blog: Advanced CUDA Graph Techniques in Inference - #399

Merged

LyH88 merged 17 commits into

lm-sys:mainfrom

Oasis-Git:blog/advanced-cuda-graph

Aug 17, 2026

Commits on Aug 17, 2026

blog: Advanced CUDA Graph Techniques in Inference

Oasis-Git
and
claude
committed
2d6bf87
blog: add Baizhou Zhang, Yusheng Su, Ke Bao to acknowledgments

Oasis-Git
and
claude
committed
db55d87
blog: thank Thinking Machines Lab in acknowledgments

Oasis-Git
and
claude
committed
a7bc6b2
blog: use gpt-oss and GLM-5.2 for the prefill memory figure

Oasis-Git
and
claude
committed
f2f2628
blog: scale down the figures

Oasis-Git
and
claude
committed
84679d3
blog: use a consistent "text #PR" form for all references

Oasis-Git
and
claude
committed
1d17080
blog: fix broken diffusion figure (missing xmlns)

Oasis-Git
and
claude
committed
ccf4215
blog: drop the diffusion SVG's inline root style

Oasis-Git
and
claude
committed
45412ca
blog: add the prefill-only latency comparison

Oasis-Git
and
claude
committed
2a0e469
blog: keep only gpt-oss in the prefill latency figure

Oasis-Git
and
claude
committed
ec63827
blog: drop the footnote line from the prefill latency figure

Oasis-Git
and
claude
committed
01bea74
blog: drop the end-of-line labels from the prefill figure

Oasis-Git
and
claude
committed
e03f6ce
blog: summarise the results in the TL;DR

Oasis-Git
and
claude
committed
2666d63
blog: reconcile the TL;DR build times with the coverage row

Oasis-Git
and
claude
committed
582c024
blog: date to Aug 17, TL;DR summary paragraph, prefill benchmark section

Oasis-Git
and
claude
committed
2e22d57
blog: bracket the PR references, tighten two passages

Oasis-Git
and
claude
committed
e192705
blog: fix an antecedent in the TL;DR, drop a duplicate disclaimer

Oasis-Git
and
claude
committed
1f7fdcc

Navigation Menu

Uh oh!

blog: Advanced CUDA Graph Techniques in Inference - #399

blog: Advanced CUDA Graph Techniques in Inference#399
LyH88 merged 17 commits into
lm-sys:mainfrom
Oasis-Git:blog/advanced-cuda-graph

Commits on Aug 17, 2026

blog: Advanced CUDA Graph Techniques in Inference

blog: add Baizhou Zhang, Yusheng Su, Ke Bao to acknowledgments

blog: thank Thinking Machines Lab in acknowledgments

blog: use gpt-oss and GLM-5.2 for the prefill memory figure

blog: scale down the figures

blog: use a consistent "text #PR" form for all references

blog: fix broken diffusion figure (missing xmlns)

blog: drop the diffusion SVG's inline root style

blog: add the prefill-only latency comparison

blog: keep only gpt-oss in the prefill latency figure

blog: drop the footnote line from the prefill latency figure

blog: drop the end-of-line labels from the prefill figure

blog: summarise the results in the TL;DR

blog: reconcile the TL;DR build times with the coverage row

blog: date to Aug 17, TL;DR summary paragraph, prefill benchmark section

blog: bracket the PR references, tighten two passages

blog: fix an antecedent in the TL;DR, drop a duplicate disclaimer