IBM Granite

Granite
Developer(s)	IBM Research
Initial release	November 7, 2023; 6 months ago
Platform	IBM Watsonx (initially); GitHub; Hugging Face; RHEL AI
Type	Multimodal; Large language model; Generative pre-trained transformer; Foundation model;
License	Proprietary; Code models: Open Source (Apache 2.0)

IBM Granite is a series of foundation models created by IBM. It was announced on September 7, 2023,^[3]^[4] and an initial paper was published 4 days later.^[5] Initially intended for use in the IBM's cloud-based data and generative AI platform Watsonx along with other models,^[6] IBM opened the source code of some code models.^[7] Granite models are trained on datasets curated from Internet, academic publishings, code datasets, legal and finance documents.^[8]^[9]^[1]

Foundation models[edit]

A foundation model is an AI model trained on broad data at scale such that it can be adapted to a wide range of downstream tasks.^[10]

Granite's first foundation models were Granite.13b.instruct and Granite.13b.chat. The "13b" in their name comes from 13 billion, the amount of parameters they have as models, lesser than most of the larger models of the time.^[3] Later models vary from 3 to 34 billion parameters.

On May 6, 2024, IBM releases the source code of four variations of Granite Code Models and puts them on Hugging Face for public use, which outperform Llama 3 on several tasks within similar range of parameters.^[11]

References[edit]

^ ^a ^b McDowell, Steve. "IBM's New Granite Foundation Models Enable Safe Enterprise AI". Forbes.
^ ibm-granite/granite-code-models, IBM Granite, 2024-05-08, retrieved 2024-05-08
^ ^a ^b Nirmal, Dinesh (September 7, 2023). "Building AI for business: IBM's Granite foundation models". IBM.
^ "IBM debuts Granite series of hardware-efficient language models". September 7, 2023.
^ "Granite Foundation Models" (PDF). IBM. 2023-11-30.
^ Fritts, Harold (2024-04-22). "IBM Adds Meta Llama 3 To watsonx, Expands AI Offerings". StorageReview.com. Retrieved 2024-05-08.
^ Jindal, Siddharth (2024-05-07). "IBM Releases Open-Source Granite Code Models, Outperforms Llama 3". Analytics India Magazine. Retrieved 2024-05-08.
^ Azhar, Ali (2024-04-08). "IBM Patents a Faster Method to Train LLMs for Enterprises". Datanami. Retrieved 2024-05-08.
^ Wiggers, Kyle (2023-09-07). "IBM rolls out new generative AI features and models". TechCrunch. Retrieved 2024-05-08.
^ "Introducing the Center for Research on Foundation Models (CRFM)". Stanford HAI. 18 August 2021.
^ Jindal, Siddharth (2024-05-07). "IBM Releases Open-Source Granite Code models, Outperforms Llama 3". Analytics India Magazine. Retrieved 2024-05-08.

External links[edit]

[auto-1] McDowell, Steve. "IBM's New Granite Foundation Models Enable Safe Enterprise AI". Forbes.

[2] ibm-granite/granite-code-models, IBM Granite, 2024-05-08, retrieved 2024-05-08

[auto1-3] Nirmal, Dinesh (September 7, 2023). "Building AI for business: IBM's Granite foundation models". IBM.

[4] "IBM debuts Granite series of hardware-efficient language models". September 7, 2023.

[5] "Granite Foundation Models" (PDF). IBM. 2023-11-30.

[6] Fritts, Harold (2024-04-22). "IBM Adds Meta Llama 3 To watsonx, Expands AI Offerings". StorageReview.com. Retrieved 2024-05-08.

[7] Jindal, Siddharth (2024-05-07). "IBM Releases Open-Source Granite Code Models, Outperforms Llama 3". Analytics India Magazine. Retrieved 2024-05-08.

[8] Azhar, Ali (2024-04-08). "IBM Patents a Faster Method to Train LLMs for Enterprises". Datanami. Retrieved 2024-05-08.

[9] Wiggers, Kyle (2023-09-07). "IBM rolls out new generative AI features and models". TechCrunch. Retrieved 2024-05-08.

[10] "Introducing the Center for Research on Foundation Models (CRFM)". Stanford HAI. 18 August 2021.

[11] Jindal, Siddharth (2024-05-07). "IBM Releases Open-Source Granite Code models, Outperforms Llama 3". Analytics India Magazine. Retrieved 2024-05-08.

[1]

[2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

Foundation models[edit]

See also[edit]

References[edit]

External links[edit]