List your company

Logical Reasoning on Shuffled Objects

92.1AccuracyGateKD
19.61238.43157.2576.069Jan 12, 2026Feb 1, 2026Feb 21, 2026Mar 13, 2026Apr 2, 2026Apr 22, 2026May 13, 2026

Evaluation results

MethodLinks
GateKDModel=FlanT5-base, #Params=250M2026.0592.1——
GateKDModel=T5-base, #Params=250M2026.0590.6——
In-Writing-BFBackbone=Qwen3-8B, Shot=02026.0189.9——
Vanilla-KDModel=FlanT5-large, #Params=780M2026.0589.7——
In-Writing-IFBackbone=Qwen3-8B, Shot=02026.0188.7——
In-Writing-BFBackbone=Qwen3-8B, Shot=42026.0188.7——
In-Writing-BFBackbone=Qwen3-8B, Shot=12026.0188.3——
Mentor-KDModel=FlanT5-base, #Params=250M2026.0588.3——
Vanilla-KDModel=T5-large, #Params=780M2026.0588.1——
In-Writing-BaseBackbone=Qwen3-8B, Shot=02026.0187.6——
In-Writing-IFBackbone=Qwen3-8B, Shot=42026.0186.7——
MCC-KDModel=FlanT5-base, #Params=250M2026.0585.1——
Mentor-KDModel=T5-base, #Params=250M2026.0584.9——
In-Writing-IFBackbone=Qwen3-8B, Shot=12026.0184.8——
ZS-CoT (teacher)Model=GPT-4o-mini, #Params=-2026.0584.3——
In-Writing-BaseBackbone=Qwen3-8B, Shot=42026.0184——
GateKDModel=FlanT5-small, #Params=80M2026.0583.7——
ZS-CoT (teacher)Model=GPT-4o-mini, #Params=-2026.0583.1——
In-Writing-BaseBackbone=Qwen3-8B, Shot=12026.0182.6——
Vanilla-KDModel=FlanT5-base, #Params=250M2026.0582.6——
MCC-KDModel=T5-base, #Params=250M2026.0581——
GateKDModel=T5-small, #Params=80M2026.0580.8——
Vanilla-KDModel=T5-base, #Params=250M2026.0578.4——
Mentor-KDModel=FlanT5-small, #Params=80M2026.0576.4——
Mentor-KDModel=T5-small, #Params=80M2026.0572.9——
MCC-KDModel=FlanT5-small, #Params=80M2026.0570——
Vanilla-KDModel=FlanT5-small, #Params=80M2026.0568.2——
MCC-KDModel=T5-small, #Params=80M2026.0566.1——
Vanilla-KDModel=T5-small, #Params=80M2026.0563.7——
In-Writing-BFBackbone=LLaMA3-8B, Shot=12026.0148.3——
In-Writing-IFBackbone=LLaMA3-8B, Shot=42026.0146.7——
In-Writing-BFBackbone=LLaMA3-8B, Shot=42026.0146.4——
In-Writing-IFBackbone=LLaMA3-8B, Shot=12026.0145.8——
In-Writing-BaseBackbone=LLaMA3-8B, Shot=02026.0139——
In-Writing-BaseBackbone=LLaMA3-8B, Shot=42026.0138.2——
In-Writing-IFBackbone=LLaMA3-8B, Shot=02026.0134.8——
NL-to FormatBackbone=LLaMA3-8B, Shot=02026.0127——
In-Writing-BFBackbone=LLaMA3-8B, Shot=02026.0124.3——
In-Writing-BaseBackbone=LLaMA3-8B, Shot=12026.0122.4——
zero-shotReasoning Method=Zero-shot2022.05—31.329.7
zero-shot-cotReasoning Method=Zero-shot-CoT2022.05—52.452.9