unsloth-fine-tuning
Unsloth accelerates LLM fine-tuning on consumer and datacenter GPUs through optimized LoRA and QLoRA training, cutting both speed and memory requirements dramatically. It handles supervised fine-tuning, reinforcement learning with GRPO, vision model adaptation, and TTS training across 300+ model architectures, with direct export to GGUF for deployment on Ollama and llama.cpp.
Unsloth fine-tuning speeds up LLM training 2-5x while cutting VRAM use by 50-80% via LoRA and QLoRA.
AI-generated summary based on this skill's SKILL.md
Decision gist · record as of 2026-07-27
Unsloth fine-tuning speeds up LLM training 2-5x while cutting VRAM use by 50-80% via LoRA and QLoRA. Unsloth accelerates LLM fine-tuning on consumer and datacenter GPUs through optimized LoRA and QLoRA training, cutting both speed and memory requirements dramatically. It handles supervised fine-tuning, reinforcement learning with GRPO, vision model adaptation, and TTS training across 300+ model architectures, with direct export to GGUF for deployment on Ollama and llama.cpp.
Use it when
- Unsloth-fine-tuning supports reinforcement learning with GRPO and other RL algorithms for training reasoning models.
- Yes, unsloth-fine-tuning is designed to fine-tune vision language models and text-to-speech models on single GPU setups.
Verify before relying
Read SKILL.md below before installing (44 files). Open directory: indexed for reading, not audited.
Install
synthetic-sciences/openscience/unsloth · repository language: TypeScript
Open directory. Skills are indexed for reading, not audited. Review a skill's body before installing it.
Frequently asked questions
AI-generated answers based on this skill's SKILL.md and metadata
How does unsloth-fine-tuning achieve fast LLM fine-tuning with less VRAM?
Unsloth-fine-tuning accelerates LLM fine-tuning through optimized LoRA and QLoRA implementations that reduce both speed and memory requirements dramatically. By using low-rank adapters and quantization techniques, it enables training on consumer and datacenter GPUs with significantly lower VRAM footprint than standard approaches, making large model adaptation accessible on modest hardware.
What training methods does unsloth-fine-tuning support beyond standard supervised fine-tuning?
Unsloth-fine-tuning supports reinforcement learning with GRPO and other RL algorithms for training reasoning models, supervised fine-tuning with chat templates and instruction tuning, vision and TTS model adaptation on single GPUs, and hyperparameter optimization for memory and training efficiency across 300+ model architectures including Llama, Qwen, and Gemma.
Can unsloth-fine-tuning train vision and TTS models on a single GPU?
Yes, unsloth-fine-tuning is designed to fine-tune vision language models and text-to-speech models on single GPU setups. Its memory-efficient architecture and gradient checkpointing support enable training of these multimodal and specialized models without requiring expensive multi-GPU infrastructure.
How do I export and deploy models trained with unsloth-fine-tuning?
Unsloth-fine-tuning exports trained models directly to GGUF format for deployment on Ollama, vLLM, and llama.cpp. You can merge LoRA adapters into base models and export the combined weights, enabling fast inference and easy deployment across popular inference frameworks.
What is the typical speedup and memory savings from unsloth-fine-tuning?
Unsloth-fine-tuning delivers 2–5x faster model training compared to standard approaches while cutting VRAM requirements substantially through optimized LoRA/QLoRA implementations and gradient checkpointing. Exact savings depend on model size, batch configuration, and hardware, but the framework is engineered to maximize efficiency on both consumer and datacenter GPUs.
Which model families and sizes does unsloth-fine-tuning work with?
Unsloth-fine-tuning supports 300+ model architectures including Llama, Qwen, Gemma, and other popular LLMs, vision language models, and TTS systems. It handles models from 7B to larger scales with 4-bit quantization, LoRA, and QLoRA training modes, making it adaptable to a wide range of model sizes and types.
SKILL.md
Rendered from the published skill. Quoted content, verbatim.
Unsloth - Fast LLM Fine-Tuning
Fine-tune LLMs 2-5x faster with 50-80% less VRAM. Supports SFT, RL (GRPO), vision, TTS, and 300+ models with zero accuracy loss.
When to Use Unsloth
Use Unsloth when: - Fine-tuning on a single GPU with LoRA/QLoRA (consumer or datacenter) - Training reasoning models with GRPO, Dr. GRPO, DAPO, BNPO, or GSPO - Fine-tuning vision models (Qwen3-VL, Gemma 3, Llama 3.2 Vision) - Fine-tuning TTS models (Orpheus, Sesame-CSM, Whisper) - Exporting to GGUF for Ollama, llama.cpp, or LM Studio - Need padding-free training and
(truncated - see the full file via the links below)
File tree — 15 files
backend/cli/skills/ml-training/unsloth/SKILL.md
backend/cli/skills/ml-training/unsloth/docs/advanced-rl.md
backend/cli/skills/ml-training/unsloth/docs/chat-templates.md
backend/cli/skills/ml-training/unsloth/docs/datasets.md
backend/cli/skills/ml-training/unsloth/docs/docker-extended.md
backend/cli/skills/ml-training/unsloth/docs/dynamic-ggufs-2.0.md
backend/cli/skills/ml-training/unsloth/docs/dynamic-ggufs-aider.md
backend/cli/skills/ml-training/unsloth/docs/faq.md
backend/cli/skills/ml-training/unsloth/docs/fp16-vs-bf16.md
backend/cli/skills/ml-training/unsloth/docs/fp8-rl.md
backend/cli/skills/ml-training/unsloth/docs/glm-4.7-flash.md
backend/cli/skills/ml-training/unsloth/docs/inference-deployment-overview.md
backend/cli/skills/ml-training/unsloth/docs/inference.md
backend/cli/skills/ml-training/unsloth/docs/installation-docker.md
backend/cli/skills/ml-training/unsloth/docs/installation-pip.md
Let your AI agent find skills like this
Example. Real query, live index.
You found this page by searching. An agent finds it by wishing: SkillFed indexes 56,283 agent skills by what they can do, searchable in plain language.
wish › “Fine-tune large language models faster with reduced VRAM using LoRA/QLoRA”
Give your agent the search over MCP, or paste the wish link into any chat. No install? Search from any chat →
Related skills
Unsloth-buddy guides you through optimized language model training using the Unsloth library, delivering 2x speedup and up to 80% VRAM reduction compared to standard approaches. It covers SFT, DPO, GRPO, vision, TTS, and embedding fine-tuning across Apple Silicon, NVIDIA, and Windows environments. The skill orchestrates a complete 7-phase lifecycle from project setup through deployment, managing templates, dashboards, and hardware detection automatically.
Run Unsloth-powered LLM training directly on Google Colab GPUs from openscience, connecting via WebSocket bridge for remote execution. Supports supervised fine-tuning, reinforcement learning, preference optimization, vision, and text-to-speech workflows across free T4 through paid A100 tiers.
Unsloth Training accelerates LLM fine-tuning through GRPO (reinforcement learning with reward functions) and SFT (supervised learning with input-output pairs). It cuts VRAM usage by 60% with FP8 training, speeds up mixed-length datasets via automatic packing, and supports vision model tuning, mobile deployment via ExecuTorch, and export to GGUF, Ollama, and vLLM.
This skill teaches efficient LLM fine-tuning via Unsloth, covering GRPO (reinforcement learning with reward functions) and SFT (supervised training). Learn FP8 training for 60% VRAM reduction, automatic packing for 2–5x speedup, vision model tuning, mobile deployment via QAT, Docker containerization, and export to GGUF and other formats.
Configure end-to-end LLM fine-tuning from single-GPU LoRA to multi-node distributed training. This skill covers QLoRA quantization for large models on consumer hardware, Axolotl framework setup, DeepSpeed ZeRO optimization, DPO alignment, and adapter merging for production deployment.
Train and adapt Gemma models on consumer hardware through supervised fine-tuning, direct preference optimization, and reward modeling workflows. The skill guides you through memory-efficient setups with Unsloth and TRL, dataset formatting, multimodal training for vision and audio, and deployment via GGUF or LiteRT.
More skills llamafactory (Apache-2.0) · fine-tuning-with-trl (MIT) · fine-tuning-with-trl (MIT) · fine-tuning-with-trl (MIT) · llama-cpp (MIT) · Llama Cpp (Apache-2.0) · huggingface-local-models (Apache-2.0)