FlagOS
Official@flagos-ai · China
A Unified, Open-Source AI System Software Stack
Agent Skills by FlagOS
Showing 20 vetted skills indexed across 2 GitHub repositories.
flagos-aitest-dependencies
Prepare and validate isolated Python environments for torch_fl tests without replacing the PyTorch build.
flagos-aitorch-version-port
Ports torch_fl to a new PyTorch minor version by regenerating ATen codegen and reconciling dispatcher signatures.
flagos-aipre-pr-checks
Runs CI-enforced rebase, ruff lint, and test checks before creating or updating pull requests.
flagos-aicuda-op-integration
Generates CUDA boxing operators from PyTorch schemas for the torch_fl flagos backend.
flagos-airuntime-bringup
Implement the torch_fl device runtime contract for a new accelerator backend.
flagos-aicuda-compat-vendor
Enable PyTorch operators on CUDA-compatible accelerators by bundling the vendor's libtorch_cuda.so.
flagos-aitransformers-test
Runs HuggingFace transformers tests on custom chips with triage, verification, and issue preview.
flagos-aiflaggems-integration
Enable and validate FlagGems Python or C++ operator routing on torch-fl accelerators.
flagos-ainative-op-backend
Generates native operator kernels for non-CUDA-compatible accelerators via category-based codegen.
flagos-aitemplate-skill
Create a reusable skill scaffold with YAML frontmatter for Claude agents.
flagos-aiskill-creator-flagos
Scaffold, validate, and improve FlagOS skills with structured frontmatter.
flagos-aiperf-test-flagos
Benchmark vLLM-served AI models across five workload profiles for throughput, latency, TTFT, and TPOT.
flagos-aimodel-migrate-flagos
Migrates a model from latest vLLM upstream into vllm-plugin-FL via 13-step copy-and-patch workflow with validation scripts.
flagos-aiinstall-stack-flagos
Install and validate vLLM, FlagTree, FlagGems, FlagCX, and vllm-plugin-FL in a container.
flagos-aitle-developer-flagos
Coordinate end-to-end TLE kernel development from intake to merge decision.
flagos-aiflagrelease-entrance-flagos
Orchestrate LLM deployment and benchmarking across multi-chip GPU backends.
flagos-aikernelgen-flagos
Automate GPU kernel generation and optimization across FlagGems, vLLM, and generic repositories.
flagos-aivllm-plugin-fl-setup-flagos
Automate vLLM-Plugin-FL installation and configuration across multiple hardware backends.
flagos-aigpu-container-setup-flagos
Detect multi-vendor GPU environments and launch compatible PyTorch containers.
flagos-aimodel-verify-flagos
Diff base and multi-chip serving runs to locate failing layers.
Frequently Asked Questions About FlagOS
FAQPage SchemaWhat specific tasks can engineers perform using FlagOS?▼
Engineers can benchmark vLLM-served models, migrate upstream models into specialized plugins, generate and optimize GPU kernels, and verify model layer integrity across multi-chip backends. It provides a structured environment for managing the lifecycle of high-performance computing deployments from initial intake to final merge decisions.
Which technical personas benefit most from this stack?▼
This stack is designed for infrastructure engineers, GPU kernel developers, and performance researchers. It targets professionals responsible for maintaining high-throughput serving environments, optimizing hardware-specific compute kernels, and ensuring model consistency across heterogeneous multi-vendor GPU clusters.
What are the prerequisites for deploying the FlagOS stack?▼
Deployment requires a multi-vendor GPU environment capable of supporting PyTorch containers. Users must have access to the FlagOS repository suite to initialize the installation of vLLM, FlagTree, FlagGems, FlagCX, and the vllm-plugin-FL components within their containerized infrastructure.