A large-scale empirical study from arXiv (2608.13698v1) finds that training language models to reason in their native language via Group Relative Policy Optimization (GRPO) leaves only a small gap to …
A new fine-tuning method called 'Advice String Distillation' is proposed as a safer alternative to RLVR for training AI models, using context distillation to update weights with text-associated change…
A new arXiv preprint (2608.02867v1) from researchers studying reinforcement learning with verifiable rewards (RLVR) finds that RLVR-trained large language models (LLMs) exhibit reduced semantic branch…
Researchers propose RSTG (Recovering Learning Signals via Adaptive Teacher Guidance), a method that selectively applies on-policy distillation to negative zero-variance prompts in reinforcement learni…
A developer highlights the top 10 papers on Hugging Face, covering GUI agents, LLM self-improvement via RL, parameterized long-term memory, RAG, multimodal robots, 3D generation, and role-play benchma…
Researchers propose SAF, a Stable Advantage Fusion framework that combines reinforcement learning with verifiable rewards (RLVR) and on-policy distillation (OPD) for training language models, addressi…
Researchers introduced PPO-HSC (Proximal Policy Optimization with High-order Sampling Coverage), an exploratory reinforcement learning framework that addresses mode collapse in Large Language Model fi…
Researchers formalized medical diagnosis as an iterative evidence-seeking task and used reinforcement learning with verifiable rewards to train large language models to autonomously gather clinical ev…
Researchers propose a reinforcement learning with verifiable rewards (RLVR) method for data-efficient adaptation of audio-language models to code-switched automatic speech recognition (ASR). Using Qwe…
Researchers propose Tandem Reinforcement Learning (TRL), extending the tandem training paradigm to reinforcement learning with verifiable rewards (RLVR). Training Qwen3-4B-Instruct on competition math…
Reinforcement Learning with Verifiable Rewards (RLVR) allows language models to bootstrap beyond human-level capabilities on exactly graded problems like coding and formal proofs, but alignment-flavor…
Researchers at ACL 2026 found that entropy collapse in LLM reasoning, which undermines test-time scaling, is driven by premature overconfidence at a small set of critical tokens. They proposed SCOPE, …
NVIDIA released Nemotron 3 Ultra, a 550B-parameter hybrid Mamba-Transformer model with 55B active parameters, achieving up to 6x higher inference throughput than state-of-the-art LLMs while maintainin…
Researchers introduced SDPG, a self-distilled policy-gradient framework that combines group-relative verifier advantages with normalized standard deviation and full-vocabulary on-policy self-distillat…
PopuLoRA is a method for training large language models (LLMs) that uses co-evolving populations of teacher and student adapters to generate and solve verifiable reasoning tasks, such as code and math…