spinf-serve 0.1.2

A documentation and metadata release; no code change.

• The README leads with bulk classification and decision inference on one GPU, says why the engine exists and shows a real response; it is a quarter shorter.
• The PyPI page now links to the repository, the docs, the models and the issues ([project.urls]).
• The model cards and ACKNOWLEDGEMENTS point at docs/; the 12B card's GPU list is corrected.
• The FAQ explains fp8s and w8a16 at first use.

pip install "spinf-serve[server]" then spinf-serve serve spinf/gemma-4-12b-fp8. The image ghcr.io/spinfinc/spinf-serve:0.1.2 is built from this release.

spinf-serve 0.1.1

The first public release of spinf-serve: the inference engine and OpenAI-compatible server for decision models. One input read once, hundreds of questions scored in parallel on one GPU, with batch-invariant fp8 scoring, extraction and generation slots, media, prompt packs, and the same API as the hosted spinf service.

pip install "spinf-serve[server]" then spinf-serve serve spinf/gemma-4-12b-fp8. Models: spinf/gemma-4-12b-fp8, spinf/gemma-4-31b-fp8, spinf/qwen3.8-27b-int4, spinf/pplx-decider-v1-27b-int4 on the Hugging Face Hub.

0.1.1 is 0.1.0 plus one fix: the official pack source is fetched from GitHub with the default Accept header (the tarball endpoint refused the previous one); 0.1.0 was tagged but not published to PyPI.