Manual Model Installation

Overview

Manual installation allows you to install models by typing their exact names, giving you more control and flexibility. This is useful when you know the specific model version you want or need to install models not listed in the Discover page.

Installing Models Manually

Open Downloads Page

Click Downloads in the sidebar to access the download management area.

Downloads Navigation

Click Pull Model Button

At the top-right corner of the Downloads page, click the "Pull Model" button.

Enter Model Name

A model name input field will appear. Type the complete model name including the version tag.

Format: model-name:version-tag

Examples:

gemma2:2b
llama3.1:8b
mistral:7b-instruct-v0.2
codellama:13b-python

Click Pull Button

After entering the model name, click the Pull button next to the input field.

Monitor Download

The download manager will show the progress:

Download percentage
Current speed (MB/s)
Downloaded size / Total size
Estimated time remaining

Model Naming Format

Understanding Ollama model names helps you install the right version:

Basic Format

model-family:parameter-size

Examples:

llama3:8b - Llama 3 with 8 billion parameters
mistral:7b - Mistral with 7 billion parameters

With Quantization

model-family:parameter-size-quantization

Examples:

llama3:8b-q4_0 - Llama 3 8B with Q4_0 quantization
gemma2:9b-q8_0 - Gemma 2 9B with Q8_0 quantization

With Variants

model-family:parameter-size-variant

Examples:

mistral:7b-instruct-v0.2 - Mistral 7B instruction-tuned version 0.2
codellama:13b-python - CodeLlama 13B Python specialist
llama3.1:70b-instruct - Llama 3.1 70B instruction-tuned

Using "latest" Tag

model-family:latest

Automatically downloads the recommended version:

llama3:latest - Gets the default Llama 3 version
mistral:latest - Gets the default Mistral version

Tip: Finding Model Names

Visit ollama.com/library to browse all available models and their exact names.

Common Model Examples

General Purpose Models

Llama 3.1

llama3.1:8b
llama3.1:70b
llama3.1:405b

Mistral

mistral:7b
mistral:7b-instruct
mistral-nemo:12b

Gemma 2

gemma2:2b
gemma2:9b
gemma2:27b

Specialized Models

Code Generation

codellama:13b
deepseek-coder:6.7b
starcoder2:7b

Multilingual

aya:8b
aya:35b
command-r:35b

Vision Models

llava:13b
bakllava:7b
llava-phi3:3.8b

Advanced Options

Installing Specific Versions

To install an older or specific version:

Visit the model's page on ollama.com/library
Check the Tags section for available versions
Use the exact tag in your manual installation

Example: Installing Llama 2 instead of Llama 3:

llama2:13b

Custom Model Names

If you have custom models or modified versions, you can install them using their full registry path:

registry.example.com/custom-model:version

After Installation

Once the download completes:

Check Local Models: The model appears in Installed (Local Models)
Start Using: Click Chat to begin conversations
Verify Version: Confirm you installed the correct version

Troubleshooting

Manual vs. Discover Installation

When to Use Manual Installation

Manual is better when:

You know the exact model name and version
Installing models not in the Discover library
Need a specific quantization not shown in Discover
Installing custom or experimental models
Prefer command-line style control

When to Use Discover

Discover is better when:

Browsing and exploring new models
Need detailed model descriptions
Want to compare versions side-by-side
Unsure which version to choose
Prefer visual interface

Best Practices

Installation Tips

Verify names carefully: One typo can download the wrong model
Check disk space first: Ensure you have enough room
Start with smaller versions: Test with 7B before downloading 70B
Use specific tags: Avoid :latest in production workflows
Keep a model list: Save names of models you frequently install
Test after install: Always verify the model works as expected

Quick Reference

Most Common Commands

Model	Command	Size
Llama 3.1 (8B)	`llama3.1:8b`	~4.7 GB
Mistral (7B)	`mistral:7b`	~4.1 GB
Gemma 2 (9B)	`gemma2:9b`	~5.4 GB
CodeLlama (13B)	`codellama:13b`	~7.4 GB
Llama 3.1 (70B)	`llama3.1:70b`	~40 GB