List your company
ResearchFormerly WizWand

Code Generation on LiveCodeBench v1 (test)

34.9AccuracyMarginal Sharpening
4.84412.64720.4528.253Jul 15, 2024Nov 5, 2024Feb 27, 2025Jun 20, 2025Oct 12, 2025Feb 2, 2026May 27, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Marginal SharpeningModel=Qwen3-8B, K=4, S=82026.0534.9
Marginal SharpeningModel=Qwen3-8B, K=32, S=12026.0534.3
vanilla token dropoutBackbone=Qwen-0.6B2025.1231.9
Marginal SharpeningModel=Qwen3-4B, K=4, S=82026.0530.9
Majority VotingModel=Qwen3-8B, k=322026.0530.9
EntroDropBackbone=Qwen-0.6B2025.1230.5
Power SamplingModel=Qwen3-4B2026.0530.3
Marginal SharpeningModel=Qwen3-4B, K=32, S=12026.0530.3
Temperature SamplingModel=Qwen3-8B2026.0529.6
Marginal SharpeningModel=Qwen3-1.7B, K=32, S=12026.0528
Majority VotingModel=Qwen3-4B, k=322026.0527.4
Marginal SharpeningModel=Qwen3-1.7B, K=4, S=82026.0526.9
Qwen2-7BInstruction-tuned=true2024.0726.6
Temperature SamplingModel=Qwen3-4B2026.0526.6
NEFTuneBackbone=Qwen-0.6B2025.1226.3
weight decayBackbone=Qwen-0.6B2025.1225.2
Power SamplingModel=Qwen3-1.7B2026.0523.4
hidden dropoutBackbone=Qwen-0.6B2025.1223
Majority VotingModel=Qwen3-1.7B, k=322026.0522.9
baselineBackbone=Qwen-0.6B2025.1222.7
Temperature SamplingModel=Qwen3-1.7B2026.0522.7
Llama-3-8BInstruction-tuned=true2024.0717.3
Qwen1.5-7BInstruction-tuned=true2024.076