Mathematical Reasoning on Math10K
78.2SVAMPPRISM
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| PRISMBackbone=Gemma-3-12B-pt, Type=Multimodal, r=16, epsilon=6, delta=10^-52026.05 | 78.2 | — | — | — | 65.35 | 53.15 | 69.66 | 81.93 | |
| PRISMBackbone=Gemma-2-9B, Type=Text-only, r=16, epsilon=6, delta=10^-52026.05 | 77.9 | — | — | — | 66.03 | 51.97 | 70.19 | 84.87 | |
| AdamWBackbone=Gemma-2-9B, Type=Text-only, r=16, epsilon=6, delta=10^-52026.05 | 75.7 | — | — | — | 64.73 | 49.79 | 67.79 | 80.93 | |
| LoRA+Backbone=Gemma-3-12B-pt, Type=Multimodal, r=16, epsilon=6, delta=10^-52026.05 | 74.6 | — | — | — | 63.46 | 50.39 | 67.6 | 81.93 | |
| RITESETTING=NON-DP, Backbone=Gemma-3-4B-pt2026.05 | 73.6 | — | — | — | 59.5 | 48.8 | 78.2 | 89.9 | |
| LORA+SETTING=NON-DP, Backbone=Gemma-3-4B-pt2026.05 | 71.2 | — | — | — | 59.2 | 46.5 | 76.9 | 89.1 | |
| LAMBSETTING=NON-DP, Backbone=Gemma-3-4B-pt2026.05 | 70.8 | — | — | — | 55.9 | 44.9 | 75.6 | 87.8 | |
| ADAMWSETTING=NON-DP, Backbone=Gemma-3-4B-pt2026.05 | 69.8 | — | — | — | 56.1 | 47.6 | 75.2 | 87 | |
| LoRA+Backbone=Gemma-2-9B, Type=Text-only, r=16, epsilon=6, delta=10^-52026.05 | 69.7 | — | — | — | 62.93 | 44.09 | 64.35 | 80.67 | |
| PRISMSETTING=NON-DP, Backbone=Gemma-3-4B-pt2026.05 | 69.3 | — | — | — | 55.2 | 47.2 | 73.7 | 89.5 | |
| AdamWBackbone=Gemma-3-12B-pt, Type=Multimodal, r=16, epsilon=6, delta=10^-52026.05 | 68.7 | — | — | — | 58.07 | 47.64 | 61.88 | 73.11 | |
| FFASETTING=NON-DP, Backbone=Gemma-3-4B-pt2026.05 | 67.8 | — | — | — | 51.3 | 47.6 | 71.2 | 83.6 | |
| PRISMSETTING=ε = 6, Backbone=Gemma-3-4B-pt, δ=10^-52026.05 | 62.6 | — | — | — | 46.9 | 44.5 | 69 | 81.9 | |
| PRISMBackbone=Gemma-3-4B-pt, Type=Multimodal, r=16, epsilon=6, delta=10^-52026.05 | 62.6 | — | — | — | 46.9 | 44.5 | 59 | 81.9 | |
| ADAMWSETTING=ε = 6, Backbone=Gemma-3-4B-pt, δ=10^-52026.05 | 61.5 | — | — | — | 44.1 | 46.5 | 64.1 | 76.1 | |
| AdamWBackbone=Gemma-3-4B-pt, Type=Multimodal, r=16, epsilon=6, delta=10^-52026.05 | 61.5 | — | — | — | 44.1 | 46.5 | 57.1 | 76.1 | |
| PRISMSETTING=ε = 3, Backbone=Gemma-3-4B-pt, δ=10^-52026.05 | 61.4 | — | — | — | 45.6 | 40.6 | 68 | 80.7 | |
| FFASETTING=ε = 6, Backbone=Gemma-3-4B-pt, δ=10^-52026.05 | 61.1 | — | — | — | 37.5 | 39 | 58.9 | 73.5 | |
| LORA+SETTING=ε = 6, Backbone=Gemma-3-4B-pt, δ=10^-52026.05 | 61.1 | — | — | — | 44.6 | 40.9 | 67.4 | 78.6 | |
| LoRA+Backbone=Gemma-3-4B-pt, Type=Multimodal, r=16, epsilon=6, delta=10^-52026.05 | 61.1 | — | — | — | 44.6 | 40.9 | 56.3 | 78.6 | |
| LORA+SETTING=ε = 3, Backbone=Gemma-3-4B-pt, δ=10^-52026.05 | 60.9 | — | — | — | 41.9 | 38.6 | 66.5 | 76.5 | |
| LAMBSETTING=ε = 3, Backbone=Gemma-3-4B-pt, δ=10^-52026.05 | 60.8 | — | — | — | 39.3 | 42.5 | 60.9 | 74.4 | |
| FFASETTING=ε = 3, Backbone=Gemma-3-4B-pt, δ=10^-52026.05 | 59.8 | — | — | — | 35 | 37.4 | 56.9 | 71.8 | |
| LAMBSETTING=ε = 6, Backbone=Gemma-3-4B-pt, δ=10^-52026.05 | 59.2 | — | — | — | 42.5 | 43.7 | 62.1 | 76.1 | |
| ADAMWSETTING=ε = 3, Backbone=Gemma-3-4B-pt, δ=10^-52026.05 | 59.1 | — | — | — | 44.6 | 41.3 | 63.4 | 73.1 | |
| RITESETTING=ε = 6, Backbone=Gemma-3-4B-pt, δ=10^-52026.05 | 50.3 | — | — | — | 28.2 | 36.6 | 49 | 59.7 | |
| RITESETTING=ε = 3, Backbone=Gemma-3-4B-pt, δ=10^-52026.05 | 47.4 | — | — | — | 25.5 | 36.2 | 46.2 | 52.5 | |
| GenLoRABackbone=Qwen2.5-14B, Rank (r)=32, Groups (g)=8, #Params=2.26M2026.02 | 0.878 | 0.9165 | 0.9617 | 0.9331 | 0.7468 | 0.3858 | 0.8036 | — | |
| LoRABackbone=Qwen2.5-14B, Rank (r)=8, #Params=8.65M2026.02 | 0.861 | 0.9114 | 0.9617 | 0.9252 | 0.7551 | 0.3346 | 0.7915 | — | |
| GenLoRABackbone=Qwen2.5-14B, Rank (r)=8, Groups (g)=8, #Params=1.38M2026.02 | 0.859 | 0.9215 | 0.96 | 0.9213 | 0.7619 | 0.3583 | 0.797 | — | |
| DoRABackbone=Qwen2.5-14B, Rank (r)=8, #Params=8.99M2026.02 | 0.855 | 0.9316 | 0.96 | 0.9252 | 0.7528 | 0.3386 | 0.7939 | — | |
| AuroraBackbone=Qwen2.5-14B, Rank (r)=8, #Params=8.65M2026.02 | 0.85 | 0.9316 | 0.965 | 0.9252 | 0.7574 | 0.3465 | 0.7959 | — | |
| HiRABackbone=Qwen2.5-14B, Rank (r)=8, #Params=8.65M2026.02 | 0.844 | 0.9241 | 0.9467 | 0.9114 | 0.7521 | 0.3543 | 0.7888 | — | |
| AuroraBackbone=Qwen2.5-14B, Rank (r)=2, #Params=2.16M2026.02 | 0.841 | 0.9341 | 0.95 | 0.9173 | 0.7324 | 0.3307 | 0.7843 | — | |
| MELoRABackbone=Qwen2.5-14B, Rank (r)=8, #Params=8.65M2026.02 | 0.836 | 0.9291 | 0.94 | 0.9173 | 0.7475 | 0.3189 | 0.7815 | — | |
| DoRABackbone=Gemma-7B, Rank (r)=8, #Params=5.16M2026.02 | 0.76 | 0.8658 | 0.9033 | 0.9016 | 0.5709 | 0.2638 | 0.7109 | — | |
| AuroraBackbone=Gemma-7B, Rank (r)=2, #Params=1.20M2026.02 | 0.736 | 0.8582 | 0.8733 | 0.8858 | 0.5686 | 0.2795 | 0.7003 | — | |
| AuroraBackbone=Gemma-7B, Rank (r)=8, #Params=4.82M2026.02 | 0.736 | 0.8684 | 0.8983 | 0.9075 | 0.577 | 0.2638 | 0.7085 | — | |
| LoRABackbone=Gemma-7B, Rank (r)=8, #Params=4.82M2026.02 | 0.727 | 0.8658 | 0.8933 | 0.8957 | 0.5671 | 0.3071 | 0.7093 | — | |
| HiRABackbone=Gemma-7B, Rank (r)=8, #Params=4.82M2026.02 | 0.721 | 0.8585 | 0.8867 | 0.9055 | 0.5504 | 0.2638 | 0.6976 | — | |
| GenLoRABackbone=Gemma-7B, Rank (r)=16, Groups (g)=16, #Params=1.29M2026.02 | 0.718 | 0.8734 | 0.9217 | 0.9055 | 0.5619 | 0.2835 | 0.7156 | — | |
| MELoRABackbone=Gemma-7B, Rank (r)=8, #Params=4.82M2026.02 | 0.714 | 0.8481 | 0.89 | 0.8996 | 0.5512 | 0.2953 | 0.6997 | — | |
| GenLoRABackbone=Gemma-7B, Rank (r)=8, Groups (g)=16, #Params=0.95M2026.02 | 0.71 | 0.8456 | 0.9133 | 0.9094 | 0.5686 | 0.2874 | 0.7057 | — | |
| GenLoRABackbone=LLaMA-3-8B, Rank (r)=8, Groups (g)=16, #Params=0.98M2026.02 | 0.708 | 0.8835 | 0.9067 | 0.937 | 0.539 | 0.2362 | 0.7017 | — | |
| GenLoRABackbone=LLaMA-3-8B, Rank (r)=32, Groups (g)=16, #Params=2.16M2026.02 | 0.708 | 0.8709 | 0.9271 | 0.9449 | 0.5618 | 0.2559 | 0.7105 | — | |
| HiRABackbone=LLaMA-3-8B, Rank (r)=8, #Params=4.72M2026.02 | 0.689 | 0.8582 | 0.8816 | 0.9134 | 0.5466 | 0.2755 | 0.6941 | — | |
| AuroraBackbone=LLaMA-3-8B, Rank (r)=8, #Params=4.72M2026.02 | 0.673 | 0.8025 | 0.875 | 0.8957 | 0.5686 | 0.2047 | 0.6699 | — | |
| DoRABackbone=LLaMA-3-8B, Rank (r)=8, #Params=4.92M2026.02 | 0.666 | 0.8304 | 0.8767 | 0.9114 | 0.5641 | 0.2165 | 0.6775 | — | |
| LoRABackbone=LLaMA-3-8B, Rank (r)=8, #Params=4.72M2026.02 | 0.664 | 0.8228 | 0.8583 | 0.8976 | 0.5618 | 0.2323 | 0.6787 | — | |
| AuroraBackbone=LLaMA-3-8B, Rank (r)=2, #Params=1.18M2026.02 | 0.644 | 0.8 | 0.8367 | 0.8819 | 0.5027 | 0.2559 | 0.6535 | — | |
| MELoRABackbone=LLaMA-3-8B, Rank (r)=8, #Params=4.72M2026.02 | 0.629 | 0.838 | 0.84 | 0.8661 | 0.4784 | 0.2283 | 0.647 | — |