List your company
ResearchFormerly WizWand

Code Generation on HumanEval+ (Pass@1, Avg, ΔScore)

88.3Pass@1Uni-OPD
27.543243.316659.0974.8634Mar 13, 2026Mar 25, 2026Apr 7, 2026Apr 19, 2026May 2, 2026May 14, 2026May 27, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Uni-OPDDistillation Scenario=Single-Teacher Distillation, Model Scale=4B2026.0588.3——
Uni-OPDDistillation Scenario=Multi-Teacher Distillation, Model Scale=4B2026.0588——
ExOPDDistillation Scenario=Single-Teacher Distillation, Model Scale=4B2026.0586.9——
ExPODistillation Scenario=Multi-Teacher Distillation, Model Scale=4B2026.0586.7——
SFTDistillation Scenario=Multi-Teacher Distillation, Model Scale=4B2026.0586.4——
OPDDistillation Scenario=Multi-Teacher Distillation, Model Scale=4B2026.0586.3——
ExOPDDistillation Scenario=Multi-Teacher Distillation, Model Scale=4B2026.0586.3——
TeacherTraining Strategy=RL2026.0585.2——
ExPODistillation Scenario=Single-Teacher Distillation, Model Scale=4B2026.0584.8——
OPDDistillation Scenario=Single-Teacher Distillation, Model Scale=4B2026.0582.6——
TraceLiftModel=Qwen3-4B2026.0578.66——
StudentModel Scale=4B2026.0577.4——
Exec-onlyModel=Qwen3-4B2026.0576.83——
BaseModel=Qwen3-4B2026.0575.61——
AZRBackbone=Qwen3-4B2026.0575——
COSEBackbone=Qwen3-4B2026.0573.8——
R-ZeroBackbone=Qwen3-4B2026.0573.2——
BaseBackbone=Qwen3-4B2026.0573.17——
R1-OracleBackbone=Qwen2.5-3B-Instruct, Supervision Paradigm=100% GT2026.0371.7979.120
R1-pBackbone=Qwen2.5-3B-Instruct, Supervision Paradigm=20% GT2026.0364.175.67-3.45
TraceLiftModel=Qwen2.5-7B2026.0564.02——
Exec-onlyModel=Qwen2.5-7B2026.0561.59——
MemRewardBackbone=Qwen2.5-3B-Instruct, Supervision Paradigm=20% GT + GNN2026.0361.5477.02-2.1
BaseModel=Qwen2.5-7B2026.0560.37——
R1-OracleBackbone=Qwen2.5-1.5B-Instruct, Supervision Paradigm=100% GT2026.0353.8570.470
AZRBackbone=Llama-3.2-3B-Instruct2026.0552.4——
BaseBackbone=Llama-3.2-3B-Instruct2026.0550.61——
R-ZeroBackbone=Llama-3.2-3B-Instruct2026.0550.6——
COSEBackbone=Llama-3.2-3B-Instruct2026.0549.4——
MemRewardBackbone=Qwen2.5-1.5B-Instruct, Supervision Paradigm=20% GT + GNN2026.0343.5968.1-2.37
R1-pBackbone=Qwen2.5-1.5B-Instruct, Supervision Paradigm=20% GT2026.0338.4662.72-7.75
TraceLiftModel=Llama3.1-8B2026.0532.93——
Exec-onlyModel=Llama3.1-8B2026.0531.1——
BaseModel=Llama3.1-8B2026.0529.88——
ckpt_-1SFT Status=Pre-SFT, Few-shot settings=0-shot2026.05—68.29—
ckpt_-1SFT Status=Pre-SFT, Few-shot settings=3-shot2026.05—0.61—
MONAOptimizer=MONA, SFT Status=Post-SFT, Few-shot settings=0-shot2026.05—88.41—
MONAOptimizer=MONA, SFT Status=Post-SFT, Few-shot settings=3-shot2026.05—86.59—
MuonOptimizer=Muon, SFT Status=Post-SFT, Few-shot settings=0-shot2026.05—87.2—
MuonOptimizer=Muon, SFT Status=Post-SFT, Few-shot settings=3-shot2026.05—85.98—