pip3 on macOS can install in funny directories

[ericcurtin]

Like:

/Users/ecurtin/Library/Python//3.9/bin/omlmd

[ericcurtin]

@tarilabs omlmd doesn't really work great for me at least on macOS, which is a pity, it's a very popular platform for model inferencing...

$ ./ramalama.py pull oci://quay.io/mmortari/gguf-py-example:v1
Downloading quay.io/mmortari/gguf-py-example:v1...
/Users/ecurtin/Library/Python/3.9/lib/python/site-packages/urllib3/init.py:35: NotOpenSSLWarning: urllib3 v2 only supports OpenSSL 1.1.1+, currently the 'ssl' module is compiled with 'LibreSSL 2.8.3'. See: urllib3/urllib3#3020
warnings.warn(

[tarilabs]

That's weird because I made all the demos of omlmd on Mac OSX 👉 https://containers.github.io/omlmd/demos/demo/

[tarilabs]

Is the installation you are doing in a pyenv or venv?
I think in general, or at least my experience, using the global Mac OS Python environment is in general discouraged.

For another instance this is also analogously highlighted for ilab, here: https://github.com/instructlab/instructlab#:~:text=NOTE%3A%20The%20following%20steps%20in%20this%20document%20use%20Python%20venv%20for%20virtual%20environments.

Happy to follow-up online to double check which steps you followed and potential more robust alternatives?
lmk!

[ericcurtin]

Maybe it's just a harmless warning and https://quay.io/repository/mmortari/gguf-py-example isn't a real model

[tarilabs]

and https://quay.io/repository/mmortari/gguf-py-example isn't a real model

if I recall I've taken that from the gguf example of llama.cpp repo and a valid gguf file but indeed not a Large Language Model

[tarilabs]

Maybe it's just a harmless warning

I don't get those warning anyways on Mac OSX during my demo recordings, so I think it's a matter of how the library/dependencies are installed maybe?

[rhatdan]

LGTM

[tarilabs]

fwiw, just tried straight from repo root now and with:

./ramalama.py pull oci://quay.io/mmortari/gguf-py-example:v1

I get just-and-only:

Downloading quay.io/mmortari/gguf-py-example:v1...

which then shows in my storage:

% ./ramalama.py list                                          
NAME                                                   MODIFIED    SIZE
oci://quay.io/mmortari/gguf-py-example/v1/example.gguf 1 month ago 4.0K

I believe the warn from #189 (comment) maybe depends on the local system environment and configuration used then?

Possibly some pip install ... was done against the Mac native global Python environment?
On Mac, my understanding that is discouraged,
and I suggest using a Python environment like example from InstructLab,
or use pipx

my2c and happy to discuss further!

[tarilabs]

...I think I see now; seems to me install.py is making use of the default pip3 from the default Mac global python environment:

ramalama/install.py

Lines 79 to 80 in 07555d5

	subprocess.run(["pip3", "install", "huggingface_hub[cli]"], check=True)
	subprocess.run(["pip3", "install", "omlmd==0.1.4"], check=True)

what about:

1. using pipx instead of installing in the Mac global python environment
   i.e. make for Mac pipx a depedency as homebrew brew

2. just "shading" omlmd sources inside the ramalama
   after all is just an sdk

3. I could check what distributing omlmd through brew entails

please notice 2, 3 wouldn't however solve the same issue for the other dependency huggingface_hub[cli] or any future dependencies, which imho should not be installed on the end-user default Mac global python environment my2c

hence, I believe option 1 is the most preferable for the dependencies *of ramalama when installed/used on Mac.

always happy to discuss further !

edit: typo

[ericcurtin]

@tarilabs if you could try and package ramalama and omlmd via brew that would be appreciated! We should migrate to there eventually...

[tarilabs]

could try and package ramalama and omlmd via brew that would be appreciated! We should migrate to there eventually...

noted @ericcurtin eventually this will also impact however other dependencies of ramalama itself, as I noted in #189 (comment)