Skip to content
AI News HubLIVE
Original source2 min read

Optimize the Cob, Then the Kernel: Optimizing Beam Search for Generative Recommendation Inference

Summary

An arXiv paper argues that general-purpose LLM serving engines such as vLLM waste effort on generative recommendation inference, where beams redundantly re-read the shared prompt KV cache and each beam-search step is a separate engine invocation. Using Tree Attention plus compiler- and scheduling-level optimizations, the authors cut end-to-end latency by up to 25x over vLLM on Ascend NPUs without custom kernels, matching a state-of-the-art custom-kernel solution.

SourcearXiv Machine LearningAuthor: Shing Wai Pun, Kai Wu, Pengfei Xia, Hui Zang, Wei Guo, Yong Liu
Optimize the Cob, Then the Kernel: Optimizing Beam Search for Generative Recommendation Inference
Report an error

The correction channel is not available yet. You can copy the article reference below for later.

Correction instructions
Read article

[Submitted on 17 Nov 2026]

Title:Optimize the Cob, Then the Kernel: Optimizing Beam Search for Generative Recommendation Inference

View a PDF of the paper titled Optimize the Cob, Then the Kernel: Optimizing Beam Search for Generative Recommendation Inference, by Shing Wai Pun and 5 other authors

View PDF HTML (experimental)

Abstract:Generative recommendation (GR) inference differs substantially from conventional LLM serving, featuring long input sequences, wide beam search, and only a few fixed output tokens. Existing LLM serving engines such as vLLM process beams as batched independent sequences, causing redundant KV cache accesses over the shared prompt prefix. Furthermore, vLLM executes each beam search generation step as a separate engine invocation, incurring host-side beam processing overhead between decoding steps and substantial runtime KV block management in PagedAttention. Using Tree Attention, we concatenate beams along the sequence dimension to eliminate redundant KV accesses, and further optimize execution through pre-generated partial-attention masks, static KV cache allocation, loop-unrolled generation dispatch, TorchAir graph compilation, automatic operator fusion, and vocabulary-constrained decoding. Without the use of custom kernels, our approach reduces end-to-end latency by up to 25x over vLLM on Ascend NPUs, and achieves generation latency comparable to a SOTA solution using custom kernels, showing that the generation pipeline should be optimized holistically before resorting to custom kernel development.

Comments: 8 pages, 4 figures

Subjects:

Machine Learning (cs.LG); Artificial Intelligence (cs.AI)

Cite as: arXiv:2611.23339 [cs.LG]

(or arXiv:2611.23339v1 [cs.LG] for this version)

https://doi.org/10.48550/arXiv.2611.23339

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Shing Wai Pun [view email] [v1] Tue, 17 Nov 2026 07:14:52 UTC (498 KB)

Full-text links:

Access Paper:

View a PDF of the paper titled Optimize the Cob, Then the Kernel: Optimizing Beam Search for Generative Recommendation Inference, by Shing Wai Pun and 5 other authors

View PDF

HTML (experimental)

TeX Source

view license

Additional Features

Audio Summary

Current browse context:

cs.LG

new | recent | 2026-11

Change to browse by:

cs cs.AI

References & Citations

SciX

Google Scholar

Semantic Scholar

Loading...

Data provided by:

Bibliographic Tools

Bibliographic and Citation Tools

Bibliographic Explorer Toggle

Bibliographic Explorer (What is the Explorer?)

Connected Papers Toggle

Connected Papers (What is Connected Papers?)

Litmaps Toggle

Litmaps (What is Litmaps?)

scite.ai Toggle

scite Smart Citations (What are Smart Citations?)

Code, Data, Media

Code, Data and Media Associated with this Article

alphaXiv Toggle

alphaXiv (What is alphaXiv?)

Links to Code Toggle

CatalyzeX Code Finder for Papers (What is CatalyzeX?)

DagsHub Toggle

DagsHub (What is DagsHub?)

GotitPub Toggle

Gotit.pub (What is GotitPub?)

Huggingface Toggle

Hugging Face (What is Huggingface?)

ScienceCast Toggle

ScienceCast (What is ScienceCast?)

Demos

Demos

Replicate Toggle

Replicate (What is Replicate?)

Spaces Toggle

Hugging Face Spaces (What is Spaces?)

Spaces Toggle

TXYZ.AI (What is TXYZ.AI?)

Related Papers

Recommenders and Search Tools

Link to Influence Flower

Influence Flower (What are Influence Flowers?)

Core recommender toggle

CORE Recommender (What is CORE?)

IArxiv recommender toggle

IArxiv Recommender (What is IArxiv?)

Author

Venue

Institution

Topic

About arXivLabs

arXivLabs: experimental projects with community collaborators

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.

Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)

Key points and analysis

Article intelligence

EngineersAdvanced

Key points

  • Generative recommendation inference differs from standard LLM serving: long inputs, wide beams, and only a few fixed output tokens.
  • Tree Attention concatenates beams along the sequence dimension to eliminate redundant KV cache accesses over the shared prompt prefix.
  • Pre-generated partial-attention masks, static KV cache allocation, loop-unrolled dispatch, TorchAir graph compilation, operator fusion, and vocabulary-constrained decoding remove further overhead.
  • The approach reduces end-to-end latency by up to 25x over vLLM on Ascend NPUs, and the authors conclude the generation pipeline should be optimized holistically before building custom kernels.

Highlights and analysis are generated automatically and may contain errors. Check the original source.