DeepSeek Coder
ReleasedState-of-the-art open-source code generation model
Released on 2024.01.17
Overview
DeepSeek Coder is a series of code language models trained from scratch on 2T tokens, with a composition of 87% code and 13% natural language. It achieves state-of-the-art performance among open-source code models.
Key Features
- Trained on 87% code data
- Supports 86+ programming languages
- Fill-in-the-middle capability
- 16K context window
Specifications
- Parameters
- 1.3B / 6.7B / 33B
- Architecture
- Transformer Decoder
- Context Length
- 16K tokens
- Training Tokens
- 2T tokens
- License
- DeepSeek License