Tiny GGUF models
62726 GGUF models with 3B parameters or fewer. Sizes run from 1M to 3.5B parameters. The longest trained context is 795898K tokens.
What counts as a tiny model?
On this site tiny means 3B parameters or fewer. A Q4_K_M file is usually 2 GB or less, so these models run on almost any GPU, on a phone-class device or on a CPU. 62,726 models fit this class. The most downloaded are YuE2 audio tokenizer, the v9 head of Mothersuperior, on MERT-v2-FullSong, Qwen3.8-27B iMatrix NVFP4 MTP and Tiel Coder 35B A3B GGUF MTP.
Their Q4_K_M files run 0.0 to 444.7 GB. The families you will see most are Qwen (12810), Llama (7735) and Gemma (4035). The size class sorts models by parameter count, so a model with more experts or a longer context can need more memory than its class suggests. The model pages show the real file sizes.
Frequently asked questions
How much VRAM does a tiny GGUF model need?
A Q4_K_M file is usually 2 GB or less, so these models run on almost any GPU, on a phone-class device or on a CPU.
Which tiny GGUF model is the most popular?
By Hugging Face downloads, YuE2 audio tokenizer, the v9 head of Mothersuperior, on MERT-v2-FullSong.