← Back to match
simllm
Packet-level network simulator predicting LLM serving performance before buying hardware
PlatformWebfree
SimLLM is an open-source simulator that predicts the serving performance of large LLM deployments, such as time-to-first-token and throughput, before hardware is purchased or reserved. It combines real serving-framework schedulers like vLLM and SGLang with a simulated GPU executor and a packet-level network model to capture effects such as network congestion at scale. It is aimed at ML infrastructure teams planning multi-node LLM serving or training clusters.
Categories
AI InfrastructureNetworking Simulation

