How to Deploy DeepSeek R1 Reasoning Model on AMD GPUs Using SGLang

DeepSeek R1 is a first-generation reasoning-focused large language model optimized for math, coding, and logical tasks using reinforcement learning. A technical guide outlines how to deploy it using the SGLang inference framework inside a ROCm-compatible Docker container on an AMD Instinct MI300X GPU server. The setup involves downloading the model via Hugging Face CLI, building the SGLang ROCm container, and launching an inference server with tensor parallelism across eight GPUs on port 30000. Once running, the server exposes an OpenAI-compatible HTTP API that can be tested with standard curl requests. The guide also recommends securing the deployment with a reverse proxy and TLS before exposing it beyond a local network.