Developers introduced the local API jev_local for working with LFM models

Developers introduced jev_local — a local API interface designed for processing text and visual queries. The tool provides probabilistic estimates in Choice, Score, or Noul formats using the capabilities of the llama.cpp engine.

The system supports work with LFM 2.5 family models, ensuring integration for both text data and image analysis. During performance testing, the solution demonstrated a result of 8 points when processing queries for 8 hours of continuous operation.

To deploy the tool, automation scripts setup.sh and run.sh are provided, which configure the environment for local launch. The solution architecture is focused on performing calculations directly on the user side without accessing external cloud servers.

The tool allows integrating probabilistic analysis into local data processing pipelines. Using jev_local provides the ability to standardize model output when classifying content or extracting structured information in JSON format.