The Agent Skills Directory

[SAFE]: The skill provides legitimate technical instructions for using vLLM, a well-known open-source library for high-performance LLM serving.\n- [EXTERNAL_DOWNLOADS]: Recommends installing standard, official packages from PyPI including vllm, torch, transformers, locust, autoawq, auto-gptq, and flash-attn. These are standard tools for the AI/ML ecosystem.\n- [COMMAND_EXECUTION]: Documents the use of CLI tools for serving models, benchmarking performance, and managing server infrastructure (e.g., vllm serve, docker run, locust). All commands are relevant to the skill's purpose and are documented with clear usage context.\n- [REMOTE_CODE_EXECUTION]: Mentions the --trust-remote-code flag, which is a standard configuration option in the transformers and vllm ecosystems required to run models that include custom code in their HuggingFace repositories.\n- [PRIVILEGE_ESCALATION]: Includes a troubleshooting step that suggests using sudo to configure firewall settings (sudo ufw allow 8000). This is a common and appropriate administrative task for opening a network port on Linux systems.

serving-llms-vllm