Spork round 2: GRPO warmup

Qwen3.5-9B · 400 held-out puzzles (rating 800–1199) × 4 samples

Spork round 2: GRPO warmup010203040506070ScoreSolved (pass@1)Mean reward3.544.6947.8850.887.3751.9356.4358.46
Base Qwen3.5-9BRound 1 SporkRound 2 Spork with warmupRound 2 Spork, no warmup
View data table
SeriesCategoryScoreInterval / statistics
Base Qwen3.5-9BSolved (pass@1)3.5
Round 1 SporkSolved (pass@1)44.69
Round 2 Spork with warmupSolved (pass@1)47.88
Round 2 Spork, no warmupSolved (pass@1)50.88
Base Qwen3.5-9BMean reward7.37
Round 1 SporkMean reward51.93
Round 2 Spork with warmupMean reward56.43
Round 2 Spork, no warmupMean reward58.46