# LLaVA-1.5 / 1.6 / OneVision: 2-Stage Training + Projector Pretrain + Instruction Tune

> Source: https://sukruyusufkaya.com/learn/fine-tuning-cookbook/ftc-llava-family-2-stage-training
> Updated: 2026-08-24T00:33:09.553Z
> Category: Fine-Tuning Cookbook (Model-by-Model)
> Module: Part VI — Vision-Language Multimodal FT
**TLDR:** LLaVA'nın klasik 2-stage training reçetesi: (1) Projector-only pretrain (LAION-CC-SBU 558K image-caption pair üzerinde), (2) End-to-end instruction tune (LLaVA-Instruct-150K + custom). Freeze strategy ablation (vision frozen vs unfrozen, LLM frozen vs unfrozen). RTX 4090'da LLaVA-1.6 Mistral 7B FT.

