qwen3-vl-8b

Qwen3-VL-8B-Instruct, self-hosted: questions about images and pointing for development runs without API spend

• Kind: vlm
• Protocol: openai_chat
• Default port: 8801
• VRAM: 18 GB
• Load time: about 120 s
• Skills that use a vlm: pointing, vlm, manipulation, scene
• Configurations that name it: –

Weights

harness fetch qwen3-vl-8b --deploy <machine>

Directory	Source	From
<weights>/qwen3-vl-8b	modelscope_model	Qwen/Qwen3-VL-8B-Instruct

Manifest

kind = "vlm"
summary = "Qwen3-VL-8B-Instruct, self-hosted: questions about images and pointing for development runs without API spend"
protocol = "openai_chat"
command = ["{python}", "-u", "services/servers/serve_vlm.py", "--model", "{weights}/qwen3-vl-8b",
           "--name", "Qwen3-VL-8B-Instruct", "--port", "{port}"]
port = 8801
vram_gb = 18.0
load_seconds = 120
client = { backend = "openai_chat", base_url = "{url}/v1", model = "Qwen3-VL-8B-Instruct" }
needs_files = ["{weights}/qwen3-vl-8b/config.json"]

[[weights]]
dir = "qwen3-vl-8b"
modelscope_model = "Qwen/Qwen3-VL-8B-Instruct"