How to install ollama on ubuntu with specific version

[MugdhaHardikar-GSLab]

I want to install the ollama on my ubuntu server but every few days new version of ollama gets installed. I want to fix the version of the ollama getting installed on my machine. Current install.sh doesn't seem to have that functionality. IS there any way?

[remy415]

What do you mean by fix the version of the ollama getting installed? Do you mean you want to keep a specific version installed?

[MugdhaHardikar-GSLab]

Yes. Previously I was using ollama 0.1.20, now I am getting 0.1.24. It want to stick to one version even if latest versions come up lets say too 0.1.20. How to do it.

Latest install giving me:
root@5882e714c697:/usr/src# ollama --version
ollama version is 0.1.24

[jmorganca]

By default Ollama won't auto upgrade on Linux

However, you can run this script to install a previous version:

curl -fsSL https://ollama.com/install.sh | sed 's#https://ollama.com/download#https://github.com/jmorganca/ollama/releases/download/v0.1.25#' | sh

Note this is experimental and may not work forever

[telemetrieTP23]

for me your commando for installing specific version does not work anymore, it allways installs the actual version (0.1.25) on my jetson orin AGX
even if i use:

curl -fsSL https://ollama.com/install.sh | sed 's#https://ollama.com/download#https://github.com/jmorganca/ollama/releases/download/v0.1.27#' | sh

on the jetson xavier agx i used it to install 0.1.17, after i recognized that starting with 0.1.18 it doesnt find the gpu drivers anymore, so i downgraded.

[remy415]

@telemetrieTP23 I'm working on adding Jetson support. In the mean time, I have a preliminary build available that should work on your Orin AGX until it's fully integrated into the official release: https://github.com/remy415/ollama.

To save you time, ensure that you set the following environment variables:
export LD_LIBRARY_PATH="/usr/local/cuda/lib64:/usr/local/cuda/compat:/usr/local/cuda/include"
export OLLAMA_SKIP_CPU_GENERATE="1"

Also set one of the following based on which Jetpack you are using:

L4T_VERSION.major >= 36: # JetPack 6
export CMAKE_CUDA_ARCHITECTURES="87"

L4T_VERSION.major >= 34: # JetPack 5
export CMAKE_CUDA_ARCHITECTURES="72;87"

L4T_VERSION.major == 32: # JetPack 4
export CMAKE_CUDA_ARCHITECTURES="53;62;72"

[remy415]

@telemetrieTP23 yes, you should be able to build from my fork.

I made a typo on the exports for CUDA_ARCHITECTURE. I corrected my original post, sorry about that.

It’s export CMAKE_CUDA_ARCHITECTURES

[kungfu-eric]

This works now:

curl -fsSL https://ollama.com/install.sh | sed 's#https://ollama.com/download/ollama-linux-${ARCH}${VER_PARAM}#https://github.com/ollama/ollama/releases/download/v0.1.33-rc5/ollama-linux-amd64#' | sh

By default Ollama won't auto upgrade on Linux

However, you can run this script to install a previous version:

curl -fsSL https://ollama.com/install.sh | sed 's#https://ollama.com/download#https://github.com/jmorganca/ollama/releases/download/v0.1.25#' | sh

Note this is experimental and may not work forever