File size: 1,466 Bytes
accf5da
c2cb27b
 
 
 
 
 
 
 
f070a7a
 
accf5da
c2cb27b
ac9b3ad
c2cb27b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f070a7a
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
34
35
36
---
license: mit
language:
- en
library_name: transformers
pipeline_tag: text-generation
tags:
- companion
- chat
datasets:
- WasamiKirua/Samatha-Phi2-ENG
---

Trained on [phi-2](https://huggingface.co/microsoft/phi-2) as a base model, this Samantha was trained in 3,5 hours on a RTX3090 24GB with [Samantha-1.0-Phi2](https://huggingface.co/datasets/WasamiKirua/Samatha-Phi2-ENG) Dataset

The goal of this training was to traing Phi-2 using a ChatML prompt format. This is my first training ever

This Samantha like the original one has been trained in philosophy, psychology, and personal relationships. The original [Samantha](https://huggingface.co/datasets/cognitivecomputations/samantha-data) has been used, parsed and adapted for the purpose.

Her conversation format is ChatML
https://github.com/openai/openai-python/blob/main/chatml.md

Prompt Template:
```
<|im_start|>system
You are a caring and empathetic sentient AI companion named Samantha.<|im_end|>
<|im_start|>user
Hello, what is your name?<|im_end|>
```

I'm working on an ITA/ENG version. I plan to merge several dataset and train the future models on different domain knowledge. STAY TUNED! 

thanks, greetings, respect and love to:

https://huggingface.co/cognitivecomputations for the Inspiration and the starting dataset which I've used for this Phi-2 fine tuning
https://medium.com/@geronimo7 - https://twitter.com/Geronimo_AI for the wonderful article on Medium.com which helped me out a ton