Code Llama joins the likes of OpenAI’s GPT-4, Amazon’s CodeWhisperer and Google’s yet-to-launch Alpha Code, which are all AI code generators. OpenAI’s ChatGPT and Google Bard chatbots can also help with coding but are not specialised for this purpose.
Why does this matter: AI code generators are one of the most popular use cases of generative AI because they have the potential to significantly reduce the workload of coders, especially on mundane tasks. What makes Meta’s Code Llama stand out from Google and OpenAI is that Meta has released its model for free under a community license, giving access to the model for both research and commercial use (with certain restrictions for large commercial use). While there are other such open-source LLMs for code generation, Meta claims that Code Llama performs better than these models in evaluation tests.
Article continues below ⬇, you might also want to read:
- Meta Releases New AI Model For Translation And Transcription In Dozens Of Languages
- Meta To Use Generative AI To Connect Users With Online Businesses
- Meta Anticipates Higher Expenses In 2023, Plans Continued Investments In AI And Metaverse For 2024
- Meta Partners With Microsoft, Unveils Open-Source AI Model ‘Llama 2’ For Research, Commercial Use
- Here’s Why We’re Not Convinced Meta’s LLaMa 2 Is Open-Source
“Programmers are already using LLMs to assist in a variety of tasks. The goal is to make developer workflows more efficient so that they can focus on the most human-centric aspects of their job, rather than repetitive tasks. We believe that AI models, and LLMs for coding in particular, benefit most from an open approach, both in terms of innovation and safety. Publicly available, code-specific models can facilitate the development of new technologies that improve peoples’ lives. By releasing code models like Code Llama, the entire community can evaluate their capabilities, identify issues and fix vulnerabilities.” — Meta
Copyright concerns: As with other use cases of generative AI, code generation has also found itself in copyright controversy. For instance, GitHub launched Copilot, an AI coding tool powered by OpenAI’s GPT-4, but GitHub and OpenAI are being sued because CoPilot allegedly reproduces copyrighted code.
Three sizes of Code Llama are available: “We are releasing three sizes of Code Llama with 7B, 13B and 34B parameters respectively. […] The three models address different serving and latency requirements. The 7B model, for example, can be served on a single GPU. The 34B model returns the best results and allows for better coding assistance, but the smaller 7B and 13B models are faster and more suitable for tasks that require low latency, like real-time code completion,” Meta explained.
Code Llama – Python and Code Llama – Instruct: Additionally, there are further fine-tuned variations of Code Llama:
- Code Llama – Python: This variation is fine-tuned for Python code. “Because Python is the most benchmarked language for code generation, and because Python and PyTorch play an important role in the AI community – we believe a specialized model provides additional utility,” Meta explained.
- Code Llama – Instruct: This variation is fine-tuned for understanding natural language instructions. “The model is fed a natural language instruction input and the expected output. This makes it better at understanding what people expect out of their prompts. We recommend using Code Llama – Instruct variants whenever using Code Llama for code generation since Code Llama – Instruct has been fine-tuned to generate helpful and safe answers in natural language,” Meta explained.