Ollama is a tool that allows you to get up and running with large language
models locally. It provides a simple command-line interface to run and
manage models, as well as a REST API for programmatic access.
Ollama supports a wide range of models available on ollama.com/library,
including popular models like Llama 3, Gemma, and Mistral. It also
allows you to customize models and create your own.
With Ollama, you can:
- Run large language models on your own machine
- Chat with models in the terminal
- Generate text and embeddings
- Customize models with your own prompts and data
- Expose models through a REST API for use in your applications
- Run generative models that generate images
===> The following configuration options are available for ollama-0.40.13:
====> Options available for the group BACKENDS
CPU=on: Build CPU backend shared libraries for various SIMD instruction sets
VULKAN=on: Build Vulkan GPU backend shared library
MLX=on: Build MLX backend for image generation (CPU)
===> Use 'make config' to modify these settings
You installed ollama: the AI model runner.
To run ollama, plese open 2 terminals.
1. In the first terminal, please run:
$ ollama serve
2. In the second terminal, please run:
$ ollama run gemma3
or
$ ollama run mistral
This will download and run the specified AI model.
You will be able to interact with it in plain English.
Please see https://ollama.com/library for the list
of all supported models.
The command "ollama list" lists all models downloaded
into your system.
When the model fails to load into your GPU, please use
the provided ollama-limit-gpu-layers script to create
model flavors with different num_gpu parameters.
ollama uses many gigabytes of disk space in your home directory,
because advanced AI models are often very large.
Please symlink ~/.ollama to a large disk if needed.
Working examples:
(1) Coding with the model gpt-oss:20b:
1. start ollama service with 'sudo service ollama start'
or
setup and start the service 'ollama'
2. install claude-code and run:
ANTHROPIC_BASE_URL=http://localhost:11434 \
ANTHROPIC_AUTH_TOKEN=ollama \
ANTHROPIC_MODEL=gpt-oss:20b \
ANTHROPIC_DEFAULT_SONNET_MODEL=gpt-oss:20b \
ANTHROPIC_DEFAULT_OPUS_MODEL=gpt-oss:20b \
ANTHROPIC_DEFAULT_HAIKU_MODEL=gpt-oss:20b \
claude
3. Ask it to write some program.
(2) Image generation with the model x/z-image-turbo:
1. start ollama service with 'sudo service ollama start'
or
setup and start the service 'ollama'
2. run:
ollama run x/z-image-turbo {textual description of the desired image}
Please note that the x/z-image-turbo model requires a minimum of 64GB of
RAM to run or a similar amount of VRAM on a Vulkan device.
There are also a lot of text-to-text models that you can chat with.
You might also want to install the 'gollama' package. gollama allows to
run ollama models very easily.
misc/ollama: update 0.32.5 → 0.34.0
MLX image generation is now officially disabled hy the upstream
with the promise that it will come back once issues are resolved.
The VULKAN option is now enabled again since compilation issues
were resolved.
ChangeLog: https://github.com/ollama/ollama/releases/tag/v0.34.0
misc/ollama: update 0.32.1 → 0.32.3
What's Changed:
* Fixed model downloads that stall before sending data.
* Improved integrations: restored Claude Code Channels,
fixed Anthropic thinking streams, and made Hermes
Desktop respect --force-build.
* Expanded GPU support with CUDA on Windows ARM64, B200
support through CUDA 12, and lower memory use on Linux
CUDA/ROCm iGPUs.
* Added chat, thinking, and tool calling support for
Laguna 2.1 models, including a Metal inference fix.
* Fixed GLM tool calls being silently dropped at the end
of generation.
* Updated the MLX and llama.cpp engines.
Changelog: https://github.com/ollama/ollama/compare/v0.32.1...v0.32.3
misc/ollama: add patches for: image generation hanging/timeout on slow CPUs,
etc.
Enable image gen via MLX - now your local llama can finally draw Beastie
in sunglasses:
$ ollama run x/z-image-turbo "FreeBSD Beastie in sunglasses drinking coffee at
the beach"
many: Unpin Go ports
* Ports that were pinned to a deprecated version of Go (1.23 or older)
have been unpinned.
* Ports that were pinned to a still-supported version of Go (1.24 or
newer) have been converted to requesting that as their minimum Go
version.
* Ports that had been forcibly deprecated for pinning an expired Go
version have been undeprecated.
misc/ollama: Add computational backends
Options CPU and VULKAN enable various CPU backends and the VULKAN backend.
CPU backends are for different generations of SIMD instructions.
Backends are loaded automatically when they are installed.
