Artificial Intelligence1 min read

Granite Switch 4.1.1 — Update models like you update software

By · Published by Everything Blog

In short

Designed with full backwards compatibility, the release upgrades the automated speech recognition system alongside the Decision adapter in granitelib-core. The updated speech model delivers multilingual processing across 10 languages with native punctuation and capitalization to streamline downstream data pipelines. Furthermore, the refreshed Decision adapter provides improved benchmark results on the Jev Decision Index test set.

Key points

  • Modular architectures allow specialized adapters to be updated or swapped on a single end…: Modular architectures allow specialized adapters to be updated or swapped on a single endpoint without recomputing base models.
  • The patch maintains complete backwards compatibility, eliminating the operational need fo…: The patch maintains complete backwards compatibility, eliminating the operational need for complex system migrations during deployment.
  • Developers will eventually be able to build and ship custom adapters using an upcoming Ad…: Developers will eventually be able to build and ship custom adapters using an upcoming Adapter SDK and instructional tutorials.

Sources checked by our analysis pipeline: 1

Just over a week ago, we released Granite Switch 4.1, showing that our modular Switch architecture lets you configure a language model with specialized, cache-efficient adapters — adding new expertise on a single endpoint.

Granite Switch 4.1.1 is a patch release. The version number means what it means in software: fully backwards compatible with 4.1, no migration, incremental improvements under the hood.

- Updated Granite Speech ASR model — multilingual speech-to-text across 10 languages with native punctuation and capitalization, so transcripts are ready for downstream models without a formatting pass.

- Updated Decision adapter in granitelib-core — featuring improved overall performance on the Jev Decision Index test set.

While we are excited about these new features, what is notable is the speed and efficiency with which our teams could release this update. Apart from training the new ASR model and Decision adapter, nothing was recomputed: the base model is untouched. You compose the latest adapters you want and deploy.

This pattern can power far more than patch releases to an IBM-provided model. We're working on tutorials and an Adapter SDK…

Artificial Intelligence