Oracle just yesterday presented its Code Assist platform that uses LLM (Large Language Models) optimized for create programming code and help developers be more productive, ensuring the highest quality results. GitHub Copilot Workspace also aims to improve workflows related to software developmentleveraging the benefits of generative artificial intelligence.
IBManother company that is heavily investing in AI in the business sector, has just presented its generative models Granite, releasing the source publicly on GitHub. Overall, these tools have the primary objective of addressing the multi-faceted challenges that characterize computer programming.
What is the IBM Granite family of templates and what are their features
At the heart of Granite is a sophisticated architecture, capable of offering models with a number of parameters between 3 and 34 billion (3B-34B): the goal is to adapt the platform to each specific use case. Whether it’s the code generation basic or optimization tasks on complex instruction sets, IBM generative models are extremely versatile.
Yes, because write, test, debug and deploy reliable software remains a difficult task. Even the most experienced developer often finds himself looking for advice and “ways out”: i programming languages they are constantly being updated and new ways of developing are gradually taking hold.
I LLM (Large Language Models) trained from large volumes of programming code, they are revolutionizing the software development process. More and more LLMs are integrated into development environmentsin order to improve the productivity of every programmer while special agentgoverned by artificial intelligence, help manage complex tasks.
Make full use of the enormous potential of LLMs within the codingmeans combining code generation, the bug fixthe explanation and the documentation code, repository maintenance, and more.
IBM spokespeople say that the company has looked for (and identified) ways to make developers more productive by making them spend less time understanding because the code doesn’t work or, for example, how to connect code legacy (developed years before) with the most modern applications. Granite fits right into this groove, with the models that have been trained using code developed in 116 programming languages different.
What Granite models allow you to do
For many developers, writing code isn’t actually what takes up the majority of their time. Much more demanding and delicate activities are the code testing developed, ensuring that everything works as expected and, for example, identifying and correcting any bug.
With tools built using IBM Granite templates, professionals and businesses can take advantage of a competitive advantage not indifferent.
Many of the other daily tasks that are part of a developer’s day – from generating unit tests, to writing documentation, to running vulnerability tests – can be automated with IBM’s templates. AI can also analyze the codeexplain why it doesn’t behave as expected, and provide guidance on resolving the problem.
According to IBM, Granite templates help modernize applications mission-critical which must remain safe, resilient and, above all, always available online.
By leveraging the mechanisms behind Granite, developers can, for example, convert old COBOL code into new ones more modern languagescome Java. It is one of the points on which, in recent years, IBM has focused most.
Performance at the highest levels
According to IBM technicians, Granite stands out for the maximum level performance it is capable of expressing. Big Blue “certifies” that the performance of Granite can surpass some models characterized by double the size, such as Code Llama.
IBM’s evaluations also highlight that while some third-party models can perform slightly better in some tasks, such as code generation, no model to date would be able to match Granite’s performance in the combined tasks of code generation, correction and explanation. code. The behavior affects code processing in major languages, including Python, JavaScript, Java, Go, C++, and Rust.
Engineers also tend to underline how the “mixture” of reliable sources used to train Granite is a real strong point.
The future of generative AI in software development
The dawn of AI-driven software development is approaching, says IBM, which today finds itself at the forefront. In fact, it delivers to the community a set of advanced models on which countless projects can be built.
“We’re excited to see what will be built with these templates, whether it’s new code generation tools, cutting-edge editing software, or everything in between“, we read in the note published by the company.
However, IBM is keen to point out that Granite is “a taste” of what’s coming next: in the weeks to come, the company will share further information on other models that will help shape the future of computing.
In the meantime, Granite models can be downloaded from GitHub while HuggingFace is the platform that allows their testing.
The images in the article are taken from IBM’s post.