#سيف_الباري من #بنغلاديش الفائز بجائزة #مبتكرون_دون35 العالمية في نسختها السابعة لعام 2024، وذلك لابتكاره "علّام (نموذج اللغة العربية الكبير)، نموذج ذكاء اصطناعي متقدم يحافظ على الفروق اللغوية والثقافية للغة العربية، مما يتيح تطبيقات شاملة في مجالات التعليم والرعاية الصحية والخدمات الحكومية". @sbmaruf

AI dot com redirects to DeepSeek 🐋

Really cool paper! It studies how LLMs develop extended CoT reasoning, focusing on RL and compute scaling. Key insights include: SFT boosts performance – While not strictly necessary, SFT simplifies training and increases efficiency. Models fine-tuned with long CoT data achieve higher accuracy than those using short CoT sequences. Reward shaping is crucial for stable RL – They find that naive RL approaches don’t always extend CoT length effectively. To address this, it introduces a cosine length-scaling reward with repetition penalties, which balances reasoning depth and prevents meaningless length increases. Scaling verifiable reward signals – RL models trained with noisy, web-extracted “silver” supervision signals can generalize better to OOD tasks, such as STEM reasoning. Filtering such data is crucial to maintaining training stability. Emergent reasoning abilities in base models – Skills like error correction and backtracking exist in base models but require careful RL incentives to be effectively utilized in complex tasks. This paper provides a good roadmap for folks looking to refine CoT training strategies for LLMs, highlighting how RL and reward tuning impact reasoning depth.