List your company

General Capability Retention on Held-out Capability Average

82.6Unweighted Average AccuracyBase
222.92543.8564.775May 16, 2026

Evaluation results

MethodLinks
BaseModel backbone=Qwen3-4B-It, Fine-tuning method=Base2026.0582.6
MIXSDModel backbone=Qwen3-4B-It, Fine-tuning method=MIXSD, Mixing rate λ=0.72026.0579.9
MIXSDModel backbone=Qwen3-4B-It, Fine-tuning method=MIXSD, Mixing rate λ=0.52026.0577.5
MIXSDModel backbone=Qwen3-4B-It, Fine-tuning method=MIXSD, Mixing rate λ=02026.0576.4
MIXSDModel backbone=Qwen3-4B-It, Fine-tuning method=MIXSD, Mixing rate λ=0.32026.0574.3
MIXSDModel backbone=Qwen3-8B, Fine-tuning method=MIXSD, Mixing rate λ=0.52026.0572.4
BaseModel backbone=Qwen3-8B, Fine-tuning method=Base2026.0572.2
MIXSDModel backbone=Qwen3-8B, Fine-tuning method=MIXSD, Mixing rate λ=0.72026.0570.8
MIXSDModel backbone=Qwen3-8B, Fine-tuning method=MIXSD, Mixing rate λ=0.32026.0570
MIXSDModel backbone=Qwen3-8B, Fine-tuning method=MIXSD, Mixing rate λ=02026.0566.7
OPSDModel backbone=Qwen3-8B, Fine-tuning method=OPSD2026.0561.8
OPSDModel backbone=Qwen3-4B-It, Fine-tuning method=OPSD2026.0560.6
BaseModel backbone=Qwen3-1.7B, Fine-tuning method=Base2026.0556.5
SFTModel backbone=Qwen3-8B, Fine-tuning method=SFT2026.0546.1
SFTModel backbone=Qwen3-4B-It, Fine-tuning method=SFT2026.0543.2
MIXSDModel backbone=Qwen3-1.7B, Fine-tuning method=MIXSD, Mixing rate λ=0.52026.0540.3
MIXSDModel backbone=Qwen3-1.7B, Fine-tuning method=MIXSD, Mixing rate λ=0.32026.0539.9
MIXSDModel backbone=Qwen3-1.7B, Fine-tuning method=MIXSD, Mixing rate λ=0.72026.0536.1
SFTModel backbone=Qwen3-1.7B, Fine-tuning method=SFT2026.0514.3
MIXSDModel backbone=Qwen3-1.7B, Fine-tuning method=MIXSD, Mixing rate λ=02026.058.6
OPSDModel backbone=Qwen3-1.7B, Fine-tuning method=OPSD2026.055.1