Meta AI announced the launch of ‘Code Llama,’ a community-licensed artificial intelligence (AI) coding tool built on the Llama2 large language model (LLM), on Aug. 24.
The new tool is a fine-tuned version of LLama2 that’s been trained specifically for the purpose of generating and discussing computer code.
According to a blog post from Meta, Code Llama is split into several variants with one model fine-tuned for general coding in a number of languages (including Python, C++, Java, PHP, Typescript, C#, Bash and more).
Other models include Code Llama Python and Code Llama Instruct. The former is fine-tuned for Python applications. As Meta puts it, this is to further support the AI community:
“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.”
According to Meta, the emphasis in these first two model variants is on understanding, explaining, and discussing code.
Code Llama Instruct, however, is the fine-tuned version of Code Llama that Meta recommends for actually generating code. According to the blog post, it’s been engineered specifically to generate “helpful and safe answers in natural language.”
The models are also available in different parameter sizes in order to operate in different environments. Code Llama comes in 7-billion, 14-billion, and 34-billion parameter sizes, each with different functionality.
Meta says the 7B models, for example, can run on a single GPU. While the 14B and 34B models would require more substantial hardware, they’re also capable of more complex tasks — especially those requiring low-latency feedback such as real-time processes.
CodeLlama — a version of Llama2 that was fine-tuned for code tasks is live now. Available in 7B, 13B and 34B. pic.twitter.com/LEz9z44IkI
— Soumith Chintala (@soumithchintala) August 24, 2023
Code Llama is generally available under the same community license agreement as Llama2, meaning it can be used for personal or business use with proper attribution.
This could be a massive boon for businesses and individuals who have a high need use-case for LLM models for coding purposes, such as fintech institutions that are traditionally underserved by the AI and big tech communities.
Web3 innovators, trading bot developers, and cryptocurrency exchanges all operate in a constantly-shifting environment that, to date, has seen relatively little in the way of dedicated B2B or B2C solutions for day-to-day crypto and blockchain coding problems from big tech.
Related: Naver Corp unveils South Korea’s answer to ChatGPT and generative AI
Dedicated coding tools, such as GitHub’s Co-Pilot (built with ChatGPT technology) can go a long way towards aiding developers in these underserved areas, but the costs of use can be prohibitive for some users and the lack of open-source options can pose problems for proprietary software developers.
The existence of a free-to-use, community-licensed alternative based on Meta’s highly-touted Llama2 LLM could help level the playing field for blockchain and crypto projects with small development teams.
