IBM has released a family of its Granite AI models to the open-source community. The series of decoder-only Granite code models are purpose-built to write computer code for enterprise developers, with training in 116 programming languages. These Granite models range in size from 3 to 34 billion parameters in base model and instruction-tuned variants. They offer a range of uses, from modernizing older code with new languages to optimizing programs for on-device memory constraints, such as might be experienced when conforming for mobile gadgets. In addition to generation, the models can repair and explain code.

In a blog post, IBM explains “all the models were trained on data that was collected in adherence with IBM’s AI ethics principles” as well as guidance from its legal team. They are available on Hugging Face, GitHub, watsonx.ai, as well as RHEL AI, Red Hat’s new platform for developing, testing, and deploying generative AI models.

Because these Granite models are licensed under Apache 2.0 for both research and commercial use, they are truly open source, explains ZDNet, contrasting them with other so-called “open” models (like Meta’s) which are for research purposes only.

“IBM has gone to great lengths to avoid copyright or legal issues,” ZDNet writes, noting it “managed the open sourcing of Granite code by using pretraining data from publicly available datasets, such as GitHub Code Clean, Starcoder data, public code repositories, and GitHub issues.”

The Granite Code Base models “are trained on 3- to 4-terabyte tokens of code data and natural language code-related datasets,” ZDNet explains.

IBM notes they can outperform certain models twice their size, like Code Llama, adding that “while some other models may perform slightly better in some tasks like code generation, no one model could perform at a high level at generation, fixing, and explanation — apart from Granite.”

IBM says the underlying base code for the Granite open models is the same as that used to train the Watsonx Code Assistant (WCA) for specialized domains. Introduced in May 2023, IBM Watsonx is the company’s AI development platform for enterprise. Demos are available at the Watsonx Studio. The Granite models debuted in September.

For those using the open-source Granite models, IBM has also released the Granite Code Instruct models, “the instruction methodology we used on the models to fine-tune, using a combination of Git commits paired with human instructions and open-source synthetically generated code instruction datasets.”

Forbes says the Granite open-sourcing has the potential “to democratize access to advanced AI tools” and should help speed the development process for “faster time-to-market and improved software quality.”

Related:
IBM Consulting Enhances Microsoft Copilot Capabilities, IBM, 5/9/24
IBM Launches Service to Help Enterprises Add GenAI Copilots, PYMNTS, 5/10/24
IBM and SAP Plan to Expand Collaboration to Help Clients Become Next-Generation Enterprises with Generative AI, SAP, 5/8/24
IBM and SAP Expand Partnership to Drive Generative AI Capabilities, ITPro, 5/13/24
ServiceNow and IBM Collaborate to Accelerate the Impact of Generative AI Across the Enterprise, IBM, 5/8/24
IBM Throws Its Red Hat into Open Source AI Ring with RHEL AI, TechTarget, 5/7/24
Red Hat Delivers Accessible, Open Source Generative AI Innovation with Red Hat Enterprise Linux AI, Business Wire, 5/7/24
Unlocking the Power of AI Across the Hybrid Cloud with watsonx and Fusion HCI, IBM, 5/13/24