List your company

Chess Reasoning Quality Evaluation on Lichess Puzzle Database (held-out positions)

0.218WR MAEVPS
0.197440.336220.4750.61378Apr 3, 2026

Evaluation results

MethodLinks
VPSBackbone=Qwen3-8B2026.040.21837.697.8
VPSBackbone=DeepSeek-R1-Distill-Llama-8B2026.040.28636.298.5
SFT onlyBackbone=Qwen3-8B2026.040.3132.396.1
SFT onlyBackbone=DeepSeek-R1-Distill-Llama-8B2026.040.34634.197.8
SFT + GRPOBackbone=Qwen3-8B2026.040.45218.372.1
SFT + GRPOBackbone=DeepSeek-R1-Distill-Llama-8B2026.040.7323331.2