JevEmbed v0.1.0

JevEmbed v0.1.0

Initial source release of JevEmbed, a Python framework for embedding-based Choice, Score, and Noul decisions.

Included

• Python API, command-line interface, and optional HTTP server using Jev-style requests and responses.
• Configurations for KaLM, Qwen3 Embedding (0.6B, 4B, and 8B), multilingual E5, and CLM-v0.1-8B, plus configurations for the separately published JevEmbed KaLM and Qwen3 0.6B models.
• LoRA fine-tuning tools, supervised-data validation, Open-Jev conversion, and evaluation utilities.
• Supervised data synthesis for a fixed Choice, Score, or Noul question, with configurable label quotas, validation, reporting, and resume support.
• Examples and guides for installation, inference, training, synthesis, calibration, and HTTP use.

Get started

From the source checkout, install the tested local inference stack and run an example:

python -m pip install -r requirements.txt
python -m pip install -e . --no-deps
python -m jevembed --config configs/kalm-embedding-v2.5.yaml --input examples/official_choice_exchange.json

Model weights are hosted separately and are not attached to this GitHub release. The decision scores and probabilities are uncalibrated; see calibration guidance before interpreting them as likelihoods.

Source commit: 08e41f0. License: Apache-2.0.