# DeepSeek-R1-Distill (Llama-8B / Qwen-7B): Reasoning Trace Distillation — 'Think Token'ları Öğrenmek

> Source: https://sukruyusufkaya.com/learn/fine-tuning-cookbook/ftc-deepseek-r1-distill-reasoning
> Updated: 2026-08-23T04:37:10.752Z
> Category: Fine-Tuning Cookbook (Model-by-Model)
> Module: Part III — Small Open Models (1B–8B)
**TLDR:** DeepSeek-R1-Distill — R1 (671B reasoning model) traces ile distilled Llama/Qwen base'ler. <think>...</think> token format, chain-of-thought trace dataset, R1 reasoning capability'sini 7-8B'ye sıkıştırma. RTX 4090'da kendi reasoning FT'ni yapmak: 1000 R1-traced example yeter.