Idea: Proxmox support in every layout (testers wanted)
维护者通常 1 天内回复
@sharmis-tp 已经在做这个了。
开始于 2026年10月27日。
评估
- 难度
- 5/5
- 预计耗时
- 一周以上
- 新手友好度
- 25/100
- Issue 类型
- 功能
- 描述清晰度
- 基本清楚
- 活跃度
- 活跃
- 技术栈
- go, linux
调研方向
Start by reading the Berth documentation and the existing berth env, berth run, and berth vfio entry points. The issue proposes detection and behavior across Proxmox hosts, LXC containers, VMs, GPU drivers, and multiple cards, but does not identify source files or tests. Done means the applicable layouts are detected and reported as described, with berth env remaining read only; the maintainer is still seeking testers before work is ready to try.
由索引模型根据 Issue 内容生成。
描述
A lot of you run local models on Proxmox. Berth should work there in each common layout and tell you which one it is in. I have Proxmox hosts but none with an NVIDIA card, so I need your setups to test against.
| Layout | Where Berth runs | Today | Planned for v0.4 |
|---|---|---|---|
GPU on the host, /dev/nvidia* mounted into LXC containers |
the host | Berth sees every container's GPU processes, but as bare PIDs | Processes named by container ("LXC 101 (ollama)"). berth run inside a container leases the host's card. A system service for hosts with no user session |
| GPU passed through to a VM | inside the VM | Works like any Linux machine | A host-side Berth shows the card as passed through and names the VM |
| One card moved between the host and a VM | the host | berth hold around qm start |
berth vfio run drains the card and checks that nothing holds it before your hookscript switches the driver |
| Berth inside an LXC that has the GPU | the container | Works for that container's own engines; it cannot see other containers | berth env tells you to run it on the host instead |
| Several GPUs | one berthd --gpu N per card |
Works, but each engine has to pick its daemon | Real multi-GPU scheduling is v0.5 |
berth env will detect the layout: Proxmox host, LXC or VM, which driver owns each card, which VM holds a passed-through card, and which containers mount the GPU. It then says where Berth belongs. It only reads; it changes nothing.
If you run Proxmox for inference, please comment with:
- your layout from the table, or one it misses
- GPUs and VRAM
- Proxmox VE version, and privileged or unprivileged containers
- the engines you run (Ollama, llama.cpp, vLLM, ComfyUI...)
I'll ping you when there is something to try.
- 主要语言
- Go
- 星标
- 226
- 派生
- 20
- 平均合并
- 43 分钟
- 30 天内合并 PR
- 192
环境准备
这个项目没有提供开发容器、Dockerfile 或贡献指南，环境需要你自己搭建：先看它的 README，通用步骤见我们的新手贡献指南。
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
T-Crypt/berth 的其他 Issue
-
难度 2/5 1-3 小时 新手友好度 76/100
维护者通常 1 天内回复
-
难度 3/5 1-2 天 新手友好度 55/100
维护者通常 1 天内回复
-
难度 5/5 一周以上 新手友好度 42/100
维护者通常 1 天内回复
-
难度 4/5 3-5 天 新手友好度 57/100
维护者通常 1 天内回复
-
berthd --gpu-backend amd exits at startup on a desktop session: "process VRAM exceeds card usage"未关闭
难度 4/5 1-2 天 新手友好度 48/100
维护者通常 1 天内回复
相似的 Issue
-
Object Manager 1.13 resets a vCLU after sending a project可能已有人在做 @tciunelis 今天认领。 未关闭bug ready ticket
难度 2/5 1-3 小时 新手友好度 84/100
Domktorymysli/vclu#152 ·
维护者通常 1 天内回复
-
bug converter
难度 2/5 1-3 小时 新手友好度 62/100
grafana/sigma-rule-deployment#435 ·
维护者通常 1 天内回复
-
bug
难度 2/5 1-3 小时 新手友好度 82/100
维护者通常 1 天内回复
-
🤖 verification skills: harden check.sh and verify.sh argument and file handling (#391 follow-ups)未关闭
难度 2/5 1-3 小时 新手友好度 76/100
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 78/100
google/go-containerregistry#2491 ·
维护者通常 1 天内回复