mradermacher's picture
auto-patch README.md
b9341cb verified
metadata
base_model: schnapper79/lumikabra-195B_v0.3
language:
  - en
library_name: transformers
license: other
license_link: https://mistral.ai/licenses/MRL-0.1.md
license_name: mistral-ai-research-licence
quantized_by: mradermacher
tags:
  - mergekit
  - lumikabra-195B

About

weighted/imatrix quants of https://huggingface.co/schnapper79/lumikabra-195B_v0.3

static quants are available at https://huggingface.co/mradermacher/lumikabra-195B_v0.3-GGUF

Usage

If you are unsure how to use GGUF files, refer to one of TheBloke's READMEs for more details, including on how to concatenate multi-part files.

Provided Quants

(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)

Link Type Size/GB Notes
GGUF i1-IQ1_S 41.1 for the desperate
GGUF i1-IQ1_M 45.0 mostly desperate
PART 1 PART 2 i1-IQ2_XXS 51.4
PART 1 PART 2 i1-IQ2_XS 57.3
PART 1 PART 2 i1-IQ2_S 60.9
PART 1 PART 2 i1-IQ2_M 66.0
PART 1 PART 2 i1-Q2_K 71.7 IQ3_XXS probably better
PART 1 PART 2 i1-IQ3_XXS 74.6 lower quality
PART 1 PART 2 i1-IQ3_XS 79.6
PART 1 PART 2 i1-Q3_K_S 83.9 IQ3_XS probably better
PART 1 PART 2 i1-IQ3_S 84.1 beats Q3_K*
PART 1 PART 2 i1-IQ3_M 87.7
PART 1 PART 2 i1-Q3_K_M 93.8 IQ3_S probably better
PART 1 PART 2 PART 3 i1-Q3_K_L 102.5 IQ3_M probably better
PART 1 PART 2 PART 3 i1-IQ4_XS 103.9
PART 1 PART 2 PART 3 i1-Q4_0 110.0 fast, low quality
PART 1 PART 2 PART 3 i1-Q4_K_S 110.4 optimal size/speed/quality
PART 1 PART 2 PART 3 i1-Q4_K_M 116.3 fast, recommended
PART 1 PART 2 PART 3 i1-Q5_K_S 133.9
PART 1 PART 2 PART 3 i1-Q5_K_M 137.3
PART 1 PART 2 PART 3 PART 4 i1-Q6_K 159.7 practically like static Q6_K

Here is a handy graph by ikawrakow comparing some lower-quality quant types (lower is better):

image.png

And here are Artefact2's thoughts on the matter: https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9

FAQ / Model Request

See https://huggingface.co/mradermacher/model_requests for some answers to questions you might have and/or if you want some other model quantized.

Thanks

I thank my company, nethype GmbH, for letting me use its servers and providing upgrades to my workstation to enable this work in my free time. Additional thanks to @nicoboss for giving me access to his private supercomputer, enabling me to provide many more imatrix quants, at much higher quality, than I would otherwise be able to.