Pytorch (Cuda + DinD)

nvidia/cuda devel image: nvcc and the full CUDA 13.0 toolkit, no PyTorch. Pick it when you compile CUDA code or need ncu or nvcc without an apt install.

CUDA devel (nvcc, full toolkit)

PyTorch with the cluster overlay, for renting several fabric-linked nodes as one job.

Multi-node cluster

This is my second image

Stable PyTorch 2.12 CUDA 13.0 DIND development template.

Use this as the default Lium dev box for H100/Hopper and RTX PRO Blackwell/SM120 work.

Includes:
- Ubuntu 24.04 DIND base using cruizba/ubuntu-dind
- CUDA toolkit 13.0
- PyTorch 2.12.0 CUDA 13.0 wheel
- Docker Engine 27.5.1 / containerd 1.7.24 pinned for Lium nested Docker
- NVIDIA Container Toolkit configured for nested Docker
- OpenSSH server and long-lived startup command for Lium verification
- Bundled hello-world image for Lium DIND probe reliability

Why this is the stable default:
- CUDA 13.0 is PyTorch 2.12's stable CUDA line.
- CUDA 13.0 supports Hopper and Blackwell SM120 without relying on the deprecated CUDA 12.8 wheel line.
- CUDA 13.2 should remain a separate experimental template until Blackwell ecosystem packages settle.

Template fields:
- Entrypoint: blank, use image ENTRYPOINT
- Start command: blank, use image CMD
- Volume: /workspace
- Internal ports: 22, 8000, 8888


Stable PyTorch 2.12 CUDA 13.0 DIND

Systemd DIND

CUDA 13.0.2

Jammy DIND

axolotl-main-latest

SN19-FALLBACK

techx/pytorch-2.7.1-cuda12.8-dev

llama-cpp

PyTorch Latest

ms-swift

Docker-in-Docker Ubuntu Jammy image with dynamic port allocation

Jammy DIND n

lium-cuda-dind

PyTorch 2.12 CUDA 13.0 Docker-in-Docker template for Lium Sysbox pods.

Stable PyTorch 2.12 CUDA 13.0 DIND verified 20260611-1418

pytorch/pytorch:2.11.0-cuda12.8-cudnn9-devel

WAN Animate no models

lmsys dev copy(latest)

test-template-1754927339

One Click - ComfyUI Minimax H3

prism-recipe-v9-public

Docker

sn51-std-container

lium-web-test

vLLM 0.15.1 pre-installed on Datura PyTorch (CUDA 12.8, Python 3.11). No pip install needed.

Affine vLLM

Test final synchronisation VoltageGPU ↔ Lium

FINAL TEST 2025-08-29 12:39:04

Alpha-Miner

mission-miner-full-e2e

Vidaio(Cuda)

NVCR_Torch

score-vlm-systemd-qualification

CMP 170HX 64GB

comfyui

sgl fix

Mine Pearl (PRL) on AlphaPool — lowest-fee, highest revenue-per-hash pool on the Pearl network. Drop in your wallet address and start earning on any NVIDIA tensor-core GPU.

AlphaPool Pearl Miner

lium4

my_image

glm-5.3-cyber

MIUMIUM

test-refactor

SN19-QWQ-085

DeepSeek-V3-0324 | sglang

Lium image thing.

test-mik

Aphrodite Engine

lium6

Template créé via interface VoltageGPU simulée

INTERFACE TEST 2025-08-29 13:03:36

Lium 2

XXXXXXXXXX

simple-web

constation-dind-host-1785136195

This is my image

Pytorch NGC

REAL API TEST 2025-08-29 05:45:22

probe-benign

KoboldCpp

TEMPLATES