Update README.md
Browse files
README.md
CHANGED
@@ -40,7 +40,7 @@ pipeline_tag: text-generation
|
|
40 |
|
41 |
|
42 |
AI ์ ๋น
๋ฐ์ดํฐ ๋ถ์ ์ ๋ฌธ ๊ธฐ์
์ธ Linkbricks์ ๋ฐ์ดํฐ์ฌ์ด์ธํฐ์คํธ์ธ ์ง์ค์ฑ(Saxo) ์ด์ฌ๊ฐ <br>
|
43 |
-
Mistral-Nemo-Instruct-2407 ๋ฒ ์ด์ค๋ชจ๋ธ์ ์ฌ์ฉํด์ H100-80G 8๊ฐ๋ฅผ ํตํด CPT(Continue-Pretraining)->SFT->DPO ํ ํ๊ธ ์ธ์ด ๋ชจ๋ธ<br>
|
44 |
์ฒ๋ง๊ฑด์ ํ๊ธ ๋ด์ค ์ฝํผ์ค๋ฅผ ๊ธฐ์ค์ผ๋ก ๋ค์ํ ํ
์คํฌ๋ณ ํ๊ตญ์ด-์ค๊ตญ์ด-์์ด-์ผ๋ณธ์ด ๊ต์ฐจ ํ์ต ๋ฐ์ดํฐ์ ์ํ ๋ฐ ๋
ผ๋ฆฌํ๋จ ๋ฐ์ดํฐ๋ฅผ ํตํ์ฌ ํ์ค์ผ์ ์ธ์ด ๊ต์ฐจ ์ฆ๊ฐ ์ฒ๋ฆฌ์ ๋ณต์กํ ๋
ผ๋ฆฌ ๋ฌธ์ ์ญ์ ๋์ ๊ฐ๋ฅํ๋๋ก ํ๋ จํ ๋ชจ๋ธ์ด๋ค.<br>
|
45 |
-ํ ํฌ๋์ด์ ๋ ๋จ์ด ํ์ฅ ์์ด ๋ฒ ์ด์ค ๋ชจ๋ธ ๊ทธ๋๋ก ์ฌ์ฉ<br>
|
46 |
-๊ณ ๊ฐ ๋ฆฌ๋ทฐ๋ ์์
ํฌ์คํ
๊ณ ์ฐจ์ ๋ถ์ ๋ฐ ์ฝ๋ฉ๊ณผ ์๋ฌธ, ์ํ, ๋
ผ๋ฆฌํ๋จ ๋ฑ์ด ๊ฐํ๋ ๋ชจ๋ธ<br>
|
@@ -49,7 +49,7 @@ Mistral-Nemo-Instruct-2407 ๋ฒ ์ด์ค๋ชจ๋ธ์ ์ฌ์ฉํด์ H100-80G 8๊ฐ๋ฅผ ํต
|
|
49 |
-Deepspeed Stage=3, rslora ๋ฐ BAdam Layer Mode ์ฌ์ฉ <br><br><br>
|
50 |
|
51 |
Finetuned by Mr. Yunsung Ji (Saxo), a data scientist at Linkbricks, a company specializing in AI and big data analytics <br>
|
52 |
-
CPT(Continue-Pretraining)->SFT->DPO training model based on Mistral-Nemo-Instruct-2407 through 8 H100-80Gs as a Korean language model <br>
|
53 |
It is a model that has been trained to handle Korean-Chinese-English-Japanese cross-training data and 10M korean news corpus and logic judgment data for various tasks to enable cross-fertilization processing and complex Korean logic & math problems. <br>
|
54 |
-Tokenizer uses the base model without word expansion<br>
|
55 |
-Models enhanced with high-dimensional analysis of customer reviews and social posts, as well as coding, writing, math and decision making<br>
|
|
|
40 |
|
41 |
|
42 |
AI ์ ๋น
๋ฐ์ดํฐ ๋ถ์ ์ ๋ฌธ ๊ธฐ์
์ธ Linkbricks์ ๋ฐ์ดํฐ์ฌ์ด์ธํฐ์คํธ์ธ ์ง์ค์ฑ(Saxo) ์ด์ฌ๊ฐ <br>
|
43 |
+
Mistral-Nemo-Instruct-2407 ๋ฒ ์ด์ค๋ชจ๋ธ์ ์ฌ์ฉํด์ H100-80G 8๊ฐ๋ฅผ ํตํด ์ฝ 20%์ ๋์ ํ๋ผ๋ฏธํฐ๋ฅผ ํ๊ตญ์ด CPT(Continue-Pretraining)->SFT->DPO ํ ํ๊ธ ์ธ์ด ๋ชจ๋ธ<br>
|
44 |
์ฒ๋ง๊ฑด์ ํ๊ธ ๋ด์ค ์ฝํผ์ค๋ฅผ ๊ธฐ์ค์ผ๋ก ๋ค์ํ ํ
์คํฌ๋ณ ํ๊ตญ์ด-์ค๊ตญ์ด-์์ด-์ผ๋ณธ์ด ๊ต์ฐจ ํ์ต ๋ฐ์ดํฐ์ ์ํ ๋ฐ ๋
ผ๋ฆฌํ๋จ ๋ฐ์ดํฐ๋ฅผ ํตํ์ฌ ํ์ค์ผ์ ์ธ์ด ๊ต์ฐจ ์ฆ๊ฐ ์ฒ๋ฆฌ์ ๋ณต์กํ ๋
ผ๋ฆฌ ๋ฌธ์ ์ญ์ ๋์ ๊ฐ๋ฅํ๋๋ก ํ๋ จํ ๋ชจ๋ธ์ด๋ค.<br>
|
45 |
-ํ ํฌ๋์ด์ ๋ ๋จ์ด ํ์ฅ ์์ด ๋ฒ ์ด์ค ๋ชจ๋ธ ๊ทธ๋๋ก ์ฌ์ฉ<br>
|
46 |
-๊ณ ๊ฐ ๋ฆฌ๋ทฐ๋ ์์
ํฌ์คํ
๊ณ ์ฐจ์ ๋ถ์ ๋ฐ ์ฝ๋ฉ๊ณผ ์๋ฌธ, ์ํ, ๋
ผ๋ฆฌํ๋จ ๋ฑ์ด ๊ฐํ๋ ๋ชจ๋ธ<br>
|
|
|
49 |
-Deepspeed Stage=3, rslora ๋ฐ BAdam Layer Mode ์ฌ์ฉ <br><br><br>
|
50 |
|
51 |
Finetuned by Mr. Yunsung Ji (Saxo), a data scientist at Linkbricks, a company specializing in AI and big data analytics <br>
|
52 |
+
about 20% of total parameters Korean CPT(Continue-Pretraining)->SFT->DPO training model based on Mistral-Nemo-Instruct-2407 through 8 H100-80Gs as a Korean language model <br>
|
53 |
It is a model that has been trained to handle Korean-Chinese-English-Japanese cross-training data and 10M korean news corpus and logic judgment data for various tasks to enable cross-fertilization processing and complex Korean logic & math problems. <br>
|
54 |
-Tokenizer uses the base model without word expansion<br>
|
55 |
-Models enhanced with high-dimensional analysis of customer reviews and social posts, as well as coding, writing, math and decision making<br>
|