I have written a blogpost offering an explanation of why both the chosen and the rejected log-probability decreases during DPO, and more interestingly, why it is a desired phenomenon to some extent. Link:

RT @yanawei_: 🚀 Check out our ✈️ ICML 2025 work: Perception in Reflection!. A reasonable perception paradigm for LVLMs should be iterative….

RT @_zhihuixie: 🚀 Thrilled to announce Dream-Coder 7B — the most powerful open diffusion code  LLM to date.

RT @PranjalAggarw16: Super excited to see L1 accepted to #COLM2025!. We are further open-sourcing 5 new models & a dataset:.1. L1-7B & L1-….

RT @DanielKhashabi: What’s really going on inside LLMs when they handle non-English queries?. @BafnaNiyati's recent work introduces the **t….

RT @Happylemon56775: Excited to share what I worked on during my time at Meta. - We introduce a Triton-accelerated Transformer with *2-sim….

RT @BafnaNiyati: 📢When LLMs solve tasks with a mid-to-low resource input/target language, their output quality is poor. We know that. But c….

RT @gui_penedo: We have finally released the 📝paper for 🥂FineWeb2, our large multilingual pre-training dataset. Along with general (and ex….

RT @arnal_charles: ❓How to balance negative and positive rewards in off-policy RL❓. In Asymmetric REINFORCE for off-Policy RL, we show that….

RT @nouhadziri: 📢 Can LLMs really reason outside the box in math? Or are they just remixing familiar strategies? . Remember DeepSeek R1, o1….

RT @YLiiiYLiii: 🧠🤖Can LLMs outperform online therapists in single-turn counseling? Meet CounselBench - 100 mental-health professionals rate….

RT @_akhaliq: POLARIS. A POst-training recipe for scaling reinforcement Learning on Advanced ReasonIng modelS. 4B model achieves 81.2% Pass….

RT @natolambert: A common trend across recent research in using reinforcement learning to train reasoning models is that the clipping opera….

RT @aviral_kumar2: Our view on test-time scaling has been to train models to discover algos that enable them to solve harder problems. @se….

RT @alexdmartin314: Talk to me at #CVPR2025 about Multimodal RAG topics! I'll be presenting two papers on video retrieval: Video-ColBERT (l….

RT @fly51fly: [LG] Rewarding the Unlikely: Lifting GRPO Beyond Distribution Sharpening.A He, D Fried, S Welleck [CMU] (2025)..

RT @wellecks: New paper by Andre He:. Rewarding the Unlikely: Lifting GRPO Beyond Distribution Sharpening. Tired o….

RT @jaehunjung_com: Data curation is crucial for LLM reasoning, but how do we know if our dataset is not overfit to one benchmark and gener….

RT @_vaishnavh: 📢 New paper on creativity & multi-token prediction! We design minimal open-ended tasks to argue:. → LLMs are limited in cre….

RT @May_F1_: 🚀 Data to pre-train LLMs on are reaching critical bottleneck. 𝘿𝙤𝙚𝙨 𝙢𝙤𝙙𝙚𝙡-𝙜𝙚𝙣𝙚𝙧𝙖𝙩𝙚𝙙 𝙨𝙮𝙣𝙩𝙝𝙚𝙩𝙞𝙘 𝙙𝙖𝙩𝙖 𝙬𝙤𝙧𝙠 𝙨𝙞𝙢𝙞𝙡𝙖𝙧𝙡𝙮 𝙬𝙚𝙡𝙡 𝙛𝙤𝙧 𝙨𝙘𝙖𝙡….

RT @Mengyue_Yang_: Curious how training data order impacts LLMs without retraining?. Introducing FUT:. 🔍Estimate the effects of any sample….