TI

TinyMixtral 32x248M

TinymixtralGGUF by ukungUpdated 2024-05-20
Download on Hugging Face
Parameters4.93B
Context32K
Layers12
Architecturellama
Smallest quant1.8 GB
Quick answer

TinyMixtral 32x248M needs about 3.6 GB of memory at Q4_K_M (a 3.0 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 9 quants, from 1.8 GB (Q2_K) to 3.4 GB (Q5_0). Longer context needs more: at 32K tokens the KV cache alone adds about 0.4 GB.

TinyMixtral 32x248M is a 4.93B-parameter Tinymixtral model in GGUF format, quantized by ukung. Its trained context length is 32K tokens. The repository has 9 quant files, from 1.8 GB (Q2_K) to 3.4 GB (Q5_0).

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q2_K1.8 GB 2.4 GB
Fits Get file →
Q3_K_S2.1 GB 2.7 GB
Fits Get file →
Q3_K_M2.3 GB 2.9 GB
Fits Get file →
Q3_K_L2.6 GB 3.2 GB
Fits Get file →
Q4_02.8 GB 3.4 GB
Fits Get file →
Q4_K_S2.8 GB 3.4 GB
Fits Get file →
Q4_K_M3.0 GB 3.6 GB
Fits Get file →
Q4_13.1 GB 3.7 GB
Fits Get file →
Q5_03.4 GB 4.0 GB
Fits Get file →

What GPU do you need for TinyMixtral 32x248M?

These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.

  • 4 GB VRAM: the best quant that fits is Q5_0 (3.4 GB file, about 4.0 GB in use).

Which TinyMixtral 32x248M quant should you choose?

Q4_K_M is the usual starting point for TinyMixtral 32x248M: it balances file size and quality well. Here it is a 3.0 GB file.

If memory is tight, the smallest file is Q2_K at 1.8 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.

New to quant names? Read GGUF quantization explained.

Specifications

Architecture from the GGUF header

Architecturellama
Parameters4.93B (approximate, mixture of experts)
Layers12
Trained context32,768 tokens
KV cache (fp16)12 KB per token

Source

Quantized byukung
Licenseapache-2.0
Downloads (HF)69
Last updated2024-05-20

Frequently asked questions about TinyMixtral 32x248M

How much VRAM does TinyMixtral 32x248M need?

About 3.6 GB at Q4_K_M with an 8K context window. That is the 3.0 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.

Can I run TinyMixtral 32x248M on an 8 GB GPU?

Yes, with Q5_0: the 3.4 GB file needs about 4.0 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of TinyMixtral 32x248M GGUF?

It depends on the quant: Q2_K is 1.8 GB, Q3_K_S is 2.1 GB, Q3_K_M is 2.3 GB, Q3_K_L is 2.6 GB, Q4_0 is 2.8 GB, Q4_K_S is 2.8 GB and 3 more.

What context length does TinyMixtral 32x248M support?

The GGUF header lists a trained context length of 32,768 tokens (32K). The KV cache grows with the context you use: about 0.1 GB at 8K and 0.4 GB at 32K.

Where can I download TinyMixtral 32x248M GGUF?

The files are on Hugging Face in the ukung/TinyMixtral-32x248M-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.

What license does TinyMixtral 32x248M use?

Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.

Similar models

Browse all →

More from ukung

All by ukung →

More Tinymixtral models

All Tinymixtral →

Other small models

All small models →