Understanding R1-Zero-Like Training: A Critical Perspective [pdf] (github.com) 1 points by delifue 1y ago ↗ HN
[–] delifue 1y ago ↗ From: https://x.com/zzlccc/status/1903162768083259703DeepSeek-V3-Base already exhibits "Aha moment" before RL-tuningThe ever-increasing output length in RL-tuning might be due to a BIAS in GRPO
1 comment
[ 2.8 ms ] story [ 10.0 ms ] threadDeepSeek-V3-Base already exhibits "Aha moment" before RL-tuning
The ever-increasing output length in RL-tuning might be due to a BIAS in GRPO