MI

saiga mistral

MistralGGUF by maridzeUpdated 2024-03-19
Download on Hugging Face
Parameters7.24B
Context32K
Layers32
Architecturellama
Smallest quant4.1 GB
Quick answer

saiga mistral needs about 9.3 GB of memory at Q8_0 (a 7.7 GB file) with an 8K context window, so it fits on a 12 GB GPU. This repository has 3 quants, from 4.1 GB (Q4_0) to 14.5 GB (F16). Longer context needs more: at 32K tokens the KV cache alone adds about 4.3 GB.

saiga mistral is a 7.24B-parameter Mistral model in GGUF format, quantized by maridze. Its trained context length is 32K tokens. The repository has 3 quant files, from 4.1 GB (Q4_0) to 14.5 GB (F16).

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q4_04.1 GB 5.7 GB
Fits Get file →
Q8_07.7 GB 9.3 GB
Too big Get file →
F1614.5 GB 16.1 GB
Too big Get file →

What GPU do you need for saiga mistral?

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: no quant fits at an 8K context.
  • 6 GB VRAM: the best quant that fits is Q4_0 (4.1 GB file, about 5.7 GB in use).
  • 8 GB VRAM: the best quant that fits is Q4_0 (4.1 GB file, about 5.7 GB in use).
  • 12 GB VRAM: the best quant that fits is Q8_0 (7.7 GB file, about 9.3 GB in use).
  • 16 GB VRAM: the best quant that fits is Q8_0 (7.7 GB file, about 9.3 GB in use).
  • 24 GB VRAM: the best quant that fits is F16 (14.5 GB file, about 16.1 GB in use).

Which saiga mistral quant should you choose?

Q8_0 (7.7 GB) is the middle option in this repository. Close to lossless. It costs roughly twice the memory of a 4-bit file, so it suits cases where you have plenty of VRAM.

If memory is tight, the smallest file is Q4_0 at 4.1 GB, with the quality trade-offs that come with fewer bits. The usual sweet spot between file size and quality, and the quant most people start with.

New to quant names? Read GGUF quantization explained.

Specifications

Architecture from the GGUF header

Architecturellama
Parameters7.24B
Layers32
Trained context32,768 tokens
KV cache (fp16)128 KB per token

Source

Quantized bymaridze
FamilyMistral
Downloads (HF)18
Last updated2024-03-19

Frequently asked questions about saiga mistral

How much VRAM does saiga mistral need?

About 9.3 GB at Q8_0 with an 8K context window. That is the 7.7 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 saiga mistral on an 8 GB GPU?

Yes, with Q4_0: the 4.1 GB file needs about 5.7 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of saiga mistral GGUF?

It depends on the quant: Q4_0 is 4.1 GB, Q8_0 is 7.7 GB, F16 is 14.5 GB.

What context length does saiga mistral support?

The GGUF header lists a trained context length of 32,768 tokens (32K). The KV cache grows with the context you use: about 1.1 GB at 8K and 4.3 GB at 32K.

Where can I download saiga mistral GGUF?

The files are on Hugging Face in the maridze/saiga_mistral repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.

Similar models

Browse all →

More from maridze

All by maridze →

More Mistral models

All Mistral →

Other small models

All small models →