@wbavon

vllm-plugin-fl-setup-flagos installs and configures vllm-plugin-fl across multiple hardware backends (nvidia, ascend, moore threads, iluvatar) with backend-specific environment setup, dependency verification, and inference validation.

install-stack-flagos deploys a 5-package gpu software stack (vllm, flagtree, flaggems, flagcx, vllm-plugin-fl) into a running container. covers network detection, dependency ordering, per-package validation, and structured error reporting with explicit gate logic.

skill-creator-flagos scaffolds, improves, and validates skills in the flagos repository. covers the full lifecycle from intent capture through structure validation, with modes for creation, iteration, and testing.

model-verify-flagos runs inference twice on a target model - once without multi-chip libraries and once with full stack enabled - to isolate whether failures stem from the model itself or from optimization layers like flaggems/flagcx.

perf-test-flagos runs accuracy (when available) and performance benchmarks against a served model across 5 workload profiles, collecting throughput, latency, ttft, and tpot metrics. currently accuracy testing is a placeholder.

model-migrate-flagos guides backporting vllm models to the flagos plugin pinned at v0.13.0, with argument parsing, upstream reference loading, step-by-step execution, and e2e verification against ground truth.

gpu-container-setup-flagos detects gpu vendor, retrieves pytorch container images from vendor hubs with fallback chains, mounts devices and data volumes, and validates pytorch functionality across nvidia, amd, ascend, metax, and iluvatar gpus.

tle-developer-flagos orchestrates end-to-end TLE kernel development, optimization, and debugging with mandatory input structure and reproducible validation workflows. relies on external reference files for procedural detail.

Full FlagRelease pipeline orchestrator. Runs the complete LLM deployment, verification, and benchmarking pipeline for multi-chip GPU backends. Executes: inst...

Unified GPU kernel operator generation and optimization skill. Automatically detects the target repository type (FlagGems, vLLM, or general Python/Triton) an...