List your company

Preservation of General Capabilities on HellaSwag, WinoGrande, IFEval, MMLU

20.4HellaSwag DeltaWiSE-FT
-33.888-19.794-5.78.394May 19, 2026

Evaluation results

MethodLinks
WiSE-FTModel=Llama-3-8B2026.0520.43.450720.2
WiSE-FTModel=Qwen3-4B2026.0514.73.237.87.215.7
L2 RegModel=Llama-3-8B2026.0510.80.811.83.96.4
SFTModel=Llama-3-8B2026.0510.40.321.73.69
FLOWModel=Llama-3-8B2026.058.42.536.9612.2
DFTModel=Llama-3-8B2026.057.82.492.24.2
SFTModel=Qwen3-4B2026.057.329.419.69.6
FLOWModel=Qwen3-4B2026.056.40.713.712.78
L2 RegModel=Qwen3-4B2026.055.50.21.914.35.4
TALRModel=Llama-3-8B2026.055.50.612.324.8
LoRAModel=Qwen3-4B2026.054.46.99.19.27.4
FINCHModel=Llama-3-8B2026.054.22.202.61.2
SFT (small lr)Model=Qwen3-4B2026.053.82.23.13.21.5
DFTModel=Qwen3-4B2026.052.72.731.91.1
DFTModel=Qwen3-4B2026.052.6-0.65.22.32.4
TALRModel=Qwen3-4B2026.051.314.55.22.5
STMModel=Qwen3-4B2026.051.21.11.40.90.6
LoRAModel=Llama-3-8B2026.050.90.615.73.85.3
SFT (small lr)Model=Llama-3-8B2026.050.70.56.20.91.4
FINCHModel=Qwen3-4B2026.050.61.40.50.80.1
STMModel=Llama-3-8B2026.050.4012.30.73.4
TALRModel=Qwen3-4B2026.050.30.4-3.5-0.8-0.9
SFT (small lr)Model=Qwen3-4B2026.050.1-2.4-0.1-0.4-0.7
BaseModel=Qwen3-4B2026.0500000
BaseModel=Llama-3-8B2026.0500000
BaseModel=Qwen3-4B2026.0500000
BaseModel=Llama-3-8B2026.0500000
STMModel=Qwen3-4B2026.05-0.80.2-4.10.8-1
FINCHModel=Llama-3-8B2026.05-1.2-0.600-0.5
FINCHModel=Qwen3-4B2026.05-1.6-2.70-5.1-2.4
SFT (small lr)Model=Llama-3-8B2026.05-3.60.2-4.10.4-1.8
STMModel=Llama-3-8B2026.05-4.2-1.4-8.3-2-4
LoRAModel=Llama-3-8B2026.05-4.6-2.2-16.7-1-6.1
DFTModel=Llama-3-8B2026.05-5.1-3.4-20.8-5-8.6
TALRModel=Llama-3-8B2026.05-6.4-0.4-8.30-3.8
SFTModel=Qwen3-4B2026.05-10-11.6-25.1-10.4-14.3
LoRAModel=Qwen3-4B2026.05-10.85.1-45.6-2.1-13.4
SFTModel=Llama-3-8B2026.05-12.2-0.8-16.7-4.2-8.5
L2 RegModel=Llama-3-8B2026.05-12.2-1.8-16.7-4-8.7
L2 RegModel=Qwen3-4B2026.05-12.4-11.7-39.2-11.2-18.6
WiSE-FTModel=Llama-3-8B2026.05-12.6-2-16.7-5-9.1
FLOWModel=Qwen3-4B2026.05-15.5-8.7-15.6-9.4-12.3
WiSE-FTModel=Qwen3-4B2026.05-16.3-9.6-26.6-7.8-15.1
FLOWModel=Llama-3-8B2026.05-31.8-5.2-20.8-13.2-17.8