lmg-anon commited on
Commit
1a6fbf5
β€’
1 Parent(s): 768ab86

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +49 -2
README.md CHANGED
@@ -11,6 +11,53 @@ base_model: rinna/llama-3-youko-8b
11
  pipeline_tag: translation
12
  ---
13
 
14
- # Model Card
15
 
16
- TBD
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
11
  pipeline_tag: translation
12
  ---
13
 
14
+ # VNTL
15
 
16
+ This is an [LLaMA 3 Youko](https://huggingface.co/rinna/llama-3-youko-8b) qlora created using a slightly modified version of the [VNTL-v3.1-1k](https://huggingface.co/datasets/lmg-anon/VNTL-v3.1-1k) dataset, concatenated with the [VNTL-Chat](https://huggingface.co/datasets/lmg-anon/VNTL-Chat) dataset.
17
+
18
+ This was trained mostly with the same hyperparameters as the [VNTL 7B v0.3.1 lora](https://huggingface.co/lmg-anon/vntl-7b-v0.3.1-lora), the differences are:
19
+ - Added \["<\<METADATA>>", "<\<TRANSLATE>>", "<\<JAPANESE>>", "<\<ENGLISH>>", "<\<CHAT>>", "<\<HUMAN>>", "<\<LLM>>"\] as special tokens.
20
+ - Trained the \["embed_tokens", "lm_head"\] layers.
21
+ - 10x smaller learning rate, 0.00065 -> 0.000065.
22
+
23
+ This version also includes a new "chat mode", which was lazily trained just to find out how that would impact the end result. I think this ended up quite good for breaking down or explaining Japanese sentences, but it is terrible for most other things, which is expected given the nature of the VNTL-Chat dataset. To be honest, I wasn't aiming for it to be good, so the fact that it works at all is very nice.
24
+
25
+ **Eval Loss**: 0.8
26
+
27
+ ## Translation Prompt
28
+
29
+ This is an prompt example for translation:
30
+ ```
31
+ <<METADATA>>
32
+ [character] Name: Uryuu Shingo (η“œη”Ÿ 新吾) | Gender: Male | Aliases: Onii-chan (γŠε…„γ‘γ‚ƒγ‚“)
33
+ [character] Name: Uryuu Sakuno (η“œη”Ÿ ζ‘œδΉƒ) | Gender: Female
34
+ <<TRANSLATE>>
35
+ <<JAPANESE>>
36
+ [ζ‘œδΉƒ]: γ€Žβ€¦β€¦γ”γ‚γ‚“γ€
37
+ <<ENGLISH>>
38
+ [Sakuno]: γ€Ž... Sorry.』<|end_of_text|>
39
+ <<JAPANESE>>
40
+ [新吾]: γ€Œγ†γ†γ‚“γ€γ“γ†θ¨€γ£γ‘γ‚ƒγͺγ‚“γ γ‘γ©γ€θΏ·ε­γ§γ‚ˆγ‹γ£γŸγ‚ˆγ€‚ζ‘œδΉƒγ―ε―ζ„›γ„γ‹γ‚‰γ€γ„γ‚γ„γ‚εΏƒι…γ—γ‘γ‚ƒγ£γ¦γŸγ‚“γ γžδΏΊγ€
41
+ <<ENGLISH>>
42
+ ```
43
+
44
+ The generated translation for that prompt, with temperature 0, is:
45
+ ```
46
+ [Shingo]: γ€ŒNo, I know this isn't the best way to say it, but I'm glad you got lost. You're so cute that I was worried about you.」
47
+ ```
48
+
49
+ ## Chat Prompt
50
+
51
+ This is an prompt example for chat:
52
+ ```
53
+ <<CHAT>>
54
+ <<HUMAN>>
55
+ Hi! Can you break this down to me? Like, what is the meaning of "ζ‘œδΉƒγ―ε―ζ„›γ„γ‹γ‚‰"?
56
+ "ううん、こう言っけゃγͺγ‚“γ γ‘γ©γ€θΏ·ε­γ§γ‚ˆγ‹γ£γŸγ‚ˆγ€‚ζ‘œδΉƒγ―ε―ζ„›γ„γ‹γ‚‰γ€γ„γ‚γ„γ‚εΏƒι…γ—γ‘γ‚ƒγ£γ¦γŸγ‚“γ γžδΏΊγ€"
57
+ <<LLM>>
58
+ ```
59
+
60
+ The generated response for that prompt, with temperature 0, is:
61
+ ```
62
+ Certainly! The phrase "ζ‘œδΉƒγ―ε―ζ„›γ„γ‹γ‚‰" literally translates to "Because Sakura-no is cute." In Japanese culture, there is a strong emphasis on physical attractiveness and cuteness, especially when it comes to young girls or women. This phrase suggests that the speaker finds Sakura-no (presumably a girl) to be very attractive and endearing in appearance.
63
+ ```