Absortio

Email → Summary → Bookmark → Email

Unsloth AI - Open Source Fine-tuning & RL for LLMs

https://unsloth.ai/ Aug 14, 2025 18:50

Extracto

Open source fine-tuning & reinforcment learning (RL) for gpt-oss, Llama 4, DeepSeek-R1 and Qwen3 LLMs! Beginner friendly.

Resumen

Resumen Principal

Unsloth se presenta como una solución vanguardista para la optimización del entrenamiento y el finetuning de Grandes Modelos de Lenguaje (LLMs), prometiendo una aceleración sin precedentes y una eficiencia mejorada. Su principal propuesta de valor es la capacidad de entrenar modelos personalizados en tan solo 24 horas, en contraste con los 30 días que comúnmente requieren las metodologías tradicionales. Esta extraordinaria velocidad se consigue a través de un enfoque técnico avanzado: la derivación manual de los pasos matemáticos más intensivos computacionalmente y la escritura de kernels de GPU especializados, lo que permite una mejora sustancial en el rendimiento sin necesidad de actualizar el hardware existente. La plataforma destaca por ser **hasta 30 veces más rápida que Flash Attention 2 (FA2

Contenido

Easily finetune & train LLMs.
Get faster with unsloth.

6 cute pastel coloured sloths staring at their computer screens happy

Sign up to our newsletter

We'll share important news & more!

Subscribe now

Train your own custom model in 24 hrs, not 30 days.

30x faster than FA2 + 30% accuracy

90% less memory usage than FA2

TTS, BERT, FFT & more support

How are we faster?

By manually deriving all compute heavy maths steps and handwriting GPU kernels, Unsloth magically makes training faster without any hardware changes.

cute cat happy being surrounded by 3d code, objects and geometric blocks

cute sloth on top of a 3d graphics card

1 GPU or 100 GPUs

10x faster on a single GPU and up to 30x faster on multiple GPU systems compared to Flash Attention 2 (FA2).
We support NVIDIA GPUs from Tesla T4 to H100, and we’re portable to AMD and Intel GPUs.

Don’t believe us?

Why not try our fully free open source version? Finetune 2X faster on a single NVIDIA GPU for free on Google Colab or Kaggle Notebooks.

The details

We're making AI training easier for everyone

Find out more

Unsloth makes everything greener

As hardware costs rise and performance gains plateau, we use our math and coding skills to make AI LLMs and training run smarter and faster.

cute environment 3d models desert environment with different colours of cactus

Want lightning fast inference? We’re working on it!

Contact us

Contact us

2x faster inference - even faster in the works

cute environment 3d models desert environment with different colours of cactus

Don't forget to join our newsletter!

Subscribe

cute sloth reading newspaper

Pricing

Free

Freeware of our standard version of unsloth

Get started
  • Open-source
  • Supports Mistral, Gemma
  • Supports LLama 1, 2, 3
  • MultiGPU - coming soon
  • Supports 4 bit, 16 bit LoRA

unsloth Pro

2.5x faster training + 20% less VRAM

Contact us
  • 2.5x number of GPUs faster than FA2
  • 20% less memory than OSS
  • Enhanced MultiGPU support
  • Up to 8 GPUS support
  • For any usecase

unsloth Enterprise

Unlock 30x faster training + multi-node support + 30% accuracy

Contact us
  • 32x number of GPUs faster than FA2
  • up to +30% accuracy
  • 5x faster inference
  • Supports full training
  • All Pro plan features
  • Multi-node support
  • Customer support

Ready to use unsloth?

Fuente: Unsloth - Open source Fine-tuning & RL for LLMs