This is the sequel of StarCoder and The Stack (v1) - https://news.ycombinator.com/item?id=35954481 https://news.ycombinator.com/item?id=35954481 StarCoder2 has

This is the sequel of StarCoder and The Stack (v1) - https://news.ycombinator.com/item?id=35954481 https://news.ycombinator.com/item?id=35954481 StarCoder2 has 3 models, biggest one is StarCoder2-15B, trained on 600+ programming languages from The Stack v2 on 4+ trillion tokens. The Stack2 is largest open code dataset suitable for LLM pretraining.