Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Free-form answer generation on KLAR (OOD)
Loading...
38.13
Accuracy
GRPO
6.5868
14.7759
22.965
31.1541
Jun 4, 2026
Accuracy
Updated 1mo ago
Evaluation Results
Method
Method
Links
Accuracy
GRPO
Backbone=Qwen-2.5-7B
2026.06
38.13
Baseline
Backbone=Qwen-2.5-7B
2026.06
35.76
CPT + GRPO
Backbone=Qwen-2.5-7B
2026.06
30.75
CPT + SFT
Backbone=Qwen-2.5-7B
2026.06
30.21
SFT
Backbone=Qwen-2.5-7B
2026.06
29.96
CPT
Backbone=Qwen-2.5-7B
2026.06
23.75
CPT + GRPO
Backbone=OLMo-2-1124-7B
2026.06
17.6
GRPO
Backbone=OLMo-2-1124-7B
2026.06
16.7
Baseline
Backbone=OLMo-2-1124-7B
2026.06
13.3
CPT + SFT
Backbone=OLMo-2-1124-7B
2026.06
8.6
CPT
Backbone=OLMo-2-1124-7B
2026.06
8.3
SFT
Backbone=OLMo-2-1124-7B
2026.06
7.8
Feedback
Search any
task
Search any
task