AI Paper Atlas

Research map / Image generation / Fast diffusion sampling

From Bits to Rounds: Parallel Decoding with Exploration for Diffusion Language Models

Hengyu Fu, Baihe Huang, Virginia Adams, Charles Wang, Junkeun Yi, Mohammad Mahdi Kamani, Venkat Krishna Srinivasan, Jiantao Jiao

ICML 2026 · Image generation › Fast diffusion sampling · approach: Masked diffusion language model decoding · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add cross-block parallel decoding and look-ahead beam-search exploration (ETE) to confidence-based decodingLLaDA-8B-Instruct · GSM8K, HumanEval, MATH, MMLU-Proinference steps at fixed accuracy: over 30% fewer steps at almost all accuracy levelsImproved
add ETE with KV caching to confidence-based decodingLLaDA-8B-Instruct with Fast-DLLM · GSM8Kthroughput at comparable accuracy: 50% higher throughputImproved
add ETE with KV caching to confidence-based decodingLLaDA-8B-Instruct with Fast-DLLM · MMLU-Prothroughput at comparable accuracy: up to 70% higher throughputImproved
add ETE with KV caching to confidence-based decodingLLaDA-8B-Instruct with Fast-DLLM · MATHthroughput at comparable accuracy: 30% higher throughputImproved
add ETE with KV caching to confidence-based decodingLLaDA-8B-Instruct with Fast-DLLM · GSM8Kpeak accuracy: 0.779 -> 0.797Improved
add ETE with KV caching to confidence-based decodingLLaDA-8B-Instruct with Fast-DLLM · MMLU-Propeak accuracy: 0.352 -> 0.386Improved
lower the confidence factor f in dynamic confidence decoding (stricter threshold) (ablation)dynamic confidence decoding · GSM8K (200 samples)decoding roundsHurt
batch the beam-search exploration candidates with shared KV cache (ablation)ETE · LLaDA-8B-Instruct forward latencywall-clock time per batched forward: free lunch up to about 2 on H100, 4 on H200, 7 on B200No clear effect
exploration (low-confidence token commits) versus exploitation rounds in confidence decodingdynamic confidence decoding · GSM8K (200 samples)share of information per share of rounds: exploration 23.5% of rounds, 66.7% of bits at f=0.4Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy