File size: 1,104 Bytes
7e97e78
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
This is the original FP16 result of the model created using chargoddard's frankenllama script so that others interested in further experimentation with the results may do so.
# WARNING: this model is very unpredictable.
This model is an experiment using the frankenstein script from 
https://huggingface.co/chargoddard/llama2-22b
Except I decided to use it with two models that have already been extensively finetuned.
With https://huggingface.co/TheBloke/Llama-2-13B-Chat-fp16 as the base model
and 
https://huggingface.co/Aeala/Enterredaas-33b as the donor model. 

The resulting model is surprisingly coherent and still responds well to the llama2chat prompt format 
```[INST]<<SYS>><</SYS>>[/INST]``` and still has most of llama2chat's bubbly/giddy personality but more gritty and visceral. 
It makes occasional "typos" along with some other quirks so it was not completely unscathed by the frankensteining process. 
I plan to massage it over with a LoRA in the near future to bring it into more harmony but in the meantime it is available now for your enjoyment. 

Use cases: Chat/RP not much else.