home/categories/backend/orchestra-research-ai-research-skills-12-inference-serving-vllm-skill-md
backenddevelopment

serving-llms-vllm

Serves LLMs with high throughput using vLLM's PagedAttention and continuous batching. Use when deploying production LLM APIs, optimizing inference latency/throughput, or serving models with limited GPU memory. Supports OpenAI-compatible endpoints, quantization (GPTQ/AWQ/FP8), and tensor parallelism.

Orchestra-Research
maintainer
Orchestra-Research
اپ ڈیٹ ہوا 11/20/2025
اسٹارز
6563
فورکس
515
quick start

Installation and usage

Serves LLMs with high throughput using vLLM's PagedAttention and continuous batching. Use when deploying production LLM APIs, optimizing inference latency/throughput, or serving models with limited GPU memory. Supports OpenAI-compatible endpoints, quantization (GPTQ/AWQ/FP8), and tensor parallelism.

انسٹالیشن
$ install --globalskills.sh
استعمال

انسٹال کرنے کے بعد، آپ یہ اسکل ٹرمینل میں درج ذیل کمانڈ چلا کر استعمال کر سکتے ہیں:

skills use serving-llms-vllm