Code Alpaca

lastmod 2023-03-27created 2023-03-271 edits

LLaMA trained on code generation instructions

Features

• A model based on Stanford Alpaca 7B/13B, tuned so developers can use it well for coding tasks
• Replaced with 20K Instruction Folowing data related to code generation (using the Self-Instruct technique)
• Partially modified the data generation pipeline: changed the prompts to relate to code generation/editing/optimization rather than general tasks
• Fine-tuned with Hugging Face training code and Deepspeed

See also

• LLaMA
• Alpaca

Favorite site

• Github - codealpaca