Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Knowledge-Based QA on Global-MMLU Lite (5 Unseen Languages)
Loading...
71.9
MMLU Score
CAGrad
41.428
49.339
57.25
65.161
Jun 4, 2026
MMLU Score
Updated 1mo ago
Evaluation Results
Method
Method
Links
MMLU Score
CAGrad
Backbone=Qwen3-8B-Base
2026.06
71.9
MGDA
Backbone=Qwen3-8B-Base
2026.06
71
PCGrad
Backbone=Qwen3-8B-Base
2026.06
70.3
Base
Backbone=Qwen3-8B-Base
2026.06
70
V. SFT
Backbone=Qwen3-8B-Base
2026.06
69.1
MGDA
Backbone=Qwen3-4B-Base
2026.06
66
CAGrad
Backbone=Qwen3-4B-Base
2026.06
65.5
PCGrad
Backbone=Qwen3-4B-Base
2026.06
64
Base
Backbone=Qwen3-4B-Base
2026.06
63.6
V. SFT
Backbone=Qwen3-4B-Base
2026.06
62.3
PCGrad
Backbone=Llama-3.1-8B
2026.06
52.6
CAGrad
Backbone=Llama-3.1-8B
2026.06
52.1
MGDA
Backbone=Llama-3.1-8B
2026.06
51.7
MGDA
Backbone=Meta-Llama-3-8B
2026.06
50.6
CAGrad
Backbone=Meta-Llama-3-8B
2026.06
50.5
V. SFT
Backbone=Llama-3.1-8B
2026.06
50.5
PCGrad
Backbone=Meta-Llama-3-8B
2026.06
49
V. SFT
Backbone=Meta-Llama-3-8B
2026.06
48.9
Base
Backbone=Llama-3.1-8B
2026.06
48.9
Base
Backbone=Meta-Llama-3-8B
2026.06
42.6
Feedback
Search any
task
Search any
task