Choose an image
The image decides what software is in your instance: ROCm, Python, PyTorch and the tools around them. Pick it under Image in the launch form.
Start with AMD OneClick Base
Section titled “Start with AMD OneClick Base”AMD OneClick Base is the default, described in the form as Default ROCm Jupyter/OpenCode image. It is the image the platform maintains for general work. Use it unless you have a reason to pick something else.
In October 2026 it contained:
| Component | Version |
|---|---|
| OS | Ubuntu 24.04 |
| Python | 3.12, in a virtual environment at /opt/venv |
| PyTorch | 2.9.1, built for ROCm 7.2 |
| JupyterLab | 4.5 |
| Other tools | OpenCode, rocm-smi |
The image is updated from time to time, so check the versions in your own instance (see below).
Read the image list
Section titled “Read the image list”The Image field is a searchable list. Type part of a name to filter it. When an image has a description, it appears under the field after you select it.
The list depends on your account, and it also contains images made for specific courses, workshops and demos. As a rule:
| You want to | Pick |
|---|---|
| Develop, explore, or run most PyTorch code | AMD OneClick Base |
| Use the stock ROCm PyTorch environment | rocm-pytorch |
| Follow a course, workshop or tutorial | The image its instructions name |
| Serve a model with Deploy type: Model API | Nothing here; that form picks its own runtime. See Serve a model. |
Avoid images with no description unless an instruction tells you to use them.
Images in the list
Section titled “Images in the list”The tables below list the images on the China site in October 2026, by the name shown in the Image field. Your list may differ, and images are added and retired over time.
To follow a course or workshop, launch its notebook from the Gallery instead of picking an image yourself. The notebook launches with the right image.
General purpose
Section titled “General purpose”| Image | What’s in it | Use it for |
|---|---|---|
| AMD OneClick Base | ROCm 7.2, Python 3.12, PyTorch, JupyterLab and OpenCode. Maintained by the platform. See above. | Most work. The default. |
| rocm-pytorch | AMD’s official rocm/pytorch image: ROCm 7.2.4, Ubuntu 24.04, Python 3.12, PyTorch 2.10.0. | Code that expects the stock ROCm PyTorch environment, or a newer PyTorch than AMD OneClick Base. |
| rocm10-pytorch | An official rocm/pytorch image built on ROCm 10. | Trying ROCm 10. The Hello GPU notebook uses it. |
Model serving
Section titled “Model serving”| Image | What’s in it | Use it for |
|---|---|---|
| Qwen3.8-27B | vLLM set up to serve Qwen3.8-27B. Needs a 2-GPU size. | Running Qwen3.8-27B yourself in a notebook. To get an endpoint without setting up vLLM yourself, use Serve a model instead. |
Courses, workshops and events
Section titled “Courses, workshops and events”These images exist for one course or event. Pick one only when its instructions name it.
| Image | Made for |
|---|---|
| AMD-Stable-GH-Proxy | A variant of AMD OneClick Base with a proxy for reaching GitHub. Used by the Hello ROCm and AgenticAI推理自优化 notebooks. |
| Unsloth Studio | Unsloth fine-tuning. Used by the Modern LLM Notebook. |
| Muse Glimmer Llamacpp | A llama.cpp environment prepared for an event. |
| Oceanbase | SeekDB and hipVS demo, version 1.0.0, for the OceanBase database contest. Superseded by seekdb-hipvs-nb-demo_v1.4.0. |
| seekdb-hipvs-nb-demo_v1.4.0 | SeekDB and hipVS demo, version 1.4.0, for the OceanBase database contest. Used by the seekdb-hipvs notebook. |
| Ultralytics Workshop | An Ultralytics and OpenCV workshop (August 2026). |
| Ultralytics YOLO26 | Ultralytics YOLO26 notebooks. |
| ultralytics-yolo26-workshop-20260911-unified | Used by the Ultralytics Workshop2 notebook. |
| ultralytics_workshop_1 | Used by the Ultralytics Workshop 1 notebook. |
| robotwin | RoboTwin simulation with LingBot-VLA, for embodied AI. |
| FPGA-CHINA | FPGA RTL and HLS tools on ROCm 10, for the FPGA China event. |
| ai-academy_in_radeon-cloud:vllm-openai-rocm-v0.26.0 | Most AI Academy courses: GPU Profiling with rocprof, GPU Debugging with ROCgdb, HIP Programming Basics, HIP Kernel Programming, CUDA-to-HIP Conversion, Triton Kernel Development, Diffusion Models - Image Generation, Agents 101 and HF on AMD. |
| ai-academy_in_radeon-cloud:vllm-openai-rocm-v0.16.0 | AI Academy: LLM Serving Inference with vLLM, and AI on AMD. |
| ai-academy_in_radeon-cloud:vllm-openai-rocm-v0.16.0-curl | The v0.16.0 image plus curl, for the API exercise in the vLLM serving course. |
| ai-academy_in_radeon-cloud:vllm-dev-open-mi300-08052025 | An alternative environment for the AI Academy vLLM serving course on one 48 GB GPU. |
| ai-academy_in_radeon-cloud:vllm-openai-rocm-v0.26.0-benchmark | AI Academy: GPU Optimization for LLM Inference, Lab 1. |
| ai-academy_in_radeon-cloud:llms-benchmark-v1 | AI Academy: GPU Optimization for LLM Inference, Lab 2. |
| ai-academy_in_radeon-cloud:vllm-openai-rocm-v0.26.0-web | AI Academy: Diffusion Models - Video Generation. |
| ai-academy_in_radeon-cloud:gspo-lab-v1.0.0 | AI Academy: Reinforcement Learning with LLMs. |
| ai-academy_in_radeon-cloud:agentic-ai-adk-tutorial-20260317 | AI Academy: Agents 201. |
| ai-academy_in_radeon-cloud:unsloth-rocm-20261013 | AI Academy: Finetuning Your Own R1 Reasoning Model on Unsloth. |
Check what an image contains
Section titled “Check what an image contains”Versions change as images are updated, so check them inside the instance rather than relying on the name. From a JupyterLab terminal:
rocm-smipython --versionpython -c "import torch; print(torch.__version__, torch.version.hip, torch.cuda.is_available())"pip listtorch.version.hip shows the ROCm version PyTorch was built for, and torch.cuda.is_available() returning True means PyTorch can use the GPU.
Custom images
Section titled “Custom images”Custom images built in the classic console’s Custom Images page appear in the list with (custom) after their name once the build has finished.