SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.07223 · ICML · 2026

Vegas: Self-Speculative Decoding with Verification-Guided Sparse Attention

Jian Huang, Yuqi Xue, Yikang Yue

arXiv · PDF · Open in the Atlas

Code that ran

7 of 7 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 0 violated it and 3 have no contract verdict; 2 failures of our instrument; 0 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2602.07223.json · badge: /api/ran/2602.07223.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
platformxlab/vegas canonical (classifier label)
classifier basis: plan_row
7 ran · 0 no recorded run
FunctionStatusWhere it lives
async_request_deepspeed_mii Ran
ran on a synthesized input
platformxlab/vegas/benchmarks/backend_request_func.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from Ravi-VK/SSR, is pointer only (licence: NONE) · get_code("552e4370076a1a01")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from Ravi-VK/SSR, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

async_request_tgi Ran
ran on a synthesized input
platformxlab/vegas/benchmarks/backend_request_func.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from Ravi-VK/SSR, is pointer only (licence: NONE) · get_code("7ee13b8d142453c0")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from Ravi-VK/SSR, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

async_request_trt_llm Ran
ran on a synthesized input
platformxlab/vegas/benchmarks/backend_request_func.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from Ravi-VK/SSR, is pointer only (licence: NONE) · get_code("f0b0f51ad4ed00c9")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from Ravi-VK/SSR, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

builtin_hash Ran
ran, honoured the contract we drafted
platformxlab/vegas/benchmarks/benchmark_hash.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from Ravi-VK/SSR, is pointer only (licence: NONE) · get_code("0d3217ca36de88a0")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from Ravi-VK/SSR, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

check_wheel_size Ran
ran, honoured the contract we drafted
platformxlab/vegas/.buildkite/check-wheel-size.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from Ravi-VK/SSR, is pointer only (licence: NONE) · get_code("c8c099368807b154")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from Ravi-VK/SSR, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

format_model_inspection Ran
ran; our contract draft was wrong, not the code
platformxlab/vegas/vllm/model_inspection.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from Ravi-VK/SSR, is pointer only (licence: NONE) · get_code("a696f84b9b36ab84")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from Ravi-VK/SSR, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

repeat_prompts Ran
ran; our contract draft was wrong, not the code
platformxlab/vegas/benchmarks/benchmark_long_document_qa_throughput.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from Ravi-VK/SSR, is pointer only (licence: NONE) · get_code("5da7378b95e80006")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from Ravi-VK/SSR, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 distinct functions with an identity, 7 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 0 the same month, 7 later) and 0 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
check_wheel_size 4 other papers
3 dated earlier, 1 later
2025.emnlp-main.240 (2025 (year only))
QSPEC: Speculative Decoding with Complementary Quantization Schemes
2512.09472 (2025-12)
WarmServe: Enabling One-for-Many GPU Prewarming for Multi-LLM Serving
2601.20309 (2026-01)
SuperInfer: SLO-Aware Rotary Scheduling and Memory Management for LLM Inference on Superchips
and 1 more not named here
format_model_inspection 4 other papers
2 dated earlier, 2 later
2309.06180 (2023-09)
Efficient Memory Management for Large Language Model Serving with PagedAttention
2601.14327 (2026-01)
Yuan3.0 Ultra: A Trillion-Parameter Enterprise-Oriented MoE LLM
2608.19920 (2026-08)
Learning how to Forget: Fine-tuning for Long-Context Sparse Attention
and 1 more not named here
async_request_deepspeed_mii 2 other papers
2 later
2606.06302 (2026-06)
Tangram: Unlocking Non-Uniform KV Cache Compression for Efficient Multi-turn LLM Serving
2608.20359 (2026-08)
Self-Speculation for Faster Reasoning Models
async_request_tgi 2 other papers
2 later
2606.06302 (2026-06)
Tangram: Unlocking Non-Uniform KV Cache Compression for Efficient Multi-turn LLM Serving
2608.20359 (2026-08)
Self-Speculation for Faster Reasoning Models
async_request_trt_llm 2 other papers
2 later
2606.06302 (2026-06)
Tangram: Unlocking Non-Uniform KV Cache Compression for Efficient Multi-turn LLM Serving
2608.20359 (2026-08)
Self-Speculation for Faster Reasoning Models
builtin_hash 1 other paper
1 later
2608.20359 (2026-08)
Self-Speculation for Faster Reasoning Models
repeat_prompts 1 other paper
1 later
2608.20359 (2026-08)
Self-Speculation for Faster Reasoning Models

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Long-context large language model (LLM) inference has become the norm for today's AI applications. However, it is severely bottlenecked by the increasing memory demands of its KV cache. Previous works have shown that self-speculative decoding with sparse attention, where tokens are drafted using a subset of the KV cache and verified in parallel against the full KV cache, speeds up inference in a lossless manner. However, they rely on a standalone KV selection algorithm to select the KV entries used for drafting and overlook the fact that the criticality of each KV entry is inherently computed during verification. In this paper, we propose Vegas, a self-speculative decoding method with verification-guided sparse attention. Vegas identifies critical KV cache entries as a byproduct of verification and computes attention only over these entries when drafting subsequent tokens. This not only improves the draft token acceptance rate but also incurs low KV selection overhead, thereby improving decoding throughput. Vegas achieves a 1.25×-2.81× speedup in decoding throughput over default vLLM and a 1.15×-1.29× speedup over state-ofthe-art sparse attention-based self-speculative decoding methods. Our code is available at https: //github.com/platformxlab/vegas.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2602.07223")
get_code_for_paper("2602.07223")
have("2602.07223")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2602.07223.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2602.07223.svg)](https://syntology.ai/paper/2602.07223)

Connect an agent — have() is free.