v0.2.7

What's Changed

• update: always compute and report grad_norm by @jcao-ai in #193
• feat: add fullly on-policy support for GRPO by @jcao-ai in #196
• VAPO: Positive example lm loss by @xlu451 in #195
• Bump python version to 3.12 in Docker by @bddppq in #198
• minor update: rename fully_on_policy->on_policy by @jcao-ai in #200
• feat: support hf model revision by @jcao-ai in #199
• print env preset for users by @foreverlms in #201
• optimize: GRPO logits memory by @jcao-ai in #203
• fix: Fix minor issue in slurm launch by @lfengad in #205
• [Under Test] support dp balanced token-mean GRPO loss by @jcao-ai in #204
• feat: support lr decay scheduler by @jcao-ai in #206

Full Changelog: v0.2.6...v0.2.7