inference/research
verifiedhigh · 0.90updated 2026-10-03

SGLang: Efficient Execution of Structured Language Model Programs

paper

Fields

id
sources/flop-sglang
Publisher
arXiv (SGLang team)
Kind
paper
Published
2023-12-12
Accessed
2026-10-03
notes
RadixAttention: a radix tree over the KV cache enabling prefix-aware reuse across requests, plus compressed finite state machines for constrained decoding. Up to 6.4x throughput vs prior systems on agent, RAG and JSON-decoding workloads.

Notes

RadixAttention: a radix tree over the KV cache enabling prefix-aware reuse across requests, plus compressed finite state machines for constrained decoding. Up to 6.4x throughput vs prior systems on agent, RAG and JSON-decoding workloads.

Cross-references

This record references no other records.