Entity · paper

Mask-Aware Policy Gradients for Diffusion Language Models

paperactiveprovisionalmask-aware-policy-gradients-for-diffusion-language-models-efda64f1·1 events·first seen 2d ago

Aliases: Mask-Aware Policy Gradients for Diffusion Language Models

Co-occurring entities

More like this (12)

Knowledge Editing in Masked Diffusion Language Models Beyond Fully Random Masking: Attention-Guided Denoising and Optimization for Diffusion Language Models Masked Diffusion Models Accelerating Masked Diffusion Large Language Models: A Survey of Efficient Inference Techniques Diffusion Language Models LESS: Mutual-Stability Sampling for Diffusion Language Models Self-Augmenting Retrieval for Diffusion Language Models Evolved Policy Gradients Policy Gradient Methods Selective Disclosure Watermarking for Large Language Models Audio-Native Speech Recognition with a Frozen Discrete-Diffusion Language Model Kolmogorov Regression for Robust Diffusion Policies

Recent events (1)

5arXiv · cs.CL·2d ago·source ↗

Mask-Aware Policy Gradients improve RL training for Masked Diffusion Language Models

A new arXiv preprint introduces a two-stage action MDP formalization for applying reinforcement learning to Masked Diffusion Language Models (MDLMs), decomposing the policy gradient into a token prediction term and a masking order term. Prior approaches ignored the position-unmasking decision, leading to intractable log-likelihood estimates; the proposed method optimizes both terms jointly. The approach achieves 87.1% on GSM8K and 53.4% on MBPP, claiming state-of-the-art results for MDLM-based reasoning and coding.

Evaluation and Benchmarking Alignment and RLHF Diffusion Language Models Mask-Aware Policy Gradients for Diffusion Language Models MBPP +1 more