MI

bigstral-12b-32k-8xMoE

MistralGGUF by bartowskiUpdated 2024-03-09
Download on Hugging Face
Parameters81.5B
Context32K
Layers56
Architecturellama
Smallest quant30.2 GB
Quick answer

bigstral-12b-32k-8xMoE needs about 52.0 GB of memory at Q4_K_M (a 49.6 GB file) with an 8K context window, which is more than a 48 GB GPU holds. This repository has 7 quants, from 30.2 GB (Q2_K) to 49.6 GB (Q4_K_M). Longer context needs more: at 32K tokens the KV cache alone adds about 7.5 GB.

bigstral-12b-32k-8xMoE is a 81.5B-parameter Mistral model in GGUF format, quantized by bartowski. Its trained context length is 32K tokens. The repository has 7 quant files, from 30.2 GB (Q2_K) to 49.6 GB (Q4_K_M).

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q2_K30.2 GB 32.6 GB
Too big Get file →
Q3_K_S35.6 GB 38.0 GB
Too big Get file →
Q3_K_M39.3 GB 41.7 GB
Too big Get file →
Q3_K_L42.2 GB 44.5 GB
Too big Get file →
Q4_046.1 GB 48.5 GB
Too big Get file →
Q4_K_S46.7 GB 49.0 GB
Too big Get file →
Q4_K_M49.6 GB 52.0 GB
Too big Get file →

What GPU do you need for bigstral-12b-32k-8xMoE?

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: no quant fits at an 8K context.
  • 8 GB VRAM: no quant fits at an 8K context.
  • 12 GB VRAM: no quant fits at an 8K context.
  • 16 GB VRAM: no quant fits at an 8K context.
  • 24 GB VRAM: no quant fits at an 8K context.
  • 32 GB VRAM: no quant fits at an 8K context.
  • 48 GB VRAM: the best quant that fits is Q3_K_L (42.2 GB file, about 44.5 GB in use).

Which bigstral-12b-32k-8xMoE quant should you choose?

Q4_K_M is the usual starting point for bigstral-12b-32k-8xMoE: it balances file size and quality well. Here it is a 49.6 GB file.

If memory is tight, the smallest file is Q2_K at 30.2 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
Parameters81.5B (approximate, mixture of experts)
Layers56
Trained context32,768 tokens
KV cache (fp16)224 KB per token

Source

Quantized bybartowski
FamilyMistral
Downloads (HF)416
Last updated2024-03-09

Frequently asked questions about bigstral-12b-32k-8xMoE

How much VRAM does bigstral-12b-32k-8xMoE need?

About 52.0 GB at Q4_K_M with an 8K context window. That is the 49.6 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 bigstral-12b-32k-8xMoE on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 32.6 GB, so you would need a larger GPU or CPU offloading.

What is the file size of bigstral-12b-32k-8xMoE GGUF?

It depends on the quant: Q2_K is 30.2 GB, Q3_K_S is 35.6 GB, Q3_K_M is 39.3 GB, Q3_K_L is 42.2 GB, Q4_0 is 46.1 GB, Q4_K_S is 46.7 GB and 1 more.

What context length does bigstral-12b-32k-8xMoE support?

The GGUF header lists a trained context length of 32,768 tokens (32K). The KV cache grows with the context you use: about 1.9 GB at 8K and 7.5 GB at 32K.

Where can I download bigstral-12b-32k-8xMoE GGUF?

The files are on Hugging Face in the bartowski/bigstral-12b-32k-8xMoE-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.

Other GGUFs of Mistral-7B-Instruct-v0.2

Similar models

Browse all →

More from bartowski

All by bartowski →

More Mistral models

All Mistral →

Other large models

All large models →