Medium AI Coding · 1 min read
Code Llama 70B: Meta brings the Llama 2 code branch to GPT-3.5 level
In one sentence Meta releases Code Llama 70B (base, Python, Instruct), the largest member of the code-specialized family derived from Llama 2, with HumanEval results comparable to GPT-3.5.
Reading level
Meta releases a much larger version of Code Llama, its coding-specialized model. It had been announced months earlier; the 70-billion parameter size finally lands.
There are three variants: a base, one specialized in Python, and an "Instruct" version that follows natural-language requests like "write me this function."
The model is free to download with a license that allows commercial use, so developers can run it on powerful machines or fine-tune it without paying third-party APIs.
Companies
Meta
Tools
Code Llama, Code Llama 70B
Tags
MetaCode LlamaOpen SourceCodingHumanEval
Sources