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

Coding on HumanEval (test)

74.4Test Accuracy

KL-SFT

Updated 1mo ago

Evaluation Results

Method	Links
KL-SFT 2026.05		74.4
Base 2026.05		73.8
STM 2026.05		73.2
Low-SFT 2026.05		72.6
Anchored Learning 2026.05		70.7
DFT 2026.05		70.1
Iter-SFT 2026.05		70.1
SFT 2026.05		68.3
Self-SFT 2026.05		67.1
Qwen3-4B-Base 2026.06		65.24
FReDA-4B 2026.06		63.41
FReDA-4B 2026.06		60.98
TiDAR-8B 2026.06		57.93
Dream-7B-Base 2026.06		57.9
BlockDiff-4B 2026.06		57.32
TiDAR-8B 2026.06		55.49
LLaDA-MoE-7B-A1B-Base 2026.06		45.73
Qwen2.5-3B-Base 2026.06		42.1
SPG w/ EUBO 2025.10		41.5
UniGRPO 2025.10		39.6
SPG w/ Mixture 2025.10		39.6
WD1 2025.10		39
SPG w/ Mixture 2025.10		39
WD1 2025.10		38.4
UniGRPO 2025.10		38.4
D1 2025.10		37.8
D1 2025.10		37.2
LLaDA-8B-Inst. 2025.10		35.4
LLaDA-8B-Base 2026.06		35.4
LLaDA-8B-Inst. 2025.10		34.8
LLaDA-1.5 2025.10		34.8
SPG w/ EUBO 2025.10		33.5
LLaDA-1.5 2025.10		32.3
SPG w/ Mixture 2025.10		31.1
D1 2025.10		29.3
SPG w/ EUBO 2025.10		29.3
LLaDA-8B-Inst. 2025.10		28.1
UniGRPO 2025.10		28.1
WD1 2025.10		25.6
LLaDA-1.5 2025.10		22.6