# DeepSeek-Coder-V2 16B / 236B: MoE Code Model + Multi-File Context

> Source: https://sukruyusufkaya.com/learn/fine-tuning-cookbook/ftc-deepseek-coder-v2-moe-code
> Updated: 2026-08-22T00:46:26.797Z
> Category: Fine-Tuning Cookbook (Model-by-Model)
> Module: Part VIII — Code Models & Repo-Level FT
**TLDR:** DeepSeek-Coder-V2 (DeepSeek 2024) — MoE arch (16B / 236B), Apache 2.0 lisansla en güçlü açık code LLM'lerden. 338 programming language, 128K context, multi-file repo understanding. RTX 4090'da 16B (2.4B active) QLoRA mümkün; 236B cloud only.