← Discover
Discover · GitHub · Airllm↑ +208 today

Run giant AI models with 4GB VRAM

AirLLM allows large language models with 70 billion parameters to be run on graphics processing units with only 4 GB of video memory (VRAM). This library uses memory optimization techniques to enable the use of high-capacity models with low hardware requirements.

Updates

  • September 6, 2026: Stars 33,307 → 33,755, latest release v4.0.0 (September 5, 2026).
  • August 31, 2026: Stars 31,598 → 33,307, latest release v3.3.0 (August 28, 2026).
  • August 19, 2026: Stars 30,796 → 31,598, latest release v3.2.0 (August 18, 2026).
  • August 12, 2026: Stars 29,265 → 30,796, latest release v3.1.0 (July 29, 2026).

What you get

  • Possibility to run models with 70B parameters with 4GB VRAM.
  • Ability to use 405B Llama3.1 models with 8GB VRAM.
  • Up to 3x speed increase with block-based compression.

Installation

Using pip (PyPI)
pip install airllm

If you don't write code

🤖 Paste this into your AI agent (Claude Code · Codex · Antigravity)

I want to run a model with 70B parameters using the AirLLM library on my graphics card with low VRAM capacity. I used the pip install airllm command for installation. How can I create the Python code structure needed to load my model and output it with a simple text input, using the AutoModel class? I know I need to make sure I have enough disk space during the process, can you explain the basic steps I need to follow to start the process?

Related dictionary terms

Who it is forFor users with limited hardware resources who want to run high-capacity large language models locally.
LicenseApache-2.0

Links

TreScout did not build this tool · we found it in GitHub trends and wrote it up. This page describes the repository as of 2026-06-04: The star count and our text belong to that day, the repository may have changed since. Check the repository link for the current state. This page was machine-translated from the Turkish original · the Turkish version prevails.