CO

cosmoem 8x1B

CosmoemGGUF by mradermacherUpdated 2024-11-07
Download on Hugging Face
Parameters10.2B
Context2K
Layers24
Architecturellama
Smallest quant3.9 GB
Quick answer

cosmoem 8x1B needs about 8.4 GB of memory at Q4_K_M (a 6.3 GB file) with an 8K context window, so it fits on a 12 GB GPU. This repository has 13 quants, from 3.9 GB (Q2_K) to 20.4 GB (F16).

cosmoem 8x1B is a 10.2B-parameter Cosmoem model in GGUF format, quantized by mradermacher. Its trained context length is 2K tokens. The repository has 13 quant files, from 3.9 GB (Q2_K) to 20.4 GB (F16).

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q2_K3.9 GB 4.8 GB
Fits Get file →
Q3_K_S4.6 GB 5.5 GB
Fits Get file →
Q3_K_M5.0 GB 5.9 GB
Fits Get file →
Q3_K_L5.4 GB 6.3 GB
Fits Get file →
IQ4_XS5.6 GB 6.5 GB
Fits Get file →
Q4_05.9 GB 6.8 GB
Fits Get file →
Q4_K_S5.9 GB 6.8 GB
Fits Get file →
Q4_K_M6.3 GB 7.2 GB
Fits Get file →
Q5_K_S7.1 GB 8.0 GB
Too big Get file →
Q5_K_M7.3 GB 8.2 GB
Too big Get file →
Q6_K8.4 GB 9.3 GB
Too big Get file →
Q8_010.8 GB 11.7 GB
Too big Get file →
F1620.4 GB 21.3 GB
Too big Get file →

What GPU do you need for cosmoem 8x1B?

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 Q2_K (3.9 GB file, about 6.0 GB in use).
  • 8 GB VRAM: the best quant that fits is Q4_K_S (5.9 GB file, about 8.0 GB in use).
  • 12 GB VRAM: the best quant that fits is Q6_K (8.4 GB file, about 10.5 GB in use).
  • 16 GB VRAM: the best quant that fits is Q8_0 (10.8 GB file, about 12.9 GB in use).
  • 24 GB VRAM: the best quant that fits is F16 (20.4 GB file, about 22.5 GB in use).

Which cosmoem 8x1B quant should you choose?

Q4_K_M is the usual starting point for cosmoem 8x1B: it balances file size and quality well. Here it is a 6.3 GB file.

The Q8_0 file is 10.8 GB, about 1.7 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.

If memory is tight, the smallest file is Q2_K at 3.9 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
Parameters10.2B (approximate, mixture of experts)
Layers24
Trained context2,048 tokens
KV cache (fp16)192 KB per token

Source

Quantized bymradermacher
FamilyCosmoem
Licenseapache-2.0
Downloads (HF)82
Last updated2024-11-07

Frequently asked questions about cosmoem 8x1B

How much VRAM does cosmoem 8x1B need?

About 8.4 GB at Q4_K_M with an 8K context window. That is the 6.3 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 cosmoem 8x1B on an 8 GB GPU?

Yes, with Q4_K_S: the 5.9 GB file needs about 8.0 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of cosmoem 8x1B GGUF?

It depends on the quant: Q2_K is 3.9 GB, Q3_K_S is 4.6 GB, Q3_K_M is 5.0 GB, Q3_K_L is 5.4 GB, IQ4_XS is 5.6 GB, Q4_0 is 5.9 GB and 7 more.

What context length does cosmoem 8x1B support?

The GGUF header lists a trained context length of 2,048 tokens (2K). The KV cache grows with the context you use: about 1.6 GB at 8K and 0.4 GB at 2K.

Where can I download cosmoem 8x1B GGUF?

The files are on Hugging Face in the mradermacher/cosmoem-8x1B-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 cosmoem 8x1B 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 mradermacher

All by mradermacher →

More Cosmoem models

All Cosmoem →

Other medium models

All medium models →