New model store doubling model name

[ericcurtin]

Issue Description

$ bin/ramalama ls
Starting importing AI models to new store...
Imported /Users/ecurtin/.local/share/ramalama/models/ollama/smollm:135m -> /Users/ecurtin/.local/share/ramalama/store/ollama/smollm/smollm/snapshots/sha256-e0a34469bc46d5348026979b54d997a0be594c1edfd36928bb472d34f5a1f117/smollm:135m
Imported /Users/ecurtin/.local/share/ramalama/models/ollama/smollm:1.7b -> /Users/ecurtin/.local/share/ramalama/store/ollama/smollm/smollm/snapshots/sha256-fa9c77c540510d4ac24593497c6a56df798617c9eb94bd0e9434aea9d51c9b76/smollm:1.7b
NAME                        MODIFIED               SIZE
ollama://smollm/smollm:135m Less than a second ago 87.48 MB
ollama://smollm/smollm:1.7b Less than a second ago 944.83 MB
ecurtin@ecur arm64 12:58:36 0 cli-advancements
$ bin/ramalama ls
NAME                        MODIFIED      SIZE
ollama://smollm/smollm:135m 4 seconds ago 87.48 MB
ollama://smollm/smollm:1.7b 3 seconds ago 944.83 MB

Steps to reproduce the issue

Just run an "ls" on an existing installation

Describe the results you received

The model name is duplicated

Describe the results you expected

The model name is not duplicated

ramalama info output

NA

Upstream Latest Release

Yes

Additional environment details

No response

Additional information

No response

[ericcurtin]

@engelmi noticed this

Would be nice if we postponed next release until we fix this. The double model name

I did like how seamless this was... It just worked

[ericcurtin]

Also I wonder should we take the opportunity to remove the ":" character from all files names with the migration (illegal character on Windows file systems)

[engelmi]

@ericcurtin What do you mean by doubling model name? The smollm/smollm part is due to the structure of <source>/<organization>/<name> and if the organization is not given, then it falls back to using the model name as organization. We can easily remove this duplicate.

Also I wonder should we take the opportunity to remove the ":" character from all files names with the migration (illegal character on Windows file systems)

Good idea! The model store already does that for new pulls, but seems I missed that the file name of the old store can contain ":" characters.
I'll open a PR fixing this.

[ericcurtin]

Sounds like it's just an aesthetic change to "ramalama ls" output.

Thanks for the explanations @engelmi . Feel free to ignore my comment about Windows, since this only happens for migrated systems, doesn't matter I guess.

[ericcurtin]

FWIW in the actual ollama pulling protocol, when a model has no organization they use "library/". We could have just copied that, it's implementation details. In theory you could have "smollm:135m" and "smollm/smollm:135m" as two different models in the ollama registry, so "library/" is probably more correct.

[engelmi]

Sounds like it's just an aesthetic change to "ramalama ls" output.

Yes. One aesthetic addition is the display of partially downloaded models. Such models are listed and have an (partial) postfix.

In theory you could have "smollm:135m" and "smollm/smollm:135m" as two different models in the ollama registry, so "library/" is probably more correct.

Oh, didn't know about that. And makes sense to change... should be done in the extract_model_identifier function for Ollama, probably. Also need to start refactoring and pruning old code soon.

Feel free to ignore my comment about Windows, since this only happens for migrated systems, doesn't matter I guess.

Added a small fix nonetheless :)
#1283

[ericcurtin]

Sounds like it's just an aesthetic change to "ramalama ls" output.

Yes. One aesthetic addition is the display of partially downloaded models. Such models are listed and have an (partial) postfix.

What we tend to do most of the time is take influence from podman, docker, ollama, etc. How do they display partial downloads?

In theory you could have "smollm:135m" and "smollm/smollm:135m" as two different models in the ollama registry, so "library/" is probably more correct.

Oh, didn't know about that. And makes sense to change... should be done in the extract_model_identifier function for Ollama, probably. Also need to start refactoring and pruning old code soon.

Feel free to ignore my comment about Windows, since this only happens for migrated systems, doesn't matter I guess.

Added a small fix nonetheless :) #1283

[ericcurtin]

Another model store related thing:

$ ramalama ls
NAME                                                                                     MODIFIED    SIZE
huggingface://bartowski/google_gemma-3-4b-it-GGUF/google_gemma-3-4b-it-IQ2_M.gguf:latest 4 hours ago 1.43 GB

huggingface models shouldn't have ":latest"

[engelmi]

What we tend to do most of the time is take influence from podman, docker, ollama, etc. How do they display partial downloads?

As far as I know HF CLI doesn't offer listing the models and I think Ollama will not display them. Personally, I'd prefer listing them as well since they can hog quite some amount of memory and this way I can clean them up. Maybe we can only display the partials when using a --all option for the ls command? WDYT? @ericcurtin

huggingface models shouldn't have ":latest"

@ericcurtin HF models can be tagged as well and, similar to podman, I chose latest as default if none is given. I think that main is the default, though. Do you mean this difference or is there another reason?

[rhatdan]

Is this still a valid issue?