Docker Model Downloader

Interactive downloader for Docker ai/* GGUF models, designed for importing the downloaded blobs into other local runtimes such as Ollama and llama.cpp.

Why this exists:

• Docker Desktop's model GUI can fail to fetch model metadata even when the docker model CLI still works.
• docker model pull can fail on unreliable or corporate networks; this script automatically retries downloads.
• Docker can be an allowed route for GGUF downloads where Ollama, Hugging Face, or ModelScope are blocked.
• Downloaded Docker GGUF blobs can then be reused in Ollama, llama.cpp, or other GGUF-compatible apps.
• The script identifies GGUF blobs to simplify importing into Ollama or llama.cpp, with richer metadata when llama-gguf or gguf_dump is available.

Interactive script to download GGUF AI models via Docker, locate the downloaded GGUF files, and import them into other local runtimes.

Support

Donate to support this work: Ko-fi Enelass

Capabilities

• Download Docker AI models: browse the live Docker Hub ai/* catalog, select a model, inspect all available tags/variants, and pull the exact variant you want.
• Recover from flaky pulls: retries docker model pull failures automatically, which helps on unreliable or corporate networks.
• Inspect local downloads: scan ~/.docker/models/blobs/sha256/ for completed GGUF blobs, list incomplete downloads separately, and show useful metadata such as role, architecture, size, context length, quantization, tensor count, and cropped path.
• Use optional llama.cpp metadata tooling: detects llama-gguf from brew install llama.cpp, while still using gguf_dump when available. Downloads do not require either tool.
• Reuse Docker GGUF blobs elsewhere: prints Ollama import commands and file locations so downloaded Docker models can be used with Ollama, llama.cpp, or other GGUF-compatible runtimes.
• macOS compatibility filtering: hides vLLM-only Docker model entries on macOS because those variants are not compatible there.

Prerequisites

• Bash (macOS ships Bash 3.2)
• Docker Desktop (required) must be installed and running
• jq (used to parse the Docker Hub API)
• Ollama (optional) to run the downloaded models
• llama-gguf (from brew install llama.cpp) or gguf_dump (optional) for richer downloaded-model metadata and more precise GGUF matching

Installation & Usage

Run with a single command:

bash <(curl -s https://raw.githubusercontent.com/Enelass/Docker_Model_Downloader/refs/heads/main/download_docker_model.sh)

Changelog / Releases

• Changelog: CHANGELOG.md
• Release process: RELEASING.md

Features

• Fetches an up-to-date list of Docker Hub ai/* models every run
• Starts with a keyboard-selectable action menu for downloading models or checking local downloads
• Shows 20 models per page and crops long model names so the table stays readable
• Lists Docker model variants, not only the default tag, with parameters, quantization, context, VRAM, tool-calling, and size when Docker metadata provides it
• Filters vLLM-only entries on macOS because they are not compatible there
• Checks locally downloaded Docker GGUF blobs without starting a download, including grouped metadata, cropped paths, incomplete downloads, and an optional purge action
• Retries failed model downloads automatically
• Allows cancelling an active download with Ctrl+C and returning to variant selection
• Shows spinner feedback while retrieving models, variants, and local metadata
• Automatic GGUF file detection with llama-gguf or gguf_dump
• Ready-to-use Ollama import commands

Navigation

• Startup menu: Up/down arrows choose the action, Enter selects, and the downloader starts automatically after 5 seconds
• Model list: Left/right arrows navigate pages
• Number + Enter: Select model
• q: Quit

That's it. Run the script, pick a model, and follow the on-screen instructions.