Dialogue on AlpacaEval 2
64.2AlpacaEval2 ScoreFuseChat3.0
Evaluation results
| Method | Links | |
|---|---|---|
| FuseChat3.0#Params=8B2026.05 | 64.2 | |
| Gemma2-27B-it#Params=27B2026.05 | 58.9 | |
| SKillWeave (Ours)#Params=10B2026.05 | 52.8 | |
| SkillWeave (Delta-Come replacement)#Params=10B2026.05 | 52.7 | |
| SkillWeave (Self-Rewarding replacement)#Params=10B2026.05 | 50.8 | |
| Twin-Merging r1024#Params=21B2026.05 | 49.6 | |
| EMR-Merging#Params=16.7B2026.05 | 48.9 | |
| TALL-Mask#Params=16.7B2026.05 | 48.6 | |
| SkillWeave (PEFT replacement)#Params=10B2026.05 | 47.9 | |
| Self-Align#Params=8B2026.05 | 47.4 | |
| Routed LoRA r1024#Params=21B2026.05 | 47.1 | |
| SkillWeave (ASVD replacement)#Params=10B2026.05 | 47 | |
| Twin-Merging r512#Params=14.1B2026.05 | 46.6 | |
| Qwen2-57BA14B-it#Params=52B2026.05 | 46.4 | |
| Self-Rewarding#Params=8B2026.05 | 46.4 | |
| Routed LoRA r512#Params=14.1B2026.05 | 41 | |
| Qwen1.5-72B-Chat#Params=72B2026.05 | 40.6 | |
| Sequentially MTL#Params=8B2026.05 | 39.2 | |
| Self-MoE#Params=9B2026.05 | 38.8 | |
| Jointly MTL#Params=8B2026.05 | 37.9 | |
| Qwen2.5-14B#Params=14B2026.05 | 36.4 | |
| SkillWeave (Self-Specialize replacement)#Params=10B2026.05 | 35.6 | |
| PCB-Merge+DARE#Params=8B2026.05 | 35 | |
| PCB-Merging#Params=8B2026.05 | 34.8 | |
| Ties-Merging#Params=8B2026.05 | 33.9 | |
| FuseLLM#Params=8B2026.05 | 32.5 | |
| Task Arithmetic#Params=8B2026.05 | 29.2 | |
| Llama3.1-8B-Instruct#Params=8B2026.05 | 28.3 | |
| FuseChat3.0#Params=1.48B, Backbone=Llama-3.2-1B-Instruct2026.05 | 27.1 | |
| SkillWeave#Params=1.42B, Backbone=Llama-3.2-1B-Instruct2026.05 | 12.5 | |
| PEFT#Params=1.42B, Backbone=Llama-3.2-1B-Instruct2026.05 | 12.1 | |
| self-rewarding#Params=1.15B, Backbone=Llama-3.2-1B-Instruct2026.05 | 11.9 | |
| Twin-merging#Params=1.48B, Backbone=Llama-3.2-1B-Instruct2026.05 | 11.5 | |
| Llama3.2-1B-Instruct#Params=1.15B, Backbone=Llama-3.2-1B-Instruct2026.05 | 9.7 |