exported_from: sophosympatheia/Midnight-Rose-103B-v2.0.3
language:
- en
library_name: transformers
license: llama2
quantized_by: mradermacher
About
Weighted imatrix quantize of https://huggingface.co/sophosympatheia/Midnight-Rose-103B-v2.0.3
The weight was calculated using an experimental (potentially crappy) method that is iterative (thus the "i1"), using 270k semi-random english tokens.
static quants are available at https://huggingface.co/mradermacher/Midnight-Rose-103B-v2.0.3-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 | 22.1 | for the desperate |
| GGUF | i1-IQ2_XXS | 27.7 | |
| GGUF | i1-IQ2_XS | 30.8 | |
| GGUF | i1-IQ2_M | 35.1 | |
| GGUF | i1-Q2_K | 38.3 | IQ3_XXS probably better |
| GGUF | i1-IQ3_XXS | 40.6 | lower quality |
| GGUF | i1-Q3_K_XS | 42.4 | |
| GGUF | i1-IQ3_XS | 42.6 | |
| GGUF | i1-Q3_K_S | 44.9 | IQ3_XS probably better |
| PART 1 PART 2 | i1-Q3_K_M | 50.0 | IQ3_S probably better |
| PART 1 PART 2 | i1-Q3_K_L | 54.5 | IQ3_M probably better |
| PART 1 PART 2 | i1-IQ4_XS | 55.5 | |
| PART 1 PART 2 | i1-Q4_0 | 58.8 | |
| PART 1 PART 2 | i1-Q4_K_S | 59.0 | optimal size/speed/quality |
| PART 1 PART 2 | i1-Q4_K_M | 62.3 | fast, recommended |
| PART 1 PART 2 | i1-Q5_K_S | 71.4 | |
| PART 1 PART 2 | i1-Q5_K_M | 73.3 | |
| PART 1 PART 2 | i1-Q6_K | 85.1 | practically like static Q6_K |
Here is a handy graph by ikawrakow comparing some lower-quality quant types (lower is better):
And here are Artefact2's thoughts on the matter: https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9
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.