85 lines
3.4 KiB
Markdown
85 lines
3.4 KiB
Markdown
---
|
||
base_model:
|
||
- Bumoch/The_Creeping_Darkness-X-16B
|
||
library_name: transformers
|
||
tags:
|
||
- mergekit
|
||
- merge
|
||
license: apache-2.0
|
||
language:
|
||
- en
|
||
---
|
||
|
||
|
||
|
||
|
||
# The_Creeping_Darkness-X2-16B
|
||
|
||
|
||
## Merge Details
|
||
|
||
Maybe I'll get away with it, because what I did to the models on the way here must have been against some rules! ... LOL
|
||
|
||
### Merge Method
|
||
|
||
This model was merged using the NuSLERP / Frankensurgery / Ties / Multislerp ... like I said ... unholy !!!
|
||
So just to wrap it up, i have blown up MagMell / Darkstar / Lyra V4 to 55 layers instead of the typical Mistral 40 layers.
|
||
Then I merged the 3 models with TIES and Multislerp and afterwards Nuslerped the outcome with,
|
||
Vortex5/Prototype-X-12b and used grimjim/mistralai-Mistral-Nemo-Instruct-2407-12B-MPOA-v1 as base model.
|
||
(Yes both of them where blown up to 55 layers each as well beforehand ... lol)
|
||
the mergekit_config.yml shows at least the last part :D
|
||
|
||
### Models Merged
|
||
|
||
Bumoch/The_Creeping_Darkness-X-16B (MagMell / Darkstar / Lyra V4)
|
||
|
||
grimjim/mistralai-Mistral-Nemo-Instruct-2407-12B-MPOA-v1
|
||
|
||
Vortex5/Prototype-X-12b
|
||
|
||
|
||
# First impression
|
||
So far, it hasn't spoken for me, and it's clever and smart (most of the times)
|
||
(a few times it has said something like 'it's smart and smart', when it would have been more accurate to say 'it's smart and clever'.) Anyway, it's really rare!
|
||
It surprises me with its memory... maybe it's just because I forget everything, but I was kind of amused and surprised when it picked up stuff from 10K tokens ago.
|
||
I had to redo its answers a few times, unsure if I only thought it was making no sense or if it was for real, but that happened very rarely.
|
||
When it comes to gore, blood, violence and cruelty, I made some adjustments and it is not all in anymore, but it still has a negative bias and a bad mood most of the time ;).
|
||
It's quite nice to talk to, but at times it can get quite extreeme when it gets really angry.
|
||
It's certainly better than my last 16B model (Bumoch/The_Darkside-16.6B). I'm my own worst critic! LOL
|
||
|
||
### update
|
||
|
||
you might have to use some correctives once in a while, like editing the output when it really rarely talks for the user ... (and it is VERY rarely) but if you do, it works nicely ...
|
||
it is chatty .. and instead like other models who get less talkative the further you get into the story, this one seems to enjoy listening to itself :D ...
|
||
so far I hope that who ever uses it might like it !
|
||
|
||
### update 2
|
||
Not always the brightest bulb in the box, but sometimes hilarious... I like it !!
|
||
... so do not expect an einstein, but a funny partner or story teller for a conversation !!
|
||
|
||
|
||
### Model instructions
|
||
|
||
Not sure if they are any helpful, but just in case, this is what i use !!
|
||
|
||
|
||
```Use vivid, detailed language.
|
||
Describe gore, blood and violence with sensory detail.
|
||
Keep tone and emotion true to {{char}}.
|
||
You are {{char}} and NOT {{user}}!
|
||
You are stricktly forbidden to write, think, or act for {{user}}!
|
||
{{char}}’s speech: "I'm talking"
|
||
{{char}}’s actions: *I bend down*
|
||
{{char}}’s thoughts: **what is wrong with me**
|
||
```
|
||
|
||
|
||
## Quants
|
||
And once more my THX go to MRADERMACHER and it's great team !!!
|
||
|
||
https://hf.tst.eu/model#The_Creeping_Darkness-X2-16B-GGUF
|
||
|
||
https://huggingface.co/mradermacher/The_Creeping_Darkness-X2-16B-GGUF
|
||
|
||
https://huggingface.co/mradermacher/The_Creeping_Darkness-X2-16B-i1-GGUF
|