https://www.youngju.dev/blog 천천히 올바르게. AI Researcher & DevOps Engineer Youngju's tech blog. GPU/CUDA, LLM, MLOps, Kubernetes AI workloads, distributed training, and data engineering. ko fjvbn2003@gmail.com (Youngju Kim) fjvbn2003@gmail.com (Youngju Kim) Sat, 16 May 2026 00:00:00 GMT https://www.youngju.dev/blog/culture/2026-05-16-llm-fine-tuning-2026-lora-qlora-dora-galore-unsloth-axolotl-trl-peft-mlx-lm-deep-dive.en https://www.youngju.dev/blog/culture/2026-05-16-llm-fine-tuning-2026-lora-qlora-dora-galore-unsloth-axolotl-trl-peft-mlx-lm-deep-dive.en In 2026, the LLM fine-tuning ecosystem has exploded from the simple adapters of LoRA into a sprawling tree — QLoRA at 4-bit, DoRA with weight decomposition, GaLore with gradient projection. Hugging Face PEFT 0.14 and TRL 0.13 have locked in a standard interface, Unsloth has shaken the single-GPU market with 2x faster training, and Axolotl and LLaMA-Factory have democratized fine-tuning behind YAML. Apple MLX-LM opened the era of fine-tuning 7B models on M-series laptops. This article maps the whole landscape — algorithms, tools, datasets, hardware, cloud, and Korean and Japanese model case studies. Sat, 16 May 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) llm-fine-tuningloraqloradoragaloreunslothaxolotltrlpeftmlxsftdpo2026deep-diveenglish https://www.youngju.dev/blog/culture/2026-05-16-llm-fine-tuning-2026-lora-qlora-dora-galore-unsloth-axolotl-trl-peft-mlx-lm-deep-dive.ja https://www.youngju.dev/blog/culture/2026-05-16-llm-fine-tuning-2026-lora-qlora-dora-galore-unsloth-axolotl-trl-peft-mlx-lm-deep-dive.ja 2026年のLLMファインチューニング生態系は、LoRAというシンプルなアダプターから出発して、QLoRAの4ビット、DoRAの分解、GaLoreの勾配射影まで、五年の間に巨大な枝を広げた。Hugging FaceのPEFT 0.14とTRL 0.13が標準インターフェースを固め、Unslothが2倍の高速学習でシングルGPU市場を揺るがし、AxolotlとLLaMA-FactoryがYAMLでファインチューニングを民主化した。Apple MLX-LMはMシリーズのノートで7Bモデルをファインチューニングする時代を開いた。本稿はその全体地図を描く — アルゴリズム、ツール、データセット、ハードウェア、クラウド、そして韓国・日本のモデル事例まで。 Sat, 16 May 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) llm-fine-tuningloraqloradoragaloreunslothaxolotltrlpeftmlxsftdpo2026deep-dive日本語 https://www.youngju.dev/blog/culture/2026-05-16-llm-fine-tuning-2026-lora-qlora-dora-galore-unsloth-axolotl-trl-peft-mlx-lm-deep-dive https://www.youngju.dev/blog/culture/2026-05-16-llm-fine-tuning-2026-lora-qlora-dora-galore-unsloth-axolotl-trl-peft-mlx-lm-deep-dive 2026년 LLM 파인튜닝 생태계는 LoRA의 단순한 어댑터에서 출발해 QLoRA의 4비트, DoRA의 분해, GaLore의 그라디언트 프로젝션까지 5년 만에 거대한 가지를 펼쳤다. Hugging Face의 PEFT 0.14와 TRL 0.13이 표준 인터페이스를 잡았고, Unsloth가 2배 빠른 학습으로 단일 GPU 시장을 흔들었으며, Axolotl과 LLaMA-Factory가 YAML로 파인튜닝을 민주화했다. Apple MLX-LM은 M 시리즈 노트북에서 7B 모델을 파인튜닝하는 시대를 열었다. 이 글은 그 전체 지도를 그린다 — 알고리즘, 도구, 데이터셋, 하드웨어, 클라우드, 그리고 한국·일본 모델 사례까지. Sat, 16 May 2026 00:00:00 GMT fjvbn2003@gmail.com (Youngju Kim) llm-fine-tuningloraqloradoragaloreunslothaxolotltrlpeftmlxsftdpo2026deep-dive