feat: Add model-util CLI

[rafvasq]

• Supersedes ✨ add tgis-cli tools IBM/vllm#16, feat: add CLI tools IBM/vllm#52, feat: add TGIS CLI commands vllm#92

• Related to: [Feature] vLLM CLI for serving and querying OpenAI compatible server vllm-project/vllm#5090

Changes:

• Adds commands to the adapter using either entry point model-util or text-generation-server
  • model-util download-weights
  • model-util convert-to-safetensors
  • model-util convert-to-fast-tokenizer

[dtrifiro]

Please see the contributing guide for information on how to set up pre-commit so that CI does not fail

[dtrifiro]

vllm@main CI will keep failing until #68 solves the problem and is merged

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 15.35088% with 193 lines in your changes missing coverage. Please review.

Project coverage is 54.70%. Comparing base (7cd77d1) to head (fbb0cd7).

Files	Patch %	Lines
src/vllm_tgis_adapter/tgis_utils/scripts.py	0.00%	104 Missing ⚠️
src/vllm_tgis_adapter/tgis_utils/hub.py	25.26%	71 Missing ⚠️
tests/test_hub.py	37.93%	18 Missing ⚠️

❗ There is a different number of reports uploaded between BASE (7cd77d1) and HEAD (fbb0cd7). Click for more details.

HEAD has 3 uploads less than BASE

Flag	BASE (7cd77d1)	HEAD (fbb0cd7)
	6	3

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #59      +/-   ##
==========================================
- Coverage   62.08%   54.70%   -7.38%     
==========================================
  Files          21       24       +3     
  Lines        1216     1444     +228     
  Branches      213      259      +46     
==========================================
+ Hits          755      790      +35     
- Misses        386      579     +193     
  Partials       75       75              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[prashantgupta24]

@rafvasq can you trim down the commits?

[rafvasq]

Trimmed them down, @prashantgupta24

[njhill]

Thanks @rafvasq, and thanks @prashantgupta24 @dtrifiro for all of the reviews. This is looking great!

I know there was already some discussion about this above but I'm wondering whether we should make the command something different. Maybe even just make it text-generation-server since these are mainly for backwards compatibility?

Or even something like model-util? Even better would I guess would be to have both, i.e. keep text-generation-server as an alias if that's possible, for backwards compatibility, but we can just document the new command.

[prashantgupta24]

Thanks @rafvasq, and thanks @prashantgupta24 @dtrifiro for all of the reviews. This is looking great!

I know there was already some discussion about this above but I'm wondering whether we should make the command something different. Maybe even just make it text-generation-server since these are mainly for backwards compatibility?

Or even something like model-util? Even better would I guess would be to have both, i.e. keep text-generation-server as an alias if that's possible, for backwards compatibility, but we can just document the new command.

I was a bit hesitant in using text-generation-server just because it might confuse folks who have worked with TGIS earlier?

[njhill]

Thanks @rafvasq, and thanks @prashantgupta24 @dtrifiro for all of the reviews. This is looking great!
I know there was already some discussion about this above but I'm wondering whether we should make the command something different. Maybe even just make it text-generation-server since these are mainly for backwards compatibility?
Or even something like model-util? Even better would I guess would be to have both, i.e. keep text-generation-server as an alias if that's possible, for backwards compatibility, but we can just document the new command.

I was a bit hesitant in using text-generation-server just because it might confuse folks who have worked with TGIS earlier?

Imho it shouldn't introduce confusion, just would hopefully allow their existing scripts/processes to work without change. But if it's not straightforward to make an alias then I guess not a huge deal.

[rafvasq]

@njhill, sorry I forgot to mention it but I did introduce the text-generation-server alias:

https://github.com/opendatahub-io/vllm-tgis-adapter/pull/59/files#diff-50c86b7ed8ac2cf95bd48334961bf0530cdc77b5a56f852c5c61b89d735fd711R50

[dtrifiro]

Thanks @rafvasq. This looks very good at this point, just a few nits left