OpenClaw + SGLang on AMD Instinct MI300X

One command to launch a SGLang inference server with Qwen3.5-122B and install OpenClaw fully configured against it — no manual steps.

Quick Start

curl -fsSL https://raw.githubusercontent.com/Mahdi-CV/openclaw-amd-sglang/main/setup_openclaw_sglang.sh | bash

That's it. The script will:

1. Start the SGLang Docker container (Qwen/Qwen3.5-122B-A10B-FP8, port 8090)
2. Wait for the server to be ready — including Docker image pull and model load (up to 1 hour)
3. Install OpenClaw via npm (installs Node 22 if needed)
4. Configure OpenClaw to point at the SGLang endpoint automatically
5. Start the OpenClaw gateway
6. Print the SSH tunnel command to access the browser UI

Requirements

• AMD Instinct MI300X GPU
• ROCm drivers installed (/dev/kfd, /dev/dri available)
• Docker
• Internet access (for Docker image + model weights on first run)

Options

Pass flags after the script:

curl -fsSL https://raw.githubusercontent.com/Mahdi-CV/openclaw-amd-sglang/main/setup_openclaw_sglang.sh | bash -s -- [options]

Option	Default	Description
--port PORT	8090	Port to expose the SGLang server on
--api-key KEY	abc-123	API key for the SGLang endpoint
--model MODEL_PATH	Qwen/Qwen3.5-122B-A10B-FP8	HuggingFace model path
--served-name NAME	qwen3-5-122b	Model name used in API calls
--hf-cache PATH	/data/hf_cache	Local path to cache model weights
--tp-size N	1	Tensor parallel size
--no-wait	—	Skip waiting for SGLang to be ready before installing OpenClaw
--sglang-only	—	Only start SGLang, skip OpenClaw install
--openclaw-only	—	Only install OpenClaw (assumes SGLang is already running)

Examples

Custom API key and port:

curl -fsSL https://raw.githubusercontent.com/Mahdi-CV/openclaw-amd-sglang/main/setup_openclaw_sglang.sh | bash -s -- --api-key my-secret-key --port 8080

SGLang only (skip OpenClaw):

curl -fsSL https://raw.githubusercontent.com/Mahdi-CV/openclaw-amd-sglang/main/setup_openclaw_sglang.sh | bash -s -- --sglang-only

OpenClaw only (SGLang already running):

curl -fsSL https://raw.githubusercontent.com/Mahdi-CV/openclaw-amd-sglang/main/setup_openclaw_sglang.sh | bash -s -- --openclaw-only

After Setup

The script prints a ready-to-use dashboard URL with the auth token included. To open the browser UI:

1. On your local machine, open the SSH tunnel:

   ssh -N -L 18789:127.0.0.1:18789 root@<server-ip>

2. Open the URL printed by the script (it includes the token), or run on the server:

   openclaw dashboard

Other useful commands:

# Check OpenClaw status
openclaw status

# Tail SGLang logs
docker logs -f sglang_server

# Tail gateway logs
tail -f /tmp/openclaw-gateway.log

# Stop everything
docker rm -f sglang_server
openclaw gateway stop

Related

• OpenClaw on AMD Developer Cloud: Qwen 3.5 and SGLang
• OpenClaw Docs
• SGLang