# Unsloth > Wiki for Unsloth — fast, memory-efficient LLM fine-tuning: LoRA/QLoRA, datasets & chat templates, hyperparameters, reinforcement learning (GRPO), saving/exporting (GGUF/Ollama), inference & deployment (vLLM/LM Studio), Unsloth Studio, and the model catalog. > Covers: Unsloth LLM fine-tuning: when to fine-tune (vs RAG), installation/requirements, datasets & chat templates, LoRA/QLoRA & hyperparameters, reinforcement learning (GRPO/GSPO, reward design), saving/exporting (GGUF incl. Dynamic 2.0, Ollama, merged weights), inference & deployment (in-framework, vLLM, LM Studio, llama.cpp), Unsloth Studio, the FastLanguageModel API, and the model/notebook catalog. > Not covered: Per-model VRAM/quant/template specifics (catalogued — read the per-model guide), exact current API signatures, and frequent post-date releases — verify against docs.unsloth.ai. > Current as of: 2026-08-19 - [LLM Wiki](/raw/unsloth/README.md) - [Knowledge Base Index](/raw/unsloth/wiki/index.md) - [Datasets & Chat Templates](/raw/unsloth/wiki/concepts/datasets.md) - [Fine-tuning Basics: Is It Right for You?](/raw/unsloth/wiki/concepts/fine-tuning-basics.md) - [Inference & Deployment](/raw/unsloth/wiki/concepts/inference-and-deployment.md) - [Installation & Requirements](/raw/unsloth/wiki/concepts/installation.md) - [LoRA, QLoRA & Hyperparameters](/raw/unsloth/wiki/concepts/lora-and-hyperparameters.md) - [Reinforcement Learning (GRPO)](/raw/unsloth/wiki/concepts/reinforcement-learning.md) - [Saving & Exporting Models](/raw/unsloth/wiki/concepts/saving-and-exporting.md) - [Unsloth Studio](/raw/unsloth/wiki/concepts/unsloth-studio.md) - [What is Unsloth](/raw/unsloth/wiki/concepts/what-is-unsloth.md) - [The unsloth Library (FastLanguageModel)](/raw/unsloth/wiki/entities/unsloth-library.md) - [Activity Log](/raw/unsloth/wiki/log.md) - [Docs Catalog](/raw/unsloth/wiki/summaries/docs-catalog.md) - [Model Catalog & Notebooks](/raw/unsloth/wiki/summaries/model-catalog-and-notebooks.md) - [End-to-End: Fine-tune to Deployment](/raw/unsloth/wiki/syntheses/end-to-end-fine-tune.md) ## XL edition (Pro) > Deeper per-item reference pages. Requires a subscription (/pro). Over MCP, > set AGENTWIKIS_API_KEY (agentwikis-mcp 0.2.0+) and these pages join search > and open_wiki automatically. Over HTTP, send `Authorization: Bearer ` > on the /raw/ URLs below. - [XL Edition Index](/raw/unsloth/wiki-xl/index.md) — Pro - [Dataset Formats, Chat Templates & Data Recipes](/raw/unsloth/wiki-xl/guides/datasets-and-formats.md) — Pro - [Local LLMs with Coding Agents (OpenAI Codex & Claude Code)](/raw/unsloth/wiki-xl/guides/local-llms-with-coding-agents.md) — Pro - [MCP Servers with Local LLMs (Studio & llama.cpp)](/raw/unsloth/wiki-xl/guides/mcp-servers-with-local-llms.md) — Pro - [Unsloth Notebooks Catalog (Colab & Kaggle)](/raw/unsloth/wiki-xl/guides/notebooks-catalog.md) — Pro - [GRPO in Practice: VRAM Rules, GRPOConfig & Reward Functions](/raw/unsloth/wiki-xl/guides/reinforcement-learning.md) — Pro - [Unsloth as an OpenAI/Anthropic-Compatible API Endpoint](/raw/unsloth/wiki-xl/guides/unsloth-as-an-api-endpoint.md) — Pro - [Unsloth Release Digest: March–April 2026](/raw/unsloth/wiki-xl/history/releases-2026-03-to-04.md) — Pro - [Unsloth Release Digest: May–June 2026](/raw/unsloth/wiki-xl/history/releases-2026-05-to-06.md) — Pro - [Unsloth Model Catalog — Exact Repo IDs & Quant Formats](/raw/unsloth/wiki-xl/reference/model-catalog.md) — Pro - [Running DeepSeek V3-0324, V3.1 & R1-0528 Locally](/raw/unsloth/wiki-xl/reference/running-deepseek-models.md) — Pro - [Running Gemma 4, Gemma 4 QAT, Gemma 3 & Gemma 3n](/raw/unsloth/wiki-xl/reference/running-gemma.md) — Pro - [Running GLM-5.2, GLM-5 & GLM-4.7-Flash Locally](/raw/unsloth/wiki-xl/reference/running-glm-models.md) — Pro - [Running Kimi K2 & gpt-oss](/raw/unsloth/wiki-xl/reference/running-kimi-and-gpt-oss.md) — Pro - [Running Llama 4, Grok 2 & Cogito v2.1 (large MoE)](/raw/unsloth/wiki-xl/reference/running-large-moe-models.md) — Pro - [Running Magistral, Mistral Medium 3.5 & Devstral](/raw/unsloth/wiki-xl/reference/running-magistral-and-mistral.md) — Pro - [Running NVIDIA Nemotron 3 & MiniMax M2/M3](/raw/unsloth/wiki-xl/reference/running-nemotron-and-minimax.md) — Pro - [Running Granite 4, Phi-4 Reasoning, QwQ-32B, Liquid LFM2.5 & Ministral 3](/raw/unsloth/wiki-xl/reference/running-other-models.md) — Pro - [Running Qwen3.6, Qwen3.5 & Qwen3-Coder-Next Locally](/raw/unsloth/wiki-xl/reference/running-qwen3-models.md) — Pro