We have now released Mellea 0.4.0 alongside three Granite Libraries: granitelib-rag-r1.0,granitelib-core-r1.0,granitelib-guardian-r1.0. Together, these releases make it easier to construct structured, verifiable, and safety-aware AI workflows on top of IBM Granite models.
Mellea is an open-source Python library for writing generative programs — replacing probabilistic prompt behavior with structured, maintainable AI workflows. Unlike general-purpose orchestration frameworks, Mellea is designed to make LLM-based programs maintainable and predictable through constrained decoding, structured repair loops, and composable pipelines (Recent to Mellea? Start with our introductory blog and meet the team)
Mellea 0.4.0
Mellea 0.4.0 is the newest release of an open-source research project initiated and developed by IBM Research. Constructing on 0.3.0 foundational libraries and workflow primitives, 0.4.0 expands the library’s integration surface and introduces recent architectural patterns for structuring generative workflows.
What’s included:
- Native integration with the Granite Libraries, offering a standardized API that relies on constrained decoding to ensure schema correctness.
- Instruct-validate-repair pattern via rejection sampling strategies
- Observability hooks for event-driven callbacks to observe and track workflows
See full list of Mellea 0.4.0 features and updates here
What Are the Granite Libraries
Simply put, a Granite Library is a set of specialised model adapters designed to perform well-defined operations on portions of an input chain or conversation. Somewhat than counting on general-purpose prompting, each specialized model is fine-tuned for a particular task reminiscent of query rewriting, hallucination detection, or policy compliance checking. The usage of specialized adapters allows us to extend the accuracy of every task at a modest parameter count cost, and without disrupting the bottom model’s capabilities.
Three libraries are being released today for the granite-4.0-micro model, each targeting a definite set of pipeline tasks and comprised of a set of LoRA adapters:
- Granitelib-core-r1.0: targeting the step of necessities validation in Mellea’s instruct-validate-repair loop
- Granitelib-rag-r1.0: targeting a wide range of tasks in agentic RAG pipelines covering pre-retrieval, post-retrieval, and post-generation
- Granitelib-guardian-r1.0: specialized models for safety, factuality, and policy compliance
Mellea 0.4.0 is the newest release of an open-source research project initiated and developed by IBM Research. Constructing on the foundational libraries and workflow primitives introduced in 0.3.0, this release expands the combination surface and introduces recent architectural patterns for structuring generative workflows.
