We have the answers to your questions! - Don't miss our next open house about the data universe!

Meta goes head to head with OpenAI and GPT-4

 m de lecture
Meta goes head to head with OpenAI and GPT-4

Meta recently unveiled Code Llama-70B, an advanced version of its specialised language model for software development. With performance equivalent to or even better than GPT-4, this open source model goes head to head with the dominant model on the market.

Code Llama-70B an advanced version?

At the end of summer 2023, Meta announced Code Llama, an LLM for software development. Code Llama now comes in four versions: 7, 13, 34 and now 70 billion parameters.

Driven by 500 billion tokens, the first three models of Code Llama remain powerful, fast tools that are better suited to small configurations.

7B is capable of running on a single GPU. For its part, 70B has been trained with 1,000 billion tokens, and is presented by Meta as the perfect development assistant.

As well as 70B, Meta is also promoting two versions of Code Llama: Code Llama Python and Code Llama Instruct, in all possible versions (7, 13, 34 and 70). All these models are derived from Llama 2 and hosted by Hugging Face.

What are the advantages of Code Llama Python and Instruct?

Code Llama Python is optimised for a wide variety of programming tasks, including web scraping, data analysis and web development. Its versatility and performance make it a valuable tool for code generation.

Code Llama Instruct can perform numerous tasks and manipulations in natural language. Its capabilities include filtering, searching, sorting and manipulating data, as well as binary and factorial searching.

Llama-70B code more powerful than GPT-4?

To make its case, Meta says it has had its new model tested with HumanEval and Mostly Basic Python Programming (MBPP). HumanEval tests the model’s ability to complete code, while MBPP evaluates the model’s ability to write code based on a description.

As a result, Instruct 70B achieved a score of 67.8% on HumanEval, compared with 67% for GPT-4. Python 70B beat Llama 70B’s previous record of 62.4% to score 65.6%. These new records make Llama the most powerful open source LLM on the market.

If you’ve enjoyed this article, and you’re considering a career in Data Science or simply want to improve your skills in your field, don’t hesitate to check out our training offers or our blog articles on DataScientest.

Source : ai.meta.com


DataScientest News

Sign up for our Newsletter to receive our guides, tutorials, events, and the latest news directly in your inbox.

You are not available?

Leave us your e-mail, so that we can send you your new articles when they are published!
icon newsletter


Get monthly insider insights from experts directly in your mailbox