ollama-model-generator

Ollama Model Generator

Introduction

This NodeJS CLI script simplifies adding GGUF models to Ollama by creating symlinks and downloading necessary metadata from the Ollama Registry.

Benefits:

• Avoids model duplication within Ollama.
• Easy integration of GGUF models.
• No dependencies besides NodeJS.

Installation

Requires NodeJS version 18.11.0 or higher. Install it globally using npm:

npm install -g ollama-model-generator

Usage

ollama-model-generator [options]

  Options:
    --model <path>            Path to the GGUF model file. This will be symlinked to Ollama blob storage.
                              If the file doesn't exist, it will be downloaded from the Ollama Registry based on --from.
                              Optional. If not provided, the model will be downloaded to the Ollama blob storage.
  
    --from, -f <name>         Model name in the Ollama Registry to download as a base.
                              Default: architecture of the GGUF model.
                              
    --name, -n <name>         Name of the new model in Ollama.
                              Default: basename-size-finetune-version of the GGUF model.
                              If --model is not provided, it defaults to the name from --from.
                              
    --show, -s                Prints model metadata from the GGUF file header as JSON (requires --model).
    
    --registry, -r <registry> The Ollama Registry URL.
                              Default: registry.ollama.ai
    
    --dir, -d <path>          Directory for storing Ollama model data.
                              Default: $OLLAMA_MODELS or ~/.ollama/models

Additional files can be symlinked in the same way as --model ( see Ollama Model File):

--adapter, --embed, --license, --messages, --params, --projector, --prompt, --system, --template

Example

Download a model from the Ollama Registry

ollama-model-generator --from gemma2

This will download the Gemma 2 model from the Ollama Registry and configure it in Ollama (same as ollama pull gemma2).

Use a local GGUF model

ollama-model-generator --from llama3.1 --model my-model.gguf --name LLama3.1-MyModel

This will use the local my-model.gguf file and configure it in Ollama with the name LLama3.1-MyModel. The Ollama metadata (template, params etc.) is taken from the Llama 3.1 model.

Use custom template

ollama-model-generator --from gemma2 --template my-template.txt

This will download the Gemma 2 model but use the local my-template.txt file as prompt template.

Print GGUF metadata

ollama-model-generator --show --model my-model.gguf

Prints the GGUF metadata of the model file as JSON.