Share your thoughts, 1 month free Claude Pro on usSee more

Multiple-choice Reasoning on Hellaswag (bo)

39.16Accuracy

Ours-MoE-SFT

Updated 20d ago

Evaluation Results

Method	Links
Ours-MoE-SFT 2025.07		39.16	45.71
Ours-SFT 2025.07		30.64	36.71
Ours-SFT 2025.07		30.64	36.71
Ours-Base 2025.07		30.3	36.2
Ours-Base 2025.07		30.3	36.2
Ours-Base-32k 2025.07		30.09	36.58
Ours-Base-32k 2025.07		30.09	36.58
Ours-MoE-Base-8k 2025.07		29.68	36.29
Ours-MoE-Base 2025.07		29.54	36.16
LLaMA3.1-8B-Instruct 2025.07		27.28	32
Qwen3-8B 2025.07		27.07	30.56
Qwen2.5-7B-base 2025.07		26.69	31.57
Qwen2.5-7B-Instruct 2025.07		26.46	31.03
Tibetan-Llama2-7B 2025.07		26.09	27.58
DeepSeek-R1-Distill-Llama-8B 2025.07		25.96	30.67
Tibetan-Alpaca-7B 2025.07		25.95	28.23
Yak-Llama2-7B 2025.07		24.67	26.4