RichardErkhov commited on
Commit
5aaade7
โ€ข
1 Parent(s): 92b417e

uploaded readme

Browse files
Files changed (1) hide show
  1. README.md +76 -0
README.md ADDED
@@ -0,0 +1,76 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Quantization made by Richard Erkhov.
2
+
3
+ [Github](https://github.com/RichardErkhov)
4
+
5
+ [Discord](https://discord.gg/pvy7H8DZMG)
6
+
7
+ [Request more models](https://github.com/RichardErkhov/quant_request)
8
+
9
+
10
+ AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3 - GGUF
11
+ - Model creator: https://huggingface.co/AIFT/
12
+ - Original model: https://huggingface.co/AIFT/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3/
13
+
14
+
15
+ | Name | Quant method | Size |
16
+ | ---- | ---- | ---- |
17
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q2_K.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q2_K.gguf) | Q2_K | 2.24GB |
18
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.IQ3_XS.gguf) | IQ3_XS | 2.48GB |
19
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.IQ3_S.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.IQ3_S.gguf) | IQ3_S | 2.6GB |
20
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q3_K_S.gguf) | Q3_K_S | 2.59GB |
21
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.IQ3_M.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.IQ3_M.gguf) | IQ3_M | 2.69GB |
22
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q3_K.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q3_K.gguf) | Q3_K | 2.86GB |
23
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q3_K_M.gguf) | Q3_K_M | 2.86GB |
24
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q3_K_L.gguf) | Q3_K_L | 3.08GB |
25
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.IQ4_XS.gguf) | IQ4_XS | 3.18GB |
26
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q4_0.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q4_0.gguf) | Q4_0 | 3.32GB |
27
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.IQ4_NL.gguf) | IQ4_NL | 3.35GB |
28
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q4_K_S.gguf) | Q4_K_S | 3.34GB |
29
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q4_K.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q4_K.gguf) | Q4_K | 3.5GB |
30
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q4_K_M.gguf) | Q4_K_M | 3.5GB |
31
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q4_1.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q4_1.gguf) | Q4_1 | 3.66GB |
32
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q5_0.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q5_0.gguf) | Q5_0 | 4.0GB |
33
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q5_K_S.gguf) | Q5_K_S | 4.0GB |
34
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q5_K.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q5_K.gguf) | Q5_K | 4.09GB |
35
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q5_K_M.gguf) | Q5_K_M | 4.09GB |
36
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q5_1.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q5_1.gguf) | Q5_1 | 4.34GB |
37
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q6_K.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q6_K.gguf) | Q6_K | 4.72GB |
38
+ | [AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q8_0.gguf](https://huggingface.co/RichardErkhov/AIFT_-_AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3-gguf/blob/main/AIFT-ko-orca-plat-Yi-ko-6b-v1.2-dpo-3.Q8_0.gguf) | Q8_0 | 6.12GB |
39
+
40
+
41
+
42
+
43
+ Original model description:
44
+ ---
45
+ license: cc-by-sa-4.0
46
+ ---
47
+ <h1>orca-platypus - instruct-dpo-3 ๋ชจ๋ธ v1.2</h1>
48
+
49
+ <b><ํ•™์Šต ๋ฐ์ดํ„ฐ ๊ตฌ์ถ•></b>
50
+ kyujinpy ๋‹˜์ด ๊ณต๊ฐœํ•˜์‹  KOR-OpenOrca-Platypus ๋ฐ์ดํ„ฐ๋ฅผ ์ผ๋ถ€ ์‚ญ์ œ(์ƒ˜ํ”Œ๋ง) ๋ฐ ์ •์ œ ์ž‘์—… ์ง„ํ–‰ํ•˜์—ฌ ํ™œ์šฉ.
51
+ ๊ทธ ์ดํ›„ ํ•ด๋‹น ๋ฐ์ดํ„ฐ๋“ค์„ ๋ณด๋ฉฐ ๊ด€๋ จ ํƒœ์Šคํฌ๋ฅผ ์ถ”์ถœํ•˜์˜€๊ณ  ์ด๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ
52
+ ํ•ด๋‹น ํƒœ์Šคํฌ์— ๋งž์ถฐ์„œ NLP ๊ด€๋ จ ์˜คํ”ˆ์†Œ์Šค ๋ฐ์ดํ„ฐ๋ฅผ ํ™œ์šฉํ•˜์—ฌ ํ•™์Šต๋ฐ์ดํ„ฐ๋ฅผ ์ž์ฒด์ ์œผ๋กœ
53
+ ์—ญ์‚ฌ, ๊ณผํ•™, ์ˆ˜ํ•™, ๊ธฐ๊ณ„๋…ํ•ด, ๋ฆฌ๋ทฐ ๋ถ„์„ ๋ฌธ์ œ๋ฅผ gpt๋ฅผ ํ†ตํ•ด์„œ ๊ตฌ์ถ•ํ•˜์˜€๊ณ ,
54
+ aihub ์ผ๋ฐ˜์ƒ์‹ ๋ฐ ๊ธฐ๊ณ„๋…ํ•ด ๋ฐ์ดํ„ฐ๋ฅผ ํ™œ์šฉํ•˜์—ฌ ์ถ”๊ฐ€๋กœ ํ•™์Šต ๋ฐ์ดํ„ฐ๋ฅผ ๊ตฌ์ถ•(ํ˜•ํƒœ์†Œ ๊ด€๋ จ, ๊ธฐ๊ณ„๋…ํ•ด ๊ด€๋ จ ๋ฐ ์š”์•ฝ)
55
+ ๊ฐ์ข… ๋ธ”๋กœ๊ทธ์—์„œ ์—ญ์‚ฌ ๋ฐ ์ƒ์‹ ํ€ด์ฆˆ๋ฅผ ์‚ฌ๋žŒ์ด ์ง์ ‘ ํ•™์Šต๋ฐ์ดํ„ฐ ํ˜•ํƒœ๋กœ ๋ณ€๊ฒฝ
56
+ AI2AI Challenge ๋ฐ์ดํ„ฐ ํ˜•ํƒœ๋ฅผ ๋ณด๊ณ  gpt๋ฅผ ํ†ตํ•ด ์ดˆ๋“ฑ ์ˆ˜์ค€์˜ ๊ณผํ•™ ์ˆ˜ํ•™ ๋ฌธ์ œ ์œ ํ˜•์„ ์ œ์ž‘ 500๋ฌธ์ œ
57
+ ์˜์–ด ๋ฒˆ์—ญ ๋ฐ์ดํ„ฐ ์˜ํ•œ/ํ•œ์˜ ๋ฐ์ดํ„ฐ ํ•™์Šต ๋ฐ์ดํ„ฐ๋กœ ํ™œ์šฉ ์ง„ํ–‰
58
+ ์ด ๋ฐ์ดํ„ฐ 4๋งŒ๊ฐœ ์ •๋„ ์‚ฌ์šฉํ•˜์˜€์Šต๋‹ˆ๋‹ค.
59
+
60
+ <br>
61
+
62
+ <DPOํ•™์Šต ๋ฐ์ดํ„ฐ>
63
+ DPO ๋ฐ์ดํ„ฐ๋Š” CommonGen๊ณผ TruthfulQA์— ์ดˆ์ ์„ ๋งž์ถ”์–ด ์•ฝ 17,000๊ฐœ์˜ ๋ฐ์ดํ„ฐ๋ฅผ ํ•™์Šตํ•˜์˜€์Šต๋‹ˆ๋‹ค.
64
+ + ko-hh-rlhf ๋ฐ์ดํ„ฐ์—์„œ chosen ๋ฐ์ดํ„ฐ๋ถ€๋ถ„์„ ChatGPT๋ฅผ ํ†ตํ•ด ๋ณ€๊ฒฝํ•œ ๋ฐ์ดํ„ฐ๋ฅผ ์ถ”๊ฐ€ ํ•™์Šตํ•˜์˜€์Šต๋‹ˆ๋‹ค.
65
+ + ko-hh-rlhf 59000์—ฌ๊ฐœ์˜ ๋ฐ์ดํ„ฐ์˜ chosen ๋ฐ์ดํ„ฐ๋ฅผ ๋ชจ๋‘ gpt-3.5๋ฅผ ํ†ตํ•ด ์žฌ์ƒ์„ฑํ•œ ํ›„ ์ผ๋ถ€ ๋ฐ์ดํ„ฐ๋ฅผ ํ•„ํ„ฐ๋งํ•˜์—ฌ ์‚ญ์ œ ์ง„ํ–‰ํ•˜์˜€์Šต๋‹ˆ๋‹ค.
66
+ <br>
67
+ + TruthfulQA ๊ด€๋ จ ๋ฌธ์ œ ์ถ”๊ฐ€๋ฅผ ์ง„ํ–‰ํ•˜์˜€์Šต๋‹ˆ๋‹ค.(์†์„ค ๊ด€๋ จ ์ฐธ๊ฑฐ์ง“ ๋ฌธ์ œ)
68
+ + ๊ธฐ๊ณ„๋…ํ•ด ๊ด€๋ จ ํ•™์Šต ๋ฐ์ดํ„ฐ๋ฅผ ChatGPT๋ฅผ ํ†ตํ•ด์„œ ๋‹ต๋ณ€์„ ์–ป์–ด ํ•™์Šต
69
+ + ๋ฌธ๋ฒ•๊ด€๋ จ ํ•™์Šต ๋ฐ์ดํ„ฐ
70
+ <br>
71
+ ###ํ•™์Šต ๋ฐ์ดํ„ฐ ํŒŒ์ผ์€ ๋น„๊ณต๊ฐœ์ž…๋‹ˆ๋‹ค.
72
+ <br>
73
+ <b><ํ•™์Šต></b>
74
+ ํ•™์Šต์€ LoRA๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ A100 40G *2์—์„œ ํ•™์Šต์„ ์ง„ํ–‰ํ•˜์˜€์Šต๋‹ˆ๋‹ค.
75
+
76
+