aiqbench analysis
Fast-dLLM++ introduces a training-free decoding strategy that leverages heterogeneous confidence profiles to unlock additional parallelism in diffusion large language models. By moving beyond the homogeneous assumptions of its predecessor, this method achieves up to 37% higher throughput without altering the underlying model or cache implementation. The approach effectively bridges theoretical profiling with empirical gains on standard reasoning and coding benchmarks.
Fast-dLLM++: Fr\'{e}chet Profile Decoding for Faster Diffusion LLM Inference
arXiv:2606.02955v2 Announce Type: replace Abstract: Diffusion large language models promise parallel token generation, yet inference remains bottlenecked by deciding which masked tokens can be safely committed together. Fast-dLLM addressed this with KV caching and confidence-guided parallel decoding, but its decoding theory uses a homogeneous high-confidence assumption that effectively reduces each candidate set to its weakest selected token. We argue that this leaves speed on the table because real decoding steps exhibit heterogeneous confidence profiles. We propose \textbf{Fast-dLLM++}, a training-free extension that introduces \emph{Fr\'{e}chet profile decoding}: selecting parallel commit sets from the full sorted confidence profile rather than a single worst-case confidence. The resulting rule is a heterogeneous-confidence generalization of Fast-dLLM's factor selector and it recovers the previous rule exactly in the equal-confidence case and adds a provable \emph{heterogeneity bonus} when the selected tokens have uneven confidences. Fast-dLLM++ leaves the model, diffusion process, and cache implementation entirely unchanged, making it a drop-in replacement for existing Fast-dLLM decoding. Experiments on GSM8K, MATH, HumanEval, and MBPP with the LLaDA-8B model show that the theoretical improvement translates directly into empirical gains: profile-aware selection improves the accuracy--throughput frontier by exploiting safe parallelism that weakest-token rules miss, achieving up to 37\% higher throughput at comparable accuracy. Our code release is at https://github.com/Ringo-Star/FastdLLM_plusplus.
Abstract:Diffusion large language models promise parallel token generation, yet inference remains bottlenecked by deciding which masked tokens can be safely committed together. Fast-dLLM addressed this with KV caching and confidence-guided parallel decoding, but its decoding theory uses a homogeneous high-confidence assumption that effectively reduces each candidate set to its weakest selected token. We argue that this leaves speed on the table because real decoding steps exhibit heterogeneous confidence profiles. We propose \textbf{Fast-dLLM++}, a training-free extension that introduces \emph{Fréchet profile decoding}: selecting parallel commit sets from the full sorted confidence profile rather than a single worst-case confidence. The resulting rule is a heterogeneous-confidence generalization of Fast-dLLM's factor selector and it recovers the previous rule exactly in the equal-confidence case and adds a provable \emph{heterogeneity bonus} when the selected tokens have uneven confidences. Fast-dLLM++ leaves the model, diffusion process, and cache implementation entirely unchanged, making it a drop-in replacement for existing Fast-dLLM decoding. Experiments on GSM8K, MATH, HumanEval, and MBPP with the LLaDA-8B model show that the theoretical improvement translates directly into empirical gains: profile-aware selection improves the accuracy--throughput frontier by exploiting safe parallelism that weakest-token rules miss, achieving up to 37\% higher throughput at comparable accuracy. Our code release is at this https URL.
Initial version accepted at Workshop on Structured Probabilistic Inference & Generative Modeling, ICML 2026. Project Page: this https URL
Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
arXiv:2606.02955 [cs.CL]
(or arXiv:2606.02955v2 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2606.02955 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 321 words
Related coverage
Fast-dLLM++: Fr\'{e}chet Profile Decoding for Faster Diffusion LLM Inference
arXiv · Jun 3, 2026
Flash-dLLM: IO-Aware KV Caching and Parallel Decoding for Fast, Memory-Efficient Diffusion LLMs
arXiv · Sep 23, 2026
Prefilling-dLLM v2: Learned Chunk Routing for Long-Context Diffusion Language Models
arXiv · Oct 14, 2026
Mask-Guided KV Cache Eviction in Block Diffusion Language Models
arXiv · Oct 7, 2026