Update README.md
Browse files
README.md
CHANGED
|
@@ -40,32 +40,6 @@ Command r plus ๋ชจ๋ธ์ ์ด์ฉํ์ฌ ์์ฒด ๊ตฌ์ถํ RAG ํนํ ๋ฐ์ดํฐ์
,
|
|
| 40 |
}
|
| 41 |
```
|
| 42 |
|
| 43 |
-
## ํ์ต ํ๊ฒฝ ๋ฐ ํ๋ผ๋ฏธํฐ
|
| 44 |
-
- ํ๋ ํ๊ฒฝ
|
| 45 |
-
- H100(80GB) * 8
|
| 46 |
-
- ํ๋ผ๋ฏธํฐ
|
| 47 |
-
- tokenizer_model_mex_length 4500
|
| 48 |
-
- use_flash_attn True
|
| 49 |
-
- num_train_epochs 3.0
|
| 50 |
-
- weight_decay 0.001
|
| 51 |
-
- lr_scheduler_type "linear"
|
| 52 |
-
- per_device_train_batch_size 1
|
| 53 |
-
- gradient_accumulation_steps 64
|
| 54 |
-
- learning_rate 5e-06
|
| 55 |
-
- bf16 True
|
| 56 |
-
- deepspeed ds_stage2.json
|
| 57 |
-
|
| 58 |
-
## ์ฌ์ฉ ๋ฐ์ดํฐ์
|
| 59 |
-
- AIhub 16 ํ์ ๋ฌธ์ ๋์ ๊ธฐ๊ณ๋
ํด ๋ฐ์ดํฐ
|
| 60 |
-
- AIhub 17 ๋ด์ค ๊ธฐ์ฌ ๊ธฐ๊ณ๋
ํด ๋ฐ์ดํฐ
|
| 61 |
-
- AIhub 21 ๋์์๋ฃ ๊ธฐ๊ณ๋
ํด
|
| 62 |
-
- AIhub 149 ํ ์ ๋ณด ์ง์์๋ต ๋ฐ์ดํฐ
|
| 63 |
-
- AIhub 150 ์ซ์์ฐ์ฐ ๊ธฐ๊ณ๋
ํด ๋ฐ์ดํฐ
|
| 64 |
-
- AIhub 151 ๊ธ์ต, ๋ฒ๋ฅ ๋ฌธ์ ๊ธฐ๊ณ๋
ํด ๋ฐ์ดํฐ
|
| 65 |
-
- kyujinpy/KoCoT_2000
|
| 66 |
-
- MarkrAI/KoCommercial-Dataset
|
| 67 |
-
- CarrotAI/ko-instruction-dataset
|
| 68 |
-
- heegyu/CoT-collection-ko
|
| 69 |
|
| 70 |
## RAG Prompt
|
| 71 |
```
|
|
@@ -115,6 +89,33 @@ response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
|
|
| 115 |
print(response)
|
| 116 |
```
|
| 117 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 118 |
## Contact us
|
| 119 |
- ์จ๋ก๋ง์ธ๋
|
| 120 |
- ์์ธ์ ๊ด์
๊ตฌ ๋จ๋ถ์ํ๋ก 1802, 2์ธต
|
|
|
|
| 40 |
}
|
| 41 |
```
|
| 42 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 43 |
|
| 44 |
## RAG Prompt
|
| 45 |
```
|
|
|
|
| 89 |
print(response)
|
| 90 |
```
|
| 91 |
|
| 92 |
+
## ํ์ต ํ๊ฒฝ ๋ฐ ํ๋ผ๋ฏธํฐ
|
| 93 |
+
- ํ๋ ํ๊ฒฝ
|
| 94 |
+
- H100(80GB) * 8
|
| 95 |
+
- ํ๋ผ๋ฏธํฐ
|
| 96 |
+
- tokenizer_model_mex_length 4500
|
| 97 |
+
- use_flash_attn True
|
| 98 |
+
- num_train_epochs 3.0
|
| 99 |
+
- weight_decay 0.001
|
| 100 |
+
- lr_scheduler_type "linear"
|
| 101 |
+
- per_device_train_batch_size 1
|
| 102 |
+
- gradient_accumulation_steps 64
|
| 103 |
+
- learning_rate 5e-06
|
| 104 |
+
- bf16 True
|
| 105 |
+
- deepspeed ds_stage2.json
|
| 106 |
+
|
| 107 |
+
## ์ฌ์ฉ ๋ฐ์ดํฐ์
|
| 108 |
+
- AIhub 16 ํ์ ๋ฌธ์ ๋์ ๊ธฐ๊ณ๋
ํด ๋ฐ์ดํฐ
|
| 109 |
+
- AIhub 17 ๋ด์ค ๊ธฐ์ฌ ๊ธฐ๊ณ๋
ํด ๋ฐ์ดํฐ
|
| 110 |
+
- AIhub 21 ๋์์๋ฃ ๊ธฐ๊ณ๋
ํด
|
| 111 |
+
- AIhub 149 ํ ์ ๋ณด ์ง์์๋ต ๋ฐ์ดํฐ
|
| 112 |
+
- AIhub 150 ์ซ์์ฐ์ฐ ๊ธฐ๊ณ๋
ํด ๋ฐ์ดํฐ
|
| 113 |
+
- AIhub 151 ๊ธ์ต, ๋ฒ๋ฅ ๋ฌธ์ ๊ธฐ๊ณ๋
ํด ๋ฐ์ดํฐ
|
| 114 |
+
- kyujinpy/KoCoT_2000
|
| 115 |
+
- MarkrAI/KoCommercial-Dataset
|
| 116 |
+
- CarrotAI/ko-instruction-dataset
|
| 117 |
+
- heegyu/CoT-collection-ko
|
| 118 |
+
-
|
| 119 |
## Contact us
|
| 120 |
- ์จ๋ก๋ง์ธ๋
|
| 121 |
- ์์ธ์ ๊ด์
๊ตฌ ๋จ๋ถ์ํ๋ก 1802, 2์ธต
|