Triangle104's picture
Update README.md
8bb67f8 verified
|
raw
history blame
4.27 kB
metadata
license: other
tags:
  - llama-cpp
  - gguf-my-repo
base_model: TheDrummer/Cydonia-22B-v1.2

Triangle104/Cydonia-22B-v1.2-Q4_K_M-GGUF

This model was converted to GGUF format from TheDrummer/Cydonia-22B-v1.2 using llama.cpp via the ggml.ai's GGUF-my-repo space. Refer to the original model card for more details on the model.


Model details:

I christen this model, 'Miqu 2 Mini' - @invisietch

Description

Strange, it feels like DRY is permanently on ... In general, I like how it feels more alive. More slang has been added, maybe this is the merit of my card, but still.

The model is very cohesive, expressive, and overall intelligent. It's able to write engaging and impactful content, carry out roleplay mostly effectively, and manage to respond well.

It shocked me with the first output by introducing a character that is not written anywhere in the card. This character immediately gave the impression that there is a history there with King Severin and that there is immediately something to build off of. It's maintaining creativity and keeping things nearly constantly shifting. It's remaining aware of who is where and what they are doing. It's maintaining a good balance of action and dialogue as well as holding to the style of talking for the second character it made up and introduced. ... I feel like v1.2 is much, much better with creativity and letting the player build off what the model is able to bring in all by itself rather than, like most Mistral tunes, keeping the roleplay to solely what information is provided in the card.

When I swapped to v1.2 I was impressed that it seemed just as good as OG Small in intelligence while being a lot more creative (and much more moist)

v1.2 real good in my experience so far (i don't comment pretty much ever but i do want to put it out there that i agree)

It got creative and added a whole other person whose mannerisms and speech imply a history there. That could be fun to unravel and see what it comes up with. ... It's maintaining creativity and keeping things nearly constantly shifting. It's remaining aware of who is where and what they are doing. It's maintaining a good balance of action and dialogue.

v1.2 is much gooder. Omg. Your dataset is amazing. I'm not getting far with these two because I have to keep crawling away from my pc to cool off. 🥵

Arsenal (Supported Chat Templates)

Metharme (a.k.a. Pygmalion in ST) for RP / Story
Text Completion for RP
Mistral for Instruct / RP / Story
You can mix it up and see which works best for you.

Favorite RP Format

action Dialogue thoughts Dialogue narration in 1st person PoV

No one's gonna take me alive Time has come to make things right You and I must fight for our rights You and I must fight to survive


Use with llama.cpp

Install llama.cpp through brew (works on Mac and Linux)

brew install llama.cpp

Invoke the llama.cpp server or the CLI.

CLI:

llama-cli --hf-repo Triangle104/Cydonia-22B-v1.2-Q4_K_M-GGUF --hf-file cydonia-22b-v1.2-q4_k_m.gguf -p "The meaning to life and the universe is"

Server:

llama-server --hf-repo Triangle104/Cydonia-22B-v1.2-Q4_K_M-GGUF --hf-file cydonia-22b-v1.2-q4_k_m.gguf -c 2048

Note: You can also use this checkpoint directly through the usage steps listed in the Llama.cpp repo as well.

Step 1: Clone llama.cpp from GitHub.

git clone https://github.com/ggerganov/llama.cpp

Step 2: Move into the llama.cpp folder and build it with LLAMA_CURL=1 flag along with other hardware-specific flags (for ex: LLAMA_CUDA=1 for Nvidia GPUs on Linux).

cd llama.cpp && LLAMA_CURL=1 make

Step 3: Run inference through the main binary.

./llama-cli --hf-repo Triangle104/Cydonia-22B-v1.2-Q4_K_M-GGUF --hf-file cydonia-22b-v1.2-q4_k_m.gguf -p "The meaning to life and the universe is"

or

./llama-server --hf-repo Triangle104/Cydonia-22B-v1.2-Q4_K_M-GGUF --hf-file cydonia-22b-v1.2-q4_k_m.gguf -c 2048