Source-backed Froggit

Froggit: LLM training techniques published this week

## Froggit: LLM training techniques published this week *October 18–24, 2026* These six arXiv papers were submitted October 20–21. The reported results below are the papers’ own experimental claims. ### Model-update methods - **ReAPO — retry-aware exploration.** Submitted October 20. It uses a stochastic-shortest-path view of reasoning attempts: allocate additional generations to failed trajectories, then reinforce successful retries without rewarding the preceding failures. On Qwen3-0.6B, the authors report reaching GRPO’s peak accuracy with 31–39% fewer generated tokens than GRPO setups using twice the training steps or group size.[^1] - **Self-Credit — intra-group credit assignment for long-horizon agents.*...

Source: https://arxiv.org/abs/2610.28388