fastino

/

GLiNER2.5-Decide

Like 401

Follow

fastino 518

The Install section says pip install gliner2, but the base profile is torch-free and the next line imports AutoExtractor, which fails without the local extra (see the install section in the GLiNER2 README). Also switch pipeline_tag from token-classification to text-classification so the Hub lists this classification checkpoint under the task it serves.

Thanks for the catch. I checked the install point against the GLiNER2 README: plain pip install gliner2 is torch-free and needs the local extra, so the install line is right.

I'd hold the rest of this PR. main has moved since it was opened, and merging it as-is would also revert recent deliberate changes: library_name: transformers, pipeline_tag: schema-extraction, the decision-model tag, and the agent.fastino.ai banner links. It would also delete the Fine-tuning section that was just added in #12.

I opened #13 with only the install-line change, on current main. If you want the pipeline_tag or banner changes, they're worth proposing separately so the model owners can decide on each.

Update: the install-line change from this PR landed via #13. The one remaining idea, pipeline_tag: text-classification, is now a single-line PR on current main in #14, for the model owners to decide on since #11 chose schema-extraction deliberately. The library_name, decision-model tag and banner changes aren't included because they would revert #10, #11 and the earlier banner commits. Thanks @bkinge for the suggestions.

I rebased this PR onto current main to resolve the conflict. Its diff is now a single line, pipeline_tag: schema-extraction -> text-classification. The install-line fix already landed via #13, and the library_name, decision-model tag and banner changes are left as they are on main (they were changed deliberately in #10 and #11). Thanks @bkinge for the suggestions.

@MattThomas-fastino this is ready to merge or close at your call. #11 chose schema-extraction to match the Gliner2ForSchemaExtraction pipeline name, while the Hub's metadata validator warns that schema-extraction is not in its official pipeline tag list. config.json is untouched either way.

Merging at the maintainers' request: card tag is now text-classification (the Hub's recognized task), everything else on main is unchanged. Thanks @bkinge !