Templates
Ready-to-run images for common AI workloads. Every template includes root SSH, JupyterLab and the same isolation — pick one, choose a GPU, and you're running.
PyTorch
PyTorch 2.7 with CUDA 12.8, cuDNN 9, JupyterLab, Transformers and Accelerate. The default for training and research.
edge-pytorch:2026.10
NVIDIA CUDA Dev
Full CUDA 12.8 toolkit with nvcc, cuDNN headers, CMake and compilers — for building kernels and native GPU code.
edge-cuda:2026.10
vLLM (OpenAI-compatible API)
High-throughput LLM serving. Loads any Hugging Face model and exposes an OpenAI-compatible API protected by your rental's API key.
edge-vllm:2026.10
· exposes OpenAI-compatible APIOllama
Run quantised open models with one command. Pulls the models you choose at start-up and serves the Ollama API.
edge-ollama:2026.10
· exposes Ollama APIOpen WebUI (Ollama)
A ChatGPT-style web interface on top of Ollama — chat, documents/RAG, and model switching in the browser.
edge-openwebui:2026.10
· exposes Open WebUIComfyUI
Node-based image and video generation. Comes with ComfyUI-Manager; model files for the chosen model are downloaded at start-up.
edge-comfyui:2026.10
· exposes ComfyUIWhisper API
Speech-to-text with Whisper (PyTorch) behind an OpenAI-compatible /v1/audio/transcriptions endpoint.
edge-whisper:2026.10
· exposes Transcription APIUnsloth Fine-tuning
Fast, memory-efficient LoRA / QLoRA fine-tuning for Llama, Qwen, Gemma and more, driven from JupyterLab.
edge-unsloth:2026.10