The largest open code dataset ever assembled: 900 billion tokens across 600+ programming languages sourced from GitHub, with an opt-out mechanism for developers. Powers StarCoder2 and is the go-to pretraining corpus for code LLMs. Includes permissively-licensed repos with deduplication and near-deduplication passes.
| Provider | BigCode |
| Category | 代码 |
| Size | 900B Tokens |
| License | Various (per-file) |
| Downloads | 350k |
| Tags | Source-Code, 600+ Languages, GitHub, BigCode |
from datasets import load_dataset
ds = load_dataset("bigcode/the-stack-v2")