# aliyun/simai

> SimAI is a simulator for large-scale AI inference and training, used to estimate how long AI tasks take and to analyze process details. It models the framework, collective communication and network layers, so users can measure end-to-end performance under different settings such as collective algorithms, NCCL environment variables or congestion control. The core is written in Python and C++.

Source: https://github.com/aliyun/simai
Indexed: commit cef83f9 on 2026-09-15 by Repowise. Facts come from the source and git history, not from the maintainer.

## Facts

- Type: App
- Language: Python
- Stars: 1,301 at index time
- License: Apache-2.0
- Code health: 6.8/10, Fair
- Last commit: 2026-07-10
- Contributors: 20

## Install

From the README: https://github.com/aliyun/simai#quick-start

Clone the simai repository.
    git clone https://github.com/aliyun/simai.git

Move into the simai folder.
    cd simai

Initialize the submodules in the repository.
    git submodule update --init --recursive

Compile SimAI-Analytical.
    ./scripts/build.sh -c analytical

Compile SimAI-Simulation with ns3 support.
    ./scripts/build.sh -c ns3

Run SimAI-Analytical on the example workload, which calculates bus bandwidth automatically.
    ./bin/SimAI_analytical -w ./example/workload_analytical.txt -g 9216 -nv 360 -nic 48.5 -n_p_s 8 -g_p_s 8 -r example-

## How it works

A run starts from a chosen mode (analytical, simulation or flow-only), which selects a scenario and feeds the simulation front end. The entry point astra-sim-alibabacloud/astra-sim/network_frontend/ns3/entry.h connects the network front end to the NS3 simulator, and astra-sim-alibabacloud/astra-sim/network_frontend/analytical pairs an analytical network API with a simulation driver called AnaSim. The astra-sim-alibabacloud/astra-sim/system directory executes the collective and network models, and astra-sim-alibabacloud/astra-sim/workload defines runnable simulation layers through the Layer class. On the Python side, vidur-alibabacloud/vidur/main.py is an entry point that runs scheduling scenarios, and vidur-alibabacloud/vidur/config holds the configuration dataclasses for the models it simulates. Profiling and analysis components, including vidur-alibabacloud/vidur/config_optimizer/analyzer/dashboard/main.py, turn run outputs into performance measurements and dashboards.

- astra-sim-alibabacloud/astra-sim/network_frontend/ns3/entry.h: entry point
- vidur-alibabacloud/vidur/config_optimizer/analyzer/dashboard/main.py: entry point
- vidur-alibabacloud/vidur/config_optimizer/config_explorer/main.py: entry point
- vidur-alibabacloud/vidur/main.py: entry point
- vidur-alibabacloud/vidur/profiling/attention/main.py: entry point
Modules: vidur-alibabacloud, astra-sim-alibabacloud, root, scripts. 340 files, 32,175 lines, 2,486 symbols.

## Code health

- Score: 6.8/10 (Fair)
- Rank: #157 of 4,000+ indexed repos
- Hotspots: none
- Lowest file: astra-sim-alibabacloud/astra-sim/workload/Layer.cc
- Dead code: 37 findings

## Maintenance

- Last commit 2026-07-10, first commit 2024-10-11, 89 commits
- 20 contributors, bus factor 1

## Security

- Security findings are shown to signed-in users at https://repowise.dev/repo/aliyun/simai/security

## More

- Page: https://repowise.dev/repo/aliyun/simai
- Ask this codebase: https://repowise.dev/repo/aliyun/simai/chat
- Docs: https://repowise.dev/repo/aliyun/simai/docs
- MCP (signs in with Repowise on first use):
    claude mcp add --transport http repowise https://api.repowise.dev/mcp/aliyun/simai
