Embarrassingly Simple Self-Distillation Improves Code Generation
Open paper
Benchmark
Model
Rank
Results
code-generation-on-livecodebench-v5
GPT-OSS-20B (high) + SSD
#7
Pass@1: 70.7
code-generation-on-livecodebench-v6
GPT-OSS-20B (high) + SSD
#26
Pass@1: 74.4
Rank counts only results with a code link.