v0.2.3

What's Changed

• Add token accounting for inference requests by @yogabbagabb in #14
• Make context-size budget configurable via SYNTHEFY_MAX_ELEMENTS_BUDGET by @sagarwal-atg in #15
• Add OpenAI-compatible usage block to inference output by @yogabbagabb in #16
• Enable HF download tracking; scrub stale classification + gated-repo docs by @sagarwal-atg in #18
• Fix Truss deploy: bundle token_accounting so production /predict stops 500ing by @yogabbagabb in #19
• Add benchmark reproduction harness and pin cu128 torch build by @realPohanLi in #21
• Release 0.2.3 by @yogabbagabb in #24

New Contributors

• @realPohanLi made their first contribution in #21

Full Changelog: v0.2.2...v0.2.3

v0.2.1

Private build of synthefy-tabular v0.2.1. Install: see docs/private-distribution.md

v0.2.0

Private build of synthefy-tabular v0.2.0. Install: see docs/private-distribution.md

v0.1.0

First public release of synthefy-tabular.

Install

pip install synthefy-tabular

Authentication

The default model checkpoint at Synthefy/synthefy-tabular on Hugging Face currently requires access. See the README for details on requesting access and providing an HF token.

If you have a local checkpoint, pass model_path= to skip the Hugging Face download.

Usage

from synthefy_tabular import SynthefyTabularRegressor
model = SynthefyTabularRegressor()
model.fit(X_train, y_train)
pred = model.predict(X_test)