List your company
ResearchFormerly WizWand

Mathematical Reasoning on MATH 500 (Accuracy Avg@8)

89.05Accuracy (Avg@8)L1-Qwen-7B-Max
53.357262.623671.8981.1564May 28, 2026Jun 3, 2026Jun 10, 2026Jun 16, 2026Jun 23, 2026Jun 29, 2026Jul 6, 2026
Updated 3mo ago

Evaluation results

MethodLinks
L1-Qwen-7B-Max# Bits=BF16, Backbone=L1-Qwen-7B-Max2026.0589.05
FlexRound+LFQ# Bits=W4, Backbone=L1-Qwen-7B-Max2026.0588.4
FlexRound# Bits=W4, Backbone=L1-Qwen-7B-Max2026.0587.45
FlexRound+LFQ# Bits=W3g128, Backbone=L1-Qwen-7B-Max2026.0586.5
FlexRound# Bits=W3g128, Backbone=L1-Qwen-7B-Max2026.0585.35
TOP-DStudent Model=Qwen3-1.7B-Base, Paradigm=Distillation, Teacher Model=Qwen3-30B-A3B-Instruct-25072026.0777.33
DeepSeek-R1-Distill-Llama-8B# Bits=BF16, Backbone=DeepSeek-R1-Distill-Llama-8B2026.0572.53
FlexRound+LFQ# Bits=W4, Backbone=DeepSeek-R1-Distill-Llama-8B2026.0571.95
OPDStudent Model=Qwen3-1.7B-Base, Paradigm=Distillation, Teacher Model=Qwen3-30B-A3B-Instruct-25072026.0771.13
FlexRound# Bits=W4, Backbone=DeepSeek-R1-Distill-Llama-8B2026.0570.1
FlexRound+LFQ# Bits=W3g128, Backbone=DeepSeek-R1-Distill-Llama-8B2026.0569.25
FlexRound# Bits=W3g128, Backbone=DeepSeek-R1-Distill-Llama-8B2026.0567
DAPOStudent Model=Qwen3-1.7B-Base, Paradigm=RLVR, Teacher Model=Qwen3-30B-A3B-Instruct-25072026.0761.88
BaseStudent Model=Qwen3-1.7B-Base, Paradigm=Base, Teacher Model=Qwen3-30B-A3B-Instruct-25072026.0756.8
GRPOStudent Model=Qwen3-1.7B-Base, Paradigm=RLVR, Teacher Model=Qwen3-30B-A3B-Instruct-25072026.0754.73