mradermacher's picture
auto-patch README.md
bf6fb4f verified
metadata
base_model: davidkim205/Rhea-72b-v0.5
language:
  - en
library_name: transformers
license: apache-2.0
quantized_by: mradermacher

About

weighted/imatrix quants of https://huggingface.co/davidkim205/Rhea-72b-v0.5

the imatrix was calculated on a reduced 40k token set (the "quarter" set) as the full token set caused overflows in the model (likely a model bug)

static quants are available at https://huggingface.co/mradermacher/Rhea-72b-v0.5-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 20.5 for the desperate
GGUF i1-IQ1_M 21.8 mostly desperate
GGUF i1-IQ2_XXS 24.0
GGUF i1-IQ2_XS 26.0
GGUF i1-IQ2_S 27.6
GGUF i1-IQ2_M 29.4
GGUF i1-Q2_K 31.1 IQ3_XXS probably better
GGUF i1-IQ3_XXS 31.9 lower quality
GGUF i1-IQ3_XS 34.0
GGUF i1-IQ3_S 35.6 beats Q3_K*
GGUF i1-Q3_K_S 35.6 IQ3_XS probably better
GGUF i1-IQ3_M 37.3
GGUF i1-Q3_K_M 39.3 IQ3_S probably better
GGUF i1-Q3_K_L 42.6 IQ3_M probably better
GGUF i1-IQ4_XS 42.8
GGUF i1-IQ4_NL 45.1 prefer IQ4_XS
GGUF i1-Q4_0 45.2 fast, low quality
GGUF i1-Q4_K_S 45.3 optimal size/speed/quality
GGUF i1-Q4_K_M 47.8 fast, recommended
PART 1 PART 2 i1-Q5_K_S 53.9
PART 1 PART 2 i1-Q5_K_M 55.4
PART 1 PART 2 i1-Q6_K 63.4 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.