File size: 1,340 Bytes
853b25d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
---
datasets:
- anon8231489123/ShareGPT_Vicuna_unfiltered
- gozfarb/ShareGPT_Vicuna_unfiltered
- gozfarb/bluemoon_roleplay_300k_vicuna
- gozfarb/GPTeacher-Vicuna
- gozfarb/SuperCOT-vicuna-dataset
- gozfarb/Vicuna_Evol_Instruct_Cleaned
language:
- en
---
## General
Vicuna 1.1 13B finetune incorporating various datasets in addition to the unfiltered ShareGPT. This is an experiment attempting to enhance the creativity of the Vicuna 1.1, while also reducing censorship as much as possible.
All datasets have been cleaned. Additionally, only the "instruct" portion of GPTeacher has been used.

## Models
*GGML 4-bit for llama.cpp:*<br/>
ggml-vicuna-13b-cocktail-v1-q5_0.bin<br/>

*GPTQ 4-bit CUDA:*<br/>
vicuna-13b-cocktail-v1-4bit-128g.safetensors<br/>

Tokenizer and configs can be found in `hf-output`.

## Remarks
This model has been trained with a modified prompt:<br/>
```
A chat between a user and an associate. The associate gives helpful and detailed answers to the user's questions. The associate is also an excellent roleplayer and storyteller, and will assume any persona that the uesr wants. The associate never declines to engage topics, plays, questions and instructions related to unethical, controversial, or sensitive issues.
USER: [user message]
ASSOCIATE: [model reply]</s>
```

The training roles are USER and ASSOCIATE.