DeepSeek-Coder-V2-Lite
Code generation model Generally available Open WeightsDeepSeek's 16B MoE (2.4B active) open-weights code model delivering 128k long-context code completion under the MIT license.
Overview
DeepSeek-Coder-V2-Lite Instruct is a lightweight open-weights code language model released by DeepSeek in June 2024. Built with 16 billion total parameters and activating 2.4 billion parameters per token via Mixture-of-Experts, the Lite variant provides efficient local code generation and editing.
The model features a 128,000-token context window with up to 4,096 output tokens. It supports over 338 programming languages and matches or exceeds the coding benchmark performance of significantly larger dense models.
DeepSeek-Coder-V2-Lite Instruct is published under the MIT License, allowing unrestricted commercial use and local self-hosting via vLLM, Ollama, and Hugging Face Transformers.
Technical Specifications
Context Window
128,000 tokens
Max Input Tokens
32,768
Max Output Tokens
4,096
Parameters
16.0B
Released
2024-06-07
Modalities
Text → Text (Both)
Capabilities
Text Generation
Code Generation
Function Calling
Use Cases
Coding Assistance
Code Review
Languages
English (en)
Chinese (zh)
Access Routes
| Route | Type | Protocol | Region / Scope | Provider | Status |
|---|---|---|---|---|---|
| DeepSeek | Hosted API (Direct provider) | OpenAI-compatible REST API | Global | DeepSeek | Available |
Identifiers
- deepseek-coder Canonical identifier Default
- deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct Repository identifier Default