(Newswire.net — November 17, 2023) — Code-generation large language models (LLMs) are essential for software developers and businesses today in a growing competitive landscape that is continually being supplemented with AI technologies.
A LLM is a deep learning algorithm that can perform a variety of natural language processing (NLP) tasks. Large language models are trained using massive datasets, allowing them to recognize, translate, predict, and generate text or other content. LLMs can be trained to perform a variety of tasks like understanding protein structures, writing software code, and more. Like the human brain, large language models must be pre-trained and then fine-tuned so that they can provide adequate solutions to the complex problems they are asked to solve.
Such existing LLMs, including ChatGPT, do not support the most modern code-sets like LangChain, nor do they include the latest thinking around coding methodologies, explains Iterate.ai Brian Sathianathan in their recent release.
Iterate.ai, whose AI innovation ecosystem enables enterprises to build production-ready applications, announced the availability of Interplay-AppCoder LLM, the company’s large language model for automated code generation.
“Our venture into training a code generation LLM on state-of-the-art generative AI frameworks and libraries like YOLO V8, LangChain, and VertexAI, has culminated in Interplay-AppCoder LLM,” states Brian Sathianathan, Iterate.ai’s CTO and Co-Founder. “The creation of Interplay-AppCoder LLM was made possible by the meticulous fine-tuning of CodeLlama-7B, 34B and Wizard Coder-15B, 34B on a bespoke dataset. Essentially, we are creating our own, updated version of ChatGPT specifically for developers to generate code at an enterprise level at levels that exceed what is currently available.”
When people utilize the capabilities of programs like ChatGPT, they may ask it to provide outlines, itineraries, or other more consumer-based tasks for the everyday person. Interplay-AppCoder is specifically created for developers and to scale at the enterprise level.
Other existing coding models on the leaderboard include Meta’s Code Llama, the AI model built on top of Meta’s Llama 2 LLM and finely tuned for code generation and discussion, and WizardCoder, an open-source coding assistant.
According to the release, as of mid-October 2023, WizardCoder stood as the dominant coding LLM, which had beaten Code Llama in testing. Iterate’s newly launched Interplay-AppCoder LLM has successfully been tested to out-code previous leaderboard platforms WizardCoder.
The initial release of the Interplay-AppCoder model’s performance scored high on the ICE Benchmark, a methodology that focuses on usefulness and functionality:
- Usefulness – 2.968 of 4.0 (52% higher than WizardCoder which scored 1.825)
- Functionality – 2.476 of 4.0 (440% higher than WizardCoder which scored 0.603)
The ICE Benchmark is a new evaluation metric via instructing large language models (LLMs) for code assessments. Their metric addresses the limitations of existing approaches by achieving superior correlations with functional correctness and human preferences, without the need for test oracles or references. They evaluate the efficacy of our metric on two different aspects (human preference and execution success) and four programming languages. The results demonstrate that their metric surpasses state-of-the-art metrics for code generation, delivering high levels of accuracy and consistency across various programming languages and tasks. They also make their evaluation metrics and datasets available to the public, encouraging further research in evaluating code intelligence tasks.
Interplay-AppCoder LLM was created through fine-tuning of CodeLlama-7B, 34B and Wizard Coder-15B, 34B on a bespoke dataset of newer generative AI libraries such as LangChain, YOLO V8, and Vertex AI, according to the release.
“Businesses can harness Interplay-AppCoder to stay ahead of the competition. Companies need the best tools available to them to work with generative AI, and the current leader is Iterate’s Interplay-AppCoder, which reduces time-to-market by automating code generation and extending support for the latest models and libraries. By automating the coding process, businesses can spend their time and resources focusing on strategic initiatives, fostering innovation and growth,” Sathianathan states.
Currently, Iterate.ai is building several private LLMs for large enterprises across the United States and Asia and aims to constantly beat benchmarks and continue to innovate, a need in a constantly changing and growing AI sector.