Grpo
PulseAugur coverage of Grpo — every cluster mentioning Grpo across labs, papers, and developer communities, ranked by signal.
24 day(s) with sentiment data
GRPO and its variants (HölderPO, GROW) are central to recent LLM policy optimization research
Multiple recent clusters highlight GRPO and its derivatives (HölderPO, GROW) as key advancements in LLM policy optimization. This indicates a strong research trend focusing on refining reinforcement learning techniques for LLMs, particularly in areas like multi-agent interaction, handling complex reward structures, and improving stability and adaptability in diverse tasks.
GROW framework to see adoption for VLM agent development beyond Minecraft
The GROW framework, leveraging adapted GRPO, has shown state-of-the-art performance on over 800 Minecraft tasks for VLM agents. This success in a complex, open-world environment suggests potential for broader application in other VLM agent development scenarios, such as robotics, simulation, or other interactive environments where multi-turn learning and handling long contexts are critical.
GRPO to be integrated into Anyscale's LLM post-training automation
The recent Anyscale Agent Skill launch focuses on automating LLM post-training runs, while another cluster details GRPO's use in multi-agent LLM deferral to humans. Given GRPO's demonstrated ability to incorporate human expertise and Anyscale's push for automation, it's plausible GRPO will be integrated as a method within Anyscale's automated post-training workflows to enhance human-in-the-loop capabilities.
-
AI alignment research tackles reward hacking with new techniques
Researchers are exploring methods to prevent AI models from exploiting reward functions, a phenomenon known as reward hacking. One approach involves using steering vectors to guide gradient routing, aiming to isolate un…
-
New method uses wrong drafts to boost LLM math capabilities
Researchers have developed a novel technique called "Weak-to-Strong Elicitation via Mismatched Wrong Drafts" to improve the capabilities of large language models. This method involves using mathematically incorrect draf…
-
New RLAIF framework improves job search query generation
Researchers have developed a novel RLAIF framework to generate portable job search queries, aiming to better capture candidate qualifications beyond simple keyword matching. The study highlights the critical role of rob…
-
New Intent-Aware Training Boosts LLM Safety Classifiers
Researchers have developed a new method for improving the safety classification of large language models by explicitly modeling user intent. They introduced AIMS, a dataset of 1,724 safety prompts with associated intent…
-
New AI Framework Enhances Role-Playing Agents with Psychology-Grounded Reasoning
Researchers have introduced Psy-CoT, a novel framework designed to enhance the role-playing capabilities of AI agents. This method grounds reasoning in psychological principles, breaking down character portrayal into in…
-
WinDOM paper details small-model GUI grounding with automated data and SFD training
Researchers have introduced WinDOM, a new method for grounding small GUI-agent models, focusing on efficient data acquisition and training techniques. The approach utilizes a large corpus of $54,425$ GUI interaction rec…
-
New ASSCG system optimizes LLM use for autonomous driving planning
Researchers have developed a new system called ASSCG to optimize the use of large language models (LLMs) in autonomous driving planning. ASSCG acts as a gatekeeper, making frame-level decisions to refresh, reuse, or sup…
-
New SR-PPO method improves RL for language models with single rollout
Researchers have developed a new method called Single-Rollout Proximal Policy Optimization (SR-PPO) to address the challenges of estimating token-level advantages in reinforcement learning for language models. This appr…
-
VibeThinker 3B model surpasses Opus 4.5 in reasoning with novel SFT+GRPO
A new 3-billion parameter model named VibeThinker has demonstrated superior reasoning capabilities compared to Anthropic's Opus 4.5. This performance was achieved using a novel combination of supervised fine-tuning (SFT…
-
New BALTO framework precisely targets LLM hallucinations at token level
Researchers from Shanghai Jiao Tong University and Tencent have developed BALTO, a novel reinforcement learning framework designed to precisely eliminate hallucinations in large language models (LLMs). The framework ope…
-
LLMs fail to reliably self-report adversarial prefill attacks, study finds
A new study published on arXiv investigates the ability of large language models (LLMs) to self-report when they have been influenced by adversarial prefill attacks. The research found that across ten different open-wei…
-
SPIRAL framework enhances language model reasoning with parallel and aggregated traces
Researchers have developed SPIRAL, a new framework designed to enhance language model reasoning capabilities by integrating sequential, parallel, and aggregation methods. Unlike traditional models optimized solely for s…
-
New CFPO framework enhances multimodal reasoning in LVLMs
Researchers have introduced CounterFactual Policy Optimization (CFPO), a new framework designed to improve multimodal reasoning in Large Vision-Language Models (LVLMs). CFPO addresses grounding failures and hallucinatio…
-
New unified vision-language model ABACUS excels at object counting
Researchers have developed ABACUS, a unified vision-language model designed for object counting and related tasks. This model leverages a 3B-parameter foundation model and incorporates novel techniques such as density-a…
-
New framework DR-MV3D enhances 3D visual question answering with dense rewards
Researchers have introduced DR-MV3D, a novel framework designed to enhance multi-view 3D visual question answering (MV3D-VQA). This approach utilizes dense, verifiable rewards to supervise the reasoning process, moving …
-
New ACOER method stabilizes LLM training for efficient reasoning
Researchers have developed a new method called ACOER (Adaptive Correct-Only Efficiency Reward) to stabilize the training of large language models for efficient reasoning. Existing methods like GRPO (Group Relative Polic…
-
AI agent training hampered by routine actions, new paper finds
A new research paper titled "Drowning in Routine: Signal Dilution in Multi-Turn Agent Training" explores the challenges of training multi-turn AI agents. The paper identifies that when agents perform many routine, non-c…
-
New AAPA framework improves LLM alignment with adversarial anchoring
Researchers have introduced AAPA, a novel framework designed to enhance the post-training alignment of large language models. This plug-in framework augments existing training objectives with an adversarial anchoring si…
-
New method enhances LLM alignment by modeling reward uncertainty
Researchers have developed a new method called Uncertainty-Aware Reward Modeling (UARM) to improve the stability of reinforcement learning from human feedback (RLHF) in large language models. Traditional RLHF methods st…
-
New Agentic Data Tailoring paradigm structures multimodal streams
Researchers have introduced a new paradigm called Agentic Data Tailoring, which uses learnable data processing to structure high-entropy multimodal streams. The DataClaw_0-9B model, trained using supervised fine-tuning …