Salvato in:
| Autori principali: | , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | https://arxiv.org/abs/2605.30851 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866916064425672704 |
|---|---|
| author | He, Minghua Zhang, Lingzhe Liu, Yuan Zhou, Xiao Liu, Aiwei |
| author_facet | He, Minghua Zhang, Lingzhe Liu, Yuan Zhou, Xiao Liu, Aiwei |
| contents | Parallel decoding improves generation efficiency by processing multiple decode positions within a single decode forward, but reported speedups conflate algorithmic token utilization with the system cost of executing multiple positions. We isolate the system side by introducing Near-Free Parallelism (NFP), the maximum number of positions executable at near-free latency. Analyzing Dense FFNs, MoE FFNs, and Attention against an idle-compute baseline, we find that NFP is shaped not by memory-bound resource slack alone, but also by implementation-induced kernel-granularity slack. Based on these mechanisms, we establish a Near-Free Parallelism principle that predicts the NFP boundary from hardware balance and implementation granularity. Validation on representative Dense and MoE models -- spanning both diffusion and autoregressive decoding -- shows that the principle accurately predicts practical NFP boundaries, revealing that the standard idle-compute intuition can over-predict by up to 23x -- offering a system-side budget for parallelism selection and model-system co-design. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_30851 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | How Much Parallelism Is "Free"? A Principle of Near-Free Parallelism for Parallel Decoding He, Minghua Zhang, Lingzhe Liu, Yuan Zhou, Xiao Liu, Aiwei Performance Parallel decoding improves generation efficiency by processing multiple decode positions within a single decode forward, but reported speedups conflate algorithmic token utilization with the system cost of executing multiple positions. We isolate the system side by introducing Near-Free Parallelism (NFP), the maximum number of positions executable at near-free latency. Analyzing Dense FFNs, MoE FFNs, and Attention against an idle-compute baseline, we find that NFP is shaped not by memory-bound resource slack alone, but also by implementation-induced kernel-granularity slack. Based on these mechanisms, we establish a Near-Free Parallelism principle that predicts the NFP boundary from hardware balance and implementation granularity. Validation on representative Dense and MoE models -- spanning both diffusion and autoregressive decoding -- shows that the principle accurately predicts practical NFP boundaries, revealing that the standard idle-compute intuition can over-predict by up to 23x -- offering a system-side budget for parallelism selection and model-system co-design. |
| title | How Much Parallelism Is "Free"? A Principle of Near-Free Parallelism for Parallel Decoding |
| topic | Performance |
| url | https://arxiv.org/abs/2605.30851 |