Error: RamaLama requires the "llama-run" command to be installed on the host when running with --nocontainer.

[dougsland]

(non root) ramalama run granite-code:latest    
os.execvp(llama-run, ['llama-run', '-c', '2048', '--temp', '0.8', '--jinja', '/Users/douglaslandgraf/.local/share/ramalama/models/ollama/granite-code:latest'])
Error: RamaLama requires the "llama-run" command to be installed on the host when running with --nocontainer.
RamaLama is designed to run AI Models inside of containers, where "llama-run" is already installed.
Either install a package containing the "llama-run" command or run the workload inside of a container.
[Errno 2] No such file or directory

OS triggered the command (no container):

sw_vers 
ProductName:		macOS
ProductVersion:		15.2
BuildVersion:		24C101

How I got this ?

brew install go
brew install podman
go install github.com/cpuguy83/go-md2man/v2@latest
python3 -m venv ~/.venvs/ramalama
source ~/.venvs/ramalama/bin/activate
pip install argcomplete
sudo make install
sudo ramalama run granite-code:latest
os.execvp(llama-run, ['llama-run', '-c', '2048', '--temp', '0.8', '--jinja', '/Users/douglaslandgraf/.local/share/ramalama/models/ollama/granite-code:latest'])
Error: RamaLama requires the "llama-run" command to be installed on the host when running with --nocontainer.
RamaLama is designed to run AI Models inside of containers, where "llama-run" is already installed.
Either install a package containing the "llama-run" command or run the workload inside of a container.
[Errno 2] No such file or directory

Should we make a make target focused in llama-run?

[dougsland]

@ericcurtin ideas? :)

[ericcurtin]

Wrote a comment on the PR version of this, it's basically just run this to install on macOS:

curl -fsSL https://raw.githubusercontent.com/containers/ramalama/s/install.sh | bash

but if can make our installation procedure better or docs better we should go for it.

[dougsland]

@ericcurtin please re-open, we need to document it somehow.

[dougsland]

curl -fsSL https://raw.githubusercontent.com/containers/ramalama/s/install.sh | bash
llama-run 
<SNIP>

Worked.

[rhatdan]

Not sure this was a bad outcome. The failure told you what had to be done?

[rhatdan]

@ericcurtin what do you want done with this one?

[ericcurtin]

It is expected outcome. My immediate thought would be just close.

Centralising on this install technique would help:

curl -fsSL https://raw.githubusercontent.com/containers/ramalama/s/install.sh | bash

There's even a suggestion on the above install technique someone added to README.md to suggest this works well on macOS. Closing, but we can continue discussions.

As discussed in:

#669

I wouldn't recommend the 10 line install technique from this issue or the #669 PR, installs many things only required for development (git, go, go-md2man, etc.), installs the community version of podman, it's overly complex, etc.