PNArchives/Memoh
1
0
Fork 0
mirror of https://github.com/memohai/Memoh.git synced 2026-04-27 07:16:19 +09:00
Code Issues Packages Projects Releases Wiki Activity

feat: Add GPU CDI support for workspace containers (#332)

Browse Source 
* feat: add CDI GPU support for workspace containers

* feat: expose GPU CDI settings in bot container UI

* feat: move GPU settings into advanced container options

* docs: document advanced CDI device configuration
This commit is contained in:
Ming Lin
2026-04-10 14:52:17 +08:00
committed by GitHub
parent 19619d73a9
commit 4d3f2de7e2
22 changed files with 752 additions and 84 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/docs/getting-started/container.md
+46
View File
@@ -33,6 +33,52 @@ The **Container** tab displays real-time data about the bot's runtime:
- **Image**: The Docker/Containerd image used as the base.
- **Paths**: Host and container paths for data persistence.
- **Tasks**: Number of active background tasks running in the container.
- **CDI Devices**: The effective GPU CDI devices currently attached to the container, if any.
---
## Advanced: Provide CDI Devices
Memoh can provide host devices to a bot container through CDI (Container Device Interface). This is an advanced capability for users who want to expose host-managed devices, most commonly GPUs, to the container runtime.
In the Web UI, this capability is placed under **Advanced options** in the **Container** tab. It is optional and only needs to be configured when the bot must access CDI-backed devices from the host.
### Configure CDI Devices
1. Open the Bot's **Container** tab.
2. Click **Create** if the container does not exist, or recreate the container if you need to change GPU settings.
3. Expand **Advanced options**.
4. Enable **GPU**.
5. Enter one or more CDI device names in **CDI devices**.
You can enter CDI device names one per line or separated with commas. Common GPU-related examples:
- `nvidia.com/gpu=0`
- `nvidia.com/gpu=all`
- `amd.com/gpu=0`
- `amd.com/gpu=all`
### Host Requirements
Before configuring CDI devices in Memoh, the host machine must already provide working device drivers, vendor toolkit support where required, and valid CDI specs. In practice, this usually means:
- the host GPU works normally outside the container
- CDI spec files exist under `/etc/cdi` or `/var/run/cdi`
- the device name you enter in Memoh matches a real CDI device on the host
To discover the exact CDI device names exposed by the host, use the vendor tool on the host machine:
- NVIDIA: `nvidia-ctk cdi list`
- AMD: `amd-ctk cdi list`
If Memoh reports an error such as `unresolvable CDI devices`, the configured device name does not match any CDI device visible to the container runtime.
### Important Behavior
- CDI device settings are applied when the container is created. Updating the setting later requires recreating the container.
- Stopping and starting an existing container does not change its attached CDI devices.
- The container image still needs the appropriate user-space libraries and tools if you want to run CUDA or ROCm software inside the container.
- After creation, the **Container** tab shows the effective attached CDI devices for verification.
---