Skip to content
localllm
日本語

File sizes and memory needs of local LLMs (GGUF)

Check whether a model will run on your PC before you download it.

For local LLMs (quantized models in GGUF format), we list the actual file size of each quantization and the estimated memory for each context length. Memory figures are estimates from a published formula and are always shown separately from measured values.

Models
1,276
GGUF sources
1,659
GPUs and Macs
91

Fetched from Hugging Face: (updated daily)

Find by VRAM or unified memory

Pick the amount of GPU VRAM (unified memory on a Mac) to see the models estimated to fit.

Find by device (GeForce, Radeon, Mac and more)

Sorted by Hugging Face downloads (last 30 days). This is not a rating or ranking.

See the download ranking → All models →

New models

See all new models →
Japanese models Models developed by Japanese companies and research institutes, and derivatives tuned for Japanese. See Japanese models → Licenses and commercial use Whether you can use a model for work depends on its license. We group models by the original model's license into "commercial use allowed", "with conditions" and "not allowed", with a summary of the terms. See models you can use commercially → Is a local LLM safe? Does what you type leave your PC? Can a GGUF file contain a program? How do you tell sources apart? We explain how it works. Read how to check safety → Reading and choosing quantizations How to read Q4_K_M, Q8_0 and other names, with sizes aggregated from real files on this site and a quality guide. Read the quantization guide →

How this site works

Measured

Measured values stay measured

Quantized file sizes are the actual values on Hugging Face. Split files are added together. We never mix them with estimates.

Estimated

Estimates are marked as estimates

Memory is estimated with a formula that includes the KV cache. We publish the formula and assumptions and always mark estimates with a badge.

✕ No scores

No performance scores of our own

We do not call any model "the best". We show measured and estimated values, with sources, so you can decide.