List your company

Multiple-choice Reasoning on Hellaswag (bo)

39.16AccuracyOurs-MoE-SFT
24.090428.002731.91535.8273Jul 12, 2025

Evaluation results

MethodLinks
Ours-MoE-SFTModel=Ours-MoE-SFT2025.0739.1645.71
Ours-SFTalignment=SFT2025.0730.6436.71
Ours-SFTModel=Ours-SFT2025.0730.6436.71
Ours-Basemode=base2025.0730.336.2
Ours-BaseModel=Ours-Base2025.0730.336.2
Ours-Base-32kcontext-length=32k2025.0730.0936.58
Ours-Base-32kModel=Ours-Base-32k2025.0730.0936.58
Ours-MoE-Base-8kModel=Ours-MoE-Base-8k2025.0729.6836.29
Ours-MoE-BaseModel=Ours-MoE-Base2025.0729.5436.16
LLaMA3.1-8B-InstructModel=LLaMA3.1-8B-Instruct2025.0727.2832
Qwen3-8BModel=Qwen3-8B2025.0727.0730.56
Qwen2.5-7B-baseModel=Qwen2.5-7B-base2025.0726.6931.57
Qwen2.5-7B-InstructModel=Qwen2.5-7B-Instruct2025.0726.4631.03
Tibetan-Llama2-7Bbackbone=Llama2-7B2025.0726.0927.58
DeepSeek-R1-Distill-Llama-8BModel=DeepSeek-R1-Distill-Llama-8B2025.0725.9630.67
Tibetan-Alpaca-7Bbackbone=Alpaca-7B2025.0725.9528.23
Yak-Llama2-7Bbackbone=Llama2-7B2025.0724.6726.4