Create accountLogin

Recent searches

No recent searches

Search options

Only available when logged in.
mamot.fr is one of the many independent Mastodon servers you can use to participate in the fediverse.
Mamot.fr est un serveur Mastodon francophone, géré par La Quadrature du Net.

Administered by:

Server stats:

3.3K
active users

mamot.fr: AboutStatusProfiles directoryPrivacy policy

Mastodon: AboutGet the appKeyboard shortcutsView source codev4.3.4

#grm

2 posts2 participants0 posts today
Public
Erik Jonker

Interesting pre-paper on Inference-Time Scaling for Generalist Reward Modeling, the basis for the soon to be released new DeepSeek models I think.
#ai #deepseek #llm #grm #SPCT
arxiv.org/abs/2504.02495

arXiv logo
arXiv.orgInference-Time Scaling for Generalist Reward ModelingReinforcement learning (RL) has been widely adopted in post-training for large language models (LLMs) at scale. Recently, the incentivization of reasoning capabilities in LLMs from RL indicates that $\textit{proper learning methods could enable effective inference-time scalability}$. A key challenge of RL is to obtain accurate reward signals for LLMs in various domains beyond verifiable questions or artificial rules. In this work, we investigate how to improve reward modeling (RM) with more inference compute for general queries, i.e. the $\textbf{inference-time scalability of generalist RM}$, and further, how to improve the effectiveness of performance-compute scaling with proper learning methods. For the RM approach, we adopt pointwise generative reward modeling (GRM) to enable flexibility for different input types and potential for inference-time scaling. For the learning method, we propose Self-Principled Critique Tuning (SPCT) to foster scalable reward generation behaviors in GRMs through online RL, to generate principles adaptively and critiques accurately, resulting in $\textbf{DeepSeek-GRM}$ models. Furthermore, for effective inference-time scaling, we use parallel sampling to expand compute usage, and introduce a meta RM to guide voting process for better scaling performance. Empirically, we show that SPCT significantly improves the quality and scalability of GRMs, outperforming existing methods and models in various RM benchmarks without severe biases, and could achieve better performance compared to training-time scaling. DeepSeek-GRM still meets challenges in some tasks, which we believe can be addressed by future efforts in generalist reward systems. The models will be released and open-sourced.
ExploreLive feeds
About