FreeLLMAjuste fino por RL aumenta o sucesso repetível do Qwen3.8‑27B no ThinkingBox‑Bench
Após 50 etapas GRPO, o observado‑20/20 do Qwen3.8‑27B subiu de 7.5% para 21.1% no benchmark de 507‑tarefas.
Microsoft and Hugging Face released ThinkingBox‑Training, a GitHub tool that turns executable database checks into rewards for post‑training agents.
They selected 507 tasks from ThinkingBox‑Bench, each providing a real backend and tool set, then built an offline policy‑relative difficulty pool for training.
Fine‑tuning Qwen3.8‑27B with GRPO for 50 steps raised pass@1 from 51.70% to 60.78% and increased tasks solved on every attempt (observed‑20/20) from 38 to 107.
The improvement came with a slight drop in pass@20 (tasks solved at least once) from 453 to 446, indicating higher repeatability but marginally less breadth.
Domain‑level results show gains across all five categories, with the largest observed‑20/20 increase in retail (+24.5 points) and the biggest pass@20 decline in auto insurance (‑7 points).
O que isso significa para você
The higher repeatable success makes agents more reliable for production database operations, reducing risky “done” false positives. For users preferring free or locally run models, the results highlight that larger models like Qwen3.8‑27B can become more dependable through targeted RL fine‑tuning, though broader coverage may require additional strategies.
Fontes
- Turning Database Grading into RL Rewards: Post-Training Agents with ThinkingBox, Hugging Face (official), 22 Oct 2026
Escrito automaticamente a partir das fontes acima, depois verificado pelo nosso código: todo número e nome deve aparecer nelas, e nenhuma frase pode copiar uma fonte. Encontrou um erro? Reporte‑o.
Qual IA você pode usar grátis hoje?
Todo modelo de IA gratuito, classificado por votos humanos cegos, com a memória e velocidade que ele precisa na sua própria GPU.