Qwen2.5-72B-i1-GGUF / README.md
mradermacher's picture
auto-patch README.md
a199b7f verified
metadata
base_model: Qwen/Qwen2.5-72B
language:
  - en
library_name: transformers
license: other
license_link: https://huggingface.co/Qwen/Qwen2.5-72B/blob/main/LICENSE
license_name: qwen
quantized_by: mradermacher

About

weighted/imatrix quants of https://huggingface.co/Qwen/Qwen2.5-72B

static quants are available at https://huggingface.co/mradermacher/Qwen2.5-72B-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 22.8 for the desperate
GGUF i1-IQ1_M 23.8 mostly desperate
GGUF i1-IQ2_XXS 25.6
GGUF i1-IQ2_XS 27.2
GGUF i1-IQ2_S 28.0
GGUF i1-IQ2_M 29.4
GGUF i1-Q2_K_S 29.7 very low quality
GGUF i1-Q2_K 29.9 IQ3_XXS probably better
GGUF i1-IQ3_XXS 31.9 lower quality
GGUF i1-IQ3_XS 32.9
GGUF i1-IQ3_S 34.6 beats Q3_K*
GGUF i1-Q3_K_S 34.6 IQ3_XS probably better
GGUF i1-IQ3_M 35.6
GGUF i1-Q3_K_M 37.8 IQ3_S probably better
GGUF i1-Q3_K_L 39.6 IQ3_M probably better
GGUF i1-IQ4_XS 39.8
GGUF i1-Q4_0_4_4 41.3 fast on arm, low quality
GGUF i1-Q4_0_4_8 41.3 fast on arm+i8mm, low quality
GGUF i1-Q4_0_8_8 41.3 fast on arm+sve, low quality
GGUF i1-IQ4_NL 41.4 prefer IQ4_XS
GGUF i1-Q4_0 41.5 fast, low quality
GGUF i1-Q4_K_S 44.0 optimal size/speed/quality
GGUF i1-Q4_1 45.8
GGUF i1-Q4_K_M 47.5 fast, recommended
PART 1 PART 2 i1-Q5_0 50.4
PART 1 PART 2 i1-Q5_K_S 51.5
PART 1 PART 2 i1-Q5_K_M 54.5
PART 1 PART 2 i1-Q5_1 54.7
PART 1 PART 2 i1-Q6_K 64.4 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. 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.