List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME 24 (pass@4)

60Pass@4Hybrid-LoRA
1.03216.34131.6546.959May 12, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Hybrid-LoRABackbone=Qwen-2.5 7B Instruct, Post-training Approach=GSPO, #Tunable Params=449M2026.0560
Hybrid-LoRABackbone=Qwen-2.5 7B Instruct, Post-training Approach=GRPO, #Tunable Params=449M2026.0556
AutoLoRABackbone=Qwen-2.5 7B Instruct, Post-training Approach=GSPO, #Tunable Params=453M2026.0555.3
AutoLoRABackbone=Qwen-2.5 7B Instruct, Post-training Approach=GRPO, #Tunable Params=453M2026.0552.7
Full-FT (ref.)Backbone=Qwen-2.5 3B Instruct, Post-training Approach=GSPO, #Tunable Params=3086M2026.0549.3
Hybrid-LoRABackbone=Qwen-2.5 3B Instruct, Post-training Approach=GSPO, #Tunable Params=192M2026.0548
Full-FT (ref.)Backbone=Qwen-2.5 3B Instruct, Post-training Approach=GRPO, #Tunable Params=3086M2026.0546
AutoLoRABackbone=Qwen-2.5 3B Instruct, Post-training Approach=GSPO, #Tunable Params=194M2026.0544
Hybrid-LoRABackbone=Qwen-2.5 3B Instruct, Post-training Approach=GRPO, #Tunable Params=192M2026.0543.3
Full-FT (ref.)Backbone=Qwen-2.5 1.5B Instruct, Post-training Approach=GSPO, #Tunable Params=1544M2026.0542
Hybrid-LoRABackbone=Qwen-2.5 1.5B Instruct, Post-training Approach=GSPO, #Tunable Params=105M2026.0542
AutoLoRABackbone=Qwen-2.5 3B Instruct, Post-training Approach=GRPO, #Tunable Params=201M2026.0540.7
Full-FT (ref.)Backbone=Qwen-2.5 1.5B Instruct, Post-training Approach=GRPO, #Tunable Params=1544M2026.0540
LoRA-DropBackbone=Qwen-2.5 1.5B Instruct, Post-training Approach=GSPO, #Tunable Params=105M2026.0540
Hybrid-LoRABackbone=Qwen-2.5 1.5B Instruct, Post-training Approach=GRPO, #Tunable Params=105M2026.0538
AdaLoRABackbone=Qwen-2.5 1.5B Instruct, Post-training Approach=GSPO, #Tunable Params=106M2026.0536
AutoLoRABackbone=Qwen-2.5 1.5B Instruct, Post-training Approach=GSPO, #Tunable Params=107M2026.0535.3
AutoLoRABackbone=Qwen-2.5 1.5B Instruct, Post-training Approach=GRPO, #Tunable Params=106M2026.0534.7
LoRA-DropBackbone=Qwen-2.5 1.5B Instruct, Post-training Approach=GRPO, #Tunable Params=105M2026.0534
LoRA-FFNBackbone=Qwen-2.5 1.5B Instruct, Post-training Approach=GSPO, #Tunable Params=122M2026.0534
AdaLoRABackbone=Qwen-2.5 1.5B Instruct, Post-training Approach=GRPO, #Tunable Params=105M2026.0532
CoTBackbone=Qwen-2.5 3B Instruct, Post-training Approach=-, #Tunable Params=0.02026.0524
Full-FTBackbone=Qwen-2.5 1.5B Instruct, Post-training Approach=SFT, #Tunable Params=1544M2026.0522.7
COTBackbone=Qwen-2.5 1.5B Instruct, Post-training Approach=COT, #Tunable Params=0.02026.0514.7
IOBackbone=Qwen-2.5 1.5B Instruct, Post-training Approach=IO, #Tunable Params=0.02026.053.3