Models get patch releases too: Granite Switch 4.1.1 swaps adapters, not the base model, to ship new speech and decision skills
Synopsis
IBM's Granite Switch 4.1.1 arrives as a patch release, fully backwards compatible with 4.1 and requiring no migration. It updates a Granite Speech transcription model covering 10 languages with native punctuation and capitalization, and a Decision adapter in granitelib-core with improved overall performance on the Jev Decision Index test set. Apart from training those two new adapters, nothing was recomputed: the base model is untouched, and users compose the latest adapters and deploy.
Interpretation
Granite Switch 4.1.1 is a patch release, and the version number means what it means in software: fully backwards compatible with 4.1, no migration, incremental improvements under the hood. Just over a week earlier, Granite Switch 4.1 showed that the modular Switch architecture lets a language model be configured with specialized, cache-efficient adapters, adding new expertise on a single endpoint. The update carries two concrete components: a Granite Speech ASR model covering 10 languages with native punctuation and capitalization, and a Decision adapter in granitelib-core with improved overall performance on the Jev Decision Index test set.
The update is light because the base model stays untouched and only new adapters are trained; users compose the latest adapters they want and deploy. The modular Switch architecture places specialized expertise in cache-efficient adapters rather than changing the base model itself. Apart from training the new ASR model and Decision adapter, nothing was recomputed, and the release states that tutorials and an Adapter SDK are in progress so developers and third-party providers can build their own adapters.
Perspective
The demonstrated scope is the IBM Granite Switch 4.1.1 patch release itself: backwards compatible with 4.1 with no migration, containing the multilingual speech transcription and Decision adapter updates, with the base model not recomputed. The release says tutorials and an Adapter SDK are still in progress.
The Decision adapter improvement is described only as improved overall performance on the Jev Decision Index test set, without a stated magnitude; the tutorials and Adapter SDK for third-party adapters are described as in progress and not yet available.
