Abstract: In this session, we will present the methods used to train Code Llama, the performance we obtained, and show how you could use Code Llama in practice for many software development use cases.
Code Llama is a family of open large language models for code based on Llama 2 providing state-of-the-art performance among open models, infilling capabilities, support for large input contexts, and zero-shot instruction following ability for programming tasks. We provide multiple flavors to cover a wide range of applications: foundation models (Code Llama), Python specializations (Code Llama - Python), and instruction-following models (Code Llama - Instruct) with 7B, 13B, 34B, and now 70B parameters each. Code Llama reaches state-of-the-art performance among open models on several code benchmarks. Notably, Code Llama - Python 7B outperforms Llama 2 70B on HumanEval and MBPP, and all our models outperform every other open model on MultiPL-E. Code Llama was released under a permissive license that allows for both research and commercial use.
This session will show how to use Code Llama locally with ollama, or to complete your code in VSCode. We will also show how to prompt it to generate executable code using our largest models available on public Chatbots.
Bio: Baptiste is a research scientist at Meta AI in Paris working in the code generation team. He contributed to Llama and led Code Llama.
At Meta, Baptiste conducted research on unsupervised translation of programming languages and model pre-training for code. His work was featured in dozens of news articles in more than ten languages. He also started a collaboration between the Fundamental AI Research department and production teams putting code models in production.
Prior to Meta, Baptiste worked as an applied scientist in advertising at Amazon.