aliyun/SimAI

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1133 stars · 178 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

SimAI is a large-scale network simulation toolkit from Alibaba Cloud for modeling AI training and inference workloads on GPU clusters, published at NSDI'25. It combines workload generation (AICB), collective communication modeling (SimCCL), and request scheduling (adapted from Vidur) to simulate end-to-end distributed LLM training and prefill/decode inference.

Use cases

• simulate network traffic of large-scale LLM training on GPU clusters
• evaluate network designs for distributed AI training before deployment
• model prefill/decode disaggregated inference performance
• estimate GPU memory and decode latency for LLM inference
• generate training workloads for models like DeepSeek and Qwen
• analyze collective communication flows offline with SimCCL

When to choose

• you are designing or evaluating network topologies and congestion control for AI clusters
• you need to predict training or inference performance without running real GPU jobs
• you want to study collective communication behavior at scale

When to avoid

• you need a production training or serving framework rather than a simulator
• you want quick single-GPU model benchmarking
• you need cycle-accurate hardware simulation

Facets

library · maturity active

simulation machine-learning networking benchmarking llm-training llm-inference simulation machine-learning networking microservices large-language-models performance python network-simulation gpu-clusters collective-communication nsdi-paper training-simulation inference-simulation alibaba-cloud linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="aliyun/SimAI")