• We’re releasing Granite Speech 3.3 8B, a new speech-to-text (STT) model that excels in automatic speech recognition (ASR) and automatic speech translation (AST).
• The new audio model is built on top of Granite 3.3 8B Instruct, the latest update to our workhouse enterprise large language model (LLM). Alongside enhanced reasoning capabilities, the Granite 3.3 Instruct models now offer fill-in-the-middle (FIM) capabilities in addition to standard next-token prediction.
• To enhance existing Granite-driven applications, we’re also releasing a suite of retrieval augmented generation (RAG)-focused LoRA adapters for Granite 3.2. Feedback will inform development of LoRA adapters for Granite 3.3 Instruct, which will be released shortly, as well as for future generations of Granite LLMs.
• Alongside these conventional adapters, IBM Research has also developed a series of activated LoRAs (aLoRAs), an experimental new kind of low-rank adaption (LoRA) that cuts inference costs and memory requirements while enabling seamless switching between adapters.
• As always, all Granite models and tools are released open source under a standard Apache 2.0 license.
• All Granite 3.3 models and associated tools are available on Hugging Face. Granite 3.3 Instruct is also available on IBM watsonx.ai, as well as through platform partners including LMStudio and Replicate.

Read the full article.