (few-shot)Backbone=LLaMA2-Chat (13B), Mode=few-shot2024.066.0738.61
LLaMA2-Chat (few-shot)Backbone=LLaMA2-Chat (7B), Mode=few-shot2024.061.3225.69