The difference with normal quantizations is that I quantize the output and embed tensors to f16. and the other tensors to 15_k,q6_k or q8_0. This creates models that are little or not degraded at all and have a smaller size. They run at about 3-6 t/sec on CPU only using llama.cpp And obviously faster on computers with potent GPUs

the fast cat at ZeroWw/llama3-8B-DarkIdol-2.3-Uncensored-32K-GGUF

Model Description:

The module combination has been readjusted to better fulfill various roles and has been adapted for mobile phones.

Saving money(LLama 3)
only test en.
Input Models input text only. Output Models generate text and code only.
Uncensored
Quick response
The underlying model used is winglian/Llama-3-8b-64k-PoSE (The theoretical support is 64k, but I have only tested up to 32k. :)
A scholarly response akin to a thesis.(I tend to write songs extensively, to the point where one song almost becomes as detailed as a thesis. :)
DarkIdol:Roles that you can imagine and those that you cannot imagine.
Roleplay
Specialized in various role-playing scenarios
more look at test role. (https://huggingface.co/aifeifei798/llama3-8B-DarkIdol-1.2/tree/main/test)
more look at LM Studio presets (https://huggingface.co/aifeifei798/llama3-8B-DarkIdol-1.2/tree/main/config-presets)
image/png

virtual idol Twitter

https://x.com/aifeifei799

Questions

The model's response results are for reference only, please do not fully trust them.

Stop Strings

python

    stop = [
      "## Instruction:",
      "### Instruction:",
      "<|end_of_text|>",
      "  //:",
      "</s>",
      "<3```",
      "### Note:",
      "### Input:",
      "### Response:",
      "### Emoticons:"
    ],

Model Use

Koboldcpp https://github.com/LostRuins/koboldcpp
Since KoboldCpp is taking a while to update with the latest llama.cpp commits, I'll recommend this fork if anyone has issues.
LM Studio https://lmstudio.ai/
Please test again using the Default LM Studio Windows preset.
llama.cpp https://github.com/ggerganov/llama.cpp
Backyard AI https://backyard.ai/
Meet Layla,Layla is an AI chatbot that runs offline on your device.No internet connection required.No censorship.Complete privacy.Layla Lite https://www.layla-network.ai/
Layla Lite llama3-8B-DarkIdol-1.1-Q4_K_S-imat.gguf https://huggingface.co/LWDCLS/llama3-8B-DarkIdol-2.3-Uncensored-32K/blob/main/llama3-8B-DarkIdol-2.3-Uncensored-32K-Q4_K_S-imat.gguf?download=true
more gguf at https://huggingface.co/LWDCLS/llama3-8B-DarkIdol-2.3-Uncensored-32K-GGUF-IQ-Imatrix-Request

character

If you want to use vision functionality:

You must use the latest versions of Koboldcpp.

To use the multimodal capabilities of this model and use vision you need to load the specified mmproj file, this can be found inside this model repo. Llava MMProj

You can load the mmproj by using the corresponding section in the interface:
image/png

Thank you:

To the authors for their hard work, which has given me more options to easily create what I want. Thank you for your efforts.

Hastagaras
Gryphe
cgato
ChaoticNeutrals
mergekit
merge
transformers
llama
Nitral-AI
MLP-KTLim
rinna
hfl
Rupesh2
stephenlzc
theprint
Sao10K
turboderp
TheBossLevel123
winglian
.........

llama3-8B-DarkIdol-2.3-Uncensored-32K

This is a merge of pre-trained language models created using mergekit.

Merge Details

Merge Method

This model was merged using the Model Stock merge method using ./llama3-8B-DarkIdol-2.3b as a base.

Configuration

The following YAML configuration was used to produce this model:

yaml

models:
  - model: Sao10K/L3-8B-Niitama-v1
  - model: Hastagaras/Jamet-8B-L3-MK.V-Blackroot
  - model: Nitral-AI/Hathor_Tahsin-L3-8B-v0.85
  - model: turboderp/llama3-turbcat-instruct-8b
  - model: winglian/Llama-3-8b-64k-PoSE
merge_method: model_stock
base_model: winglian/Llama-3-8b-64k-PoSE
dtype: bfloat16

models:
  - model: maldv/badger-writer-llama-3-8b
  - model: underwoods/writer-8b
  - model: Gryphe/Pantheon-RP-1.0-8b-Llama-3
  - model: vicgalle/Roleplay-Llama-3-8B
  - model: cgato/TheSalt-RP-L3-8b-DPO-v0.3.2-e0.15.2
  - model: ./llama3-8B-DarkIdol-2.3a
merge_method: model_stock
base_model: ./llama3-8B-DarkIdol-2.3a
dtype: bfloat16

models:
  - model: Rupesh2/Meta-Llama-3-8B-abliterated
  - model: Orenguteng/Llama-3-8B-LexiFun-Uncensored-V1
  - model: Orenguteng/Llama-3-8B-Lexi-Uncensored
  - model: theprint/Llama-3-8B-Lexi-Smaug-Uncensored
  - model: vicgalle/Unsafe-Llama-3-8B  
  - model: vicgalle/Configurable-Hermes-2-Pro-Llama-3-8B
  - model: ./llama3-8B-DarkIdol-2.3b
merge_method: model_stock
base_model: ./llama3-8B-DarkIdol-2.3b
dtype: bfloat16