What's New in Mellea 0.4.0 + Granite Libraries Release

IBM Research released Mellea 0.4.0 alongside three Granite Libraries designed to build structured, verifiable, and safety-aware AI workflows on IBM Granite models.

Mellea is an open-source Python library for writing generative programs that replaces probabilistic prompt behavior with structured, maintainable AI workflows through constrained decoding, structured repair loops, and composable pipelines.

The three Granite Libraries released are granitelib-rag-r1.0, granitelib-core-r1.0, and granitelib-guardian-r1.0, each targeting distinct pipeline tasks for the granite-4.0-micro model.

A Granite Library is a collection of specialized LoRA adapters designed to perform well-defined operations, with each adapter fine-tuned for specific tasks like query rewriting, hallucination detection, or policy compliance checking.

Granitelib-core-r1.0 targets requirements validation in Mellea's instruct-validate-repair loop, while granitelib-rag-r1.0 handles pre-retrieval, post-retrieval, and post-generation tasks in agentic RAG pipelines.

Granitelib-guardian-r1.0 provides specialized models for safety, factuality, and policy compliance.

Mellea 0.4.0 introduces native integration with Granite Libraries through a standardized API using constrained decoding to guarantee schema correctness, an instruct-validate-repair pattern via rejection sampling strategies, and observability hooks for event-driven callbacks to monitor workflows.