List your company

General Reasoning on MMLU-stem and ARC-challenge (test)

61.8AccuracyOrigin
47.7651.40555.0558.695Feb 12, 2026

Evaluation results

MethodLinks
OriginBase Model=Qwen2.5-7B-instruct, Budget (FLOPs)=02026.0261.8
RPOBase Model=Qwen2.5-7B-instruct, Budget (FLOPs)=77.88 x 10^182026.0261.8
SimPOBase Model=Qwen2.5-7B-instruct, Budget (FLOPs)=77.88 x 10^182026.0261.8
DPOBase Model=Qwen2.5-7B-instruct, Budget (FLOPs)=77.88 x 10^182026.0261.6
HD+IDFTBase Model=Qwen2.5-7B-instruct, Budget (FLOPs)=44.70 x 10^182026.0261.5
Rej@4Base Model=Qwen2.5-7B-instruct, Budget (FLOPs)=60.19 x 10^182026.0261.4
CPOBase Model=Qwen2.5-7B-instruct, Budget (FLOPs)=77.88 x 10^182026.0261.4
Rej@2Base Model=Qwen2.5-7B-instruct, Budget (FLOPs)=46.03 x 10^182026.0261.2
HD+SFTBase Model=Qwen2.5-7B-instruct, Budget (FLOPs)=44.70 x 10^182026.0261.2
RPOBase Model=Phi4-mini-instruct (3.8B), Budget (FLOPs)=40.39 x 10^182026.0259.6
CPOBase Model=Phi4-mini-instruct (3.8B), Budget (FLOPs)=40.39 x 10^182026.0259.6
HD+IDFTBase Model=Phi4-mini-instruct (3.8B), Budget (FLOPs)=19.86 x 10^182026.0259.4
DPOBase Model=Phi4-mini-instruct (3.8B), Budget (FLOPs)=40.39 x 10^182026.0259.3
SimPOBase Model=Phi4-mini-instruct (3.8B), Budget (FLOPs)=40.39 x 10^182026.0259.3
OriginBase Model=Phi4-mini-instruct (3.8B), Budget (FLOPs)=02026.0259.2
Rej@4Base Model=Phi4-mini-instruct (3.8B), Budget (FLOPs)=35.81 x 10^182026.0259.2
Rej@2Base Model=Phi4-mini-instruct (3.8B), Budget (FLOPs)=25.40 x 10^182026.0258.7
HD+SFTBase Model=Phi4-mini-instruct (3.8B), Budget (FLOPs)=19.86 x 10^182026.0258.7
SFTBase Model=Qwen2.5-7B-instruct, Budget (FLOPs)=31.87 x 10^182026.0256.9
EAFTBase Model=Qwen2.5-7B-instruct, Budget (FLOPs)=31.87 x 10^182026.0255.9
EAFTBase Model=Phi4-mini-instruct (3.8B), Budget (FLOPs)=14.99 x 10^182026.0251.6
DFTBase Model=Qwen2.5-7B-instruct, Budget (FLOPs)=31.87 x 10^182026.0250.5
SFTBase Model=Phi4-mini-instruct (3.8B), Budget (FLOPs)=14.99 x 10^182026.0250.1
DFTBase Model=Phi4-mini-instruct (3.8B), Budget (FLOPs)=14.99 x 10^182026.0248.3