Coding on HumanEval (HumanEval metric)
79.9HumanEvalQwen2-57BA14B-it
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2-57BA14B-it#Params=52B2026.05 | 79.9 | |
| Gemma2-27B-it#Params=27B2026.05 | 78.7 | |
| SKillWeave (Ours)#Params=10B2026.05 | 75 | |
| SkillWeave (Delta-Come replacement)#Params=10B2026.05 | 74.9 | |
| Twin-Merging r1024#Params=73.82026.05 | 73.8 | |
| SkillWeave (ASVD replacement)#Params=10B2026.05 | 73.7 | |
| EMR-Merging#Params=16.7B2026.05 | 73.5 | |
| SkillWeave (PEFT replacement)#Params=10B2026.05 | 73.5 | |
| Routed LoRA r1024#Params=21B2026.05 | 73.2 | |
| Twin-Merging r512#Params=14.1B2026.05 | 73.2 | |
| FuseLLM#Params=8B2026.05 | 73.2 | |
| TALL-Mask#Params=16.7B2026.05 | 72.8 | |
| Routed LoRA r512#Params=14.1B2026.05 | 72.5 | |
| PCB-Merge+DARE#Params=8B2026.05 | 71.5 | |
| Qwen1.5-72B-Chat#Params=72B2026.05 | 71.3 | |
| PCB-Merging#Params=8B2026.05 | 71.3 | |
| FuseChat3.0#Params=8B2026.05 | 71.3 | |
| Ties-Merging#Params=8B2026.05 | 71.1 | |
| Jointly MTL#Params=8B2026.05 | 70.7 | |
| SkillWeave (Self-Specialize replacement)#Params=10B2026.05 | 70.7 | |
| Task Arithmetic#Params=8B2026.05 | 70.6 | |
| Self-MoE#Params=9B2026.05 | 70.6 | |
| SkillWeave (Self-Rewarding replacement)#Params=10B2026.05 | 70.3 | |
| Llama3.1-8B-Instruct#Params=8B2026.05 | 69.5 | |
| Sequentially MTL#Params=8B2026.05 | 69.4 | |
| Self-Align#Params=8B2026.05 | 69.2 | |
| Self-Rewarding#Params=8B2026.05 | 68.8 | |
| Qwen2.5-14B#Params=14B2026.05 | 56.7 |