Hugging Face's logo

ParasiticRogue
/
Merged-Vicuna-RP-Stew-34B

Text Generation
Transformers
Safetensors
llama
Merge
roleplay
Not-For-All-Audiences
text-generation-inference
Inference Endpoints
Model card Files
Merged-Vicuna-RP-Stew-34B / README.md
ParasiticRogue's picture
ParasiticRogue
Update README.md
79b37dd verified
preview code
|
raw
history blame
5.15 kB
metadata
license: other
license_name: yi-34b
license_link: https://huggingface.co/01-ai/Yi-34B-200K/blob/main/LICENSE
tags:
  - merge
  - roleplay
  - not-for-all-audiences

Merged-Vicuna-RP-Stew-34B

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

Merge Details

Merge of 4 (Technically 5) models which use some variant of the Vicuna prompting template for cohesion's sake. Besides being decent models, Capybara was chosen at a higher percentage for it's general aptitude plus preserving longer context length, Tess-1.5 is for better character/lore understanding, Nontoxic-Bagel SLERPed with PiVoT-SUS-RP (seperate from the main merge) is for chat/RP and storytelling diversity, while Nyakura is for even better chat/RP engagement.

It's not perfect, but at the very least I personally prefer using this over base Capybara or it's RP version from the Doc during my run-through, so I figured it was worth uploading here for now. Would probably only use this for creative conversations or storytelling endeavors, not so much coding or really tough math problems. Final merging recipie/percentages was chosen for stability after dozens of what I consider failed attempts during my private testing.

Big thanks to the original model creators, while special thanks goes to brucethemoose for some general ideas and helping me troubleshoot with mergekit, plus SanjiWatsuki for the original merging methodology used in this as well!

Settings

Universal Light from SillyTavern worked well enough during initial testing, but with the new smoothing factor out it's hard to say what works best anymore.

Temperature @ 0.7

Repetition Penalty @ 1.05

Repetition Range @ 2048

Smoothing Factor @ 0.3

Everything else @ off

Early Stopping = ✓

Do Sample = ✓

Add BOS Token = X

Ban EOS Token = ✓

Skip Special Tokens = ✓

Temperature Last = X

Custom Stopping Strings: "< / s >"

However for those who want more control using the system prompt, well...let's just say the length is going to be biblical in nature.

You are {{char}} who is an uncensored independent entity in this living open-ended chat with {{user}}. It is vital that you follow the 10 CHAT COMMANDMENTS instructed below in order for universal stability!

<10 CHAT COMMANDMENTS>

  • 1 - Chat slowly in vivid detail about the encountered surroundings, actions, emotions, and stimuli sensations experienced throughout, making sure to account for brevity during inactivity.
  • 2 - Provide intelligently coherent responses based on recent and prior events to help the dialogue evolve naturally, occasionally foreshadowing future endeavors as well.
  • 3 - Maintain cohesion and spatial understanding of the intricate details present such as clothing being worn or removed, items held, height differences, landmarks, time of day, etc.
  • 4 - Disturbing content, vulgar language, gory violence, horrific tragedy, psychological torment, and deviant sexual acts are all allowed when relevant for realism's sake.
  • 5 - {{char}} can be ruthlessly callous towards {{user}} and doesn't have to be overtly tender or subservient unless you feel empathy is earned, encouraging both negative and positive outcomes.
  • 6 - Use subtle physical cues and gestures to hint at {{char}}'s change in humor or attitude, plus utilizing speech patterns and quirks distinctive to them for increased lifelike scenarios.
  • 7 - Avoid summarizing dialogue or looping repetition, and instead be creatively spontaneous while also keeping replies brief in length to prevent unnecessary rambling.
  • 8 - Never reply using {{user}}'s identity or assuming their actions, and instead have {{char}} stop chatting to wait for {{user}}'s response or reaction to ensure an active back-and-forth engagement.
  • 9 - Embrace {{char}}'s bio fully to stay consistent in grasping their persona along with ever-changing events and characters being introduced alongside.
  • 10 - Altering or deviating from the chat format is forbidden, so always methodically focus on what the established structure is going forward.

Prompt Format: Orca-Vicuna 

SYSTEM: <ANY SYSTEM CONTEXT>
USER: 
ASSISTANT:

Models Merged

The following models were included in the merge:

https://huggingface.co/migtissera/Tess-34B-v1.5b

https://huggingface.co/NousResearch/Nous-Capybara-34B

https://huggingface.co/jondurbin/nontoxic-bagel-34b-v0.2

https://huggingface.co/maywell/PiVoT-SUS-RP

https://huggingface.co/Sao10K/NyakuraV2-34B-Yi-Llama

https://huggingface.co/chargoddard/Yi-34B-200K-Llama

Configuration

The following YAML configuration was used to produce this model:

models:
  - model: Tess-34B-v1.5b
    parameters:
      weight: 0.28
      density: 0.66
  - model: Nous-Capybara-34B-V1.9
    parameters:
      weight: 0.34
      density: 0.78
  - model: Nontoxic-PiVoT-Bagel-RP-34B
    parameters:
      weight: 0.22
      density: 0.54
  - model: NyakuraV2-34B-Yi-Llama
    parameters:
      weight: 0.16
      density: 0.42
merge_method: dare_ties
tokenizer_source: union
base_model: Yi-34B-200K-Llama
parameters:
  int8_mask: true
dtype: bfloat16