IBM releases AI models that chat, make quick calls and transcribe speech in one
IBM says its new Granite Switch 4.1 models handle conversation, yes-or-no style judgments and speech-to-text from one online connection, using the memory of one model.
IBM says one copy of Granite Switch 4.1 can chat, make quick judgment calls, such as screening text for harmful content, and turn speech into written text. Developers reach all three through one online connection.
The extra skills are small add-ons attached to one main AI model. IBM says the combined model needs about the same chip memory as the main model alone, so it needs fewer chips. IBM also says its speech part is the fastest on the OpenASR Leaderboard, and its safety add-on beats several rival tools on average on safety tests.
In one demo, IBM says a small custom version played the video game Doom, making 60 to 100 moves per second. Jev, a rival system, reported 10. IBM says anyone can mix their own add-ons, and plans a toolkit for building more.
Official releaseRead the original at Hugging Face →This is mainly for software developers. IBM has released both models, a demo and its code, so they can try them now; most people won't see a direct change.
The facts
- IBM released Granite-Switch-4.1-8B and Granite-Switch-4.1-30B, which handle chat, decision tasks and speech-to-text from one hosted endpoint.
- IBM says the combined model takes up about the same GPU memory as the base model alone.
- IBM says the embedded Granite-Speech-5.0-Turbo-CTC is the fastest speech-to-text model on the OpenASR Leaderboard.
- IBM says its Guardian add-on, on average, outperforms Jev 1.13 and three other decision models on safety-related decision tasks.
- IBM says Granite-Switch-4.1-30B finished a real-world 5-step workflow faster than Jev-27B in a head-to-head race.
- IBM says its custom 3B Granite Switch model made 60 to 100 decisions per second in Doom, versus 10 per second reported for Jev.
- IBM says anyone can build their own Granite Switch model from Granite Libraries add-ons, and plans to release an Adapter SDK for training adapters in Mellea.
Sources
- Hugging Face: Granite Switch 4.1: Chat, Decisions, and Speech from One Endpoint
Published
More in New models
AmazonNew modelsAyobi’s pickReflection AI's Beam, a free-to-download coding AI, is now available on Amazon's cloud
Reflection AI released Beam, an AI built to write and fix software over long jobs. Anyone can download and change it, and Amazon now offers it to run on its cloud.
Black Forest Labs releases FLUX 3 Image Klein, an image AI for home computers
The new model makes and edits pictures up to 4K size. Black Forest Labs says it runs on an NVIDIA graphics card with 8 GB of memory.
New modelsAyobi’s pickNVIDIA releases open AI models, one built to run on your own PC
NVIDIA says its new Super VL model is five times faster than rivals, and its Bolt model can run on a home PC without the cloud.