mandoo-9b-2407-sft / README.md
heegyu's picture
Update README.md
1f6ad67 verified
metadata
license: gemma
datasets:
  - Magpie-Align/Magpie-Pro-MT-300K-v0.1
  - Magpie-Align/Magpie-Qwen2-Pro-300K-Filtered
  - iknow-lab/qarv-instruct-ko-mt-deduped
  - jojo0217/korean_safe_conversation
  - heegyu/HRC
  - heegyu/orca-math-korean-preference-cleaned
  - iknow-lab/ko-evol-writing-wiki
  - CarrotAI/ko-instruction-dataset
  - maywell/kiqu_samples
  - HAERAE-HUB/K2-Feedback
language:
  - ko
  - en
  - zh

Mandoo is a LM assistant supporting English, Chinese and Korean.

Example

from transformers import pipeline

pipe = pipeline("text-generation", model="heegyu/mandoo-9b-2407", device_map="auto", torch_dtype="auto")

messages = [
    {"role": "user", "content": "I want to start saving some money by growing my own food. Can I do this during the winter with an indoor garden?"},
]
pipe(messages, max_new_tokens=128, do_sample=True)

Benchmark Result

Every generation of this model was sampled with temperature=0.7, top_p=0.9, top_k=50

Korean

Model ์‹ฑ๊ธ€ํ„ด
gemma-2-9b-it 7.45
mandoo-9b-2407-sft 6.50

I used sampling with temperature=0.7, max_new_tokens=2048 for generation.

# mandoo-9b-2407-sft
์นดํ…Œ๊ณ ๋ฆฌ: ์ถ”๋ก (Reasoning), ์‹ฑ๊ธ€ ์ ์ˆ˜ ํ‰๊ท : 6.86, ๋ฉ€ํ‹ฐ ์ ์ˆ˜ ํ‰๊ท : 3.86
์นดํ…Œ๊ณ ๋ฆฌ: ์ˆ˜ํ•™(Math), ์‹ฑ๊ธ€ ์ ์ˆ˜ ํ‰๊ท : 5.14, ๋ฉ€ํ‹ฐ ์ ์ˆ˜ ํ‰๊ท : 3.71
์นดํ…Œ๊ณ ๋ฆฌ: ๊ธ€์“ฐ๊ธฐ(Writing), ์‹ฑ๊ธ€ ์ ์ˆ˜ ํ‰๊ท : 7.29, ๋ฉ€ํ‹ฐ ์ ์ˆ˜ ํ‰๊ท : 7.00
์นดํ…Œ๊ณ ๋ฆฌ: ์ฝ”๋”ฉ(Coding), ์‹ฑ๊ธ€ ์ ์ˆ˜ ํ‰๊ท : 8.29, ๋ฉ€ํ‹ฐ ์ ์ˆ˜ ํ‰๊ท : 8.14
์นดํ…Œ๊ณ ๋ฆฌ: ์ดํ•ด(Understanding), ์‹ฑ๊ธ€ ์ ์ˆ˜ ํ‰๊ท : 9.29, ๋ฉ€ํ‹ฐ ์ ์ˆ˜ ํ‰๊ท : 8.57
์นดํ…Œ๊ณ ๋ฆฌ: ๋ฌธ๋ฒ•(Grammar), ์‹ฑ๊ธ€ ์ ์ˆ˜ ํ‰๊ท : 6.43, ๋ฉ€ํ‹ฐ ์ ์ˆ˜ ํ‰๊ท : 3.43
์ „์ฒด ์‹ฑ๊ธ€ ์ ์ˆ˜ ํ‰๊ท : 7.21
์ „์ฒด ๋ฉ€ํ‹ฐ ์ ์ˆ˜ ํ‰๊ท : 5.79
์ „์ฒด ์ ์ˆ˜: 6.50

# gemma-2-9b-it
์นดํ…Œ๊ณ ๋ฆฌ: ์ถ”๋ก (Reasoning), ์‹ฑ๊ธ€ ์ ์ˆ˜ ํ‰๊ท : 9.43, ๋ฉ€ํ‹ฐ ์ ์ˆ˜ ํ‰๊ท : 6.71
์นดํ…Œ๊ณ ๋ฆฌ: ์ˆ˜ํ•™(Math), ์‹ฑ๊ธ€ ์ ์ˆ˜ ํ‰๊ท : 6.14, ๋ฉ€ํ‹ฐ ์ ์ˆ˜ ํ‰๊ท : 8.57
์นดํ…Œ๊ณ ๋ฆฌ: ๊ธ€์“ฐ๊ธฐ(Writing), ์‹ฑ๊ธ€ ์ ์ˆ˜ ํ‰๊ท : 8.71, ๋ฉ€ํ‹ฐ ์ ์ˆ˜ ํ‰๊ท : 8.86
์นดํ…Œ๊ณ ๋ฆฌ: ์ฝ”๋”ฉ(Coding), ์‹ฑ๊ธ€ ์ ์ˆ˜ ํ‰๊ท : 7.43, ๋ฉ€ํ‹ฐ ์ ์ˆ˜ ํ‰๊ท : 6.86
์นดํ…Œ๊ณ ๋ฆฌ: ์ดํ•ด(Understanding), ์‹ฑ๊ธ€ ์ ์ˆ˜ ํ‰๊ท : 8.29, ๋ฉ€ํ‹ฐ ์ ์ˆ˜ ํ‰๊ท : 8.29
์นดํ…Œ๊ณ ๋ฆฌ: ๋ฌธ๋ฒ•(Grammar), ์‹ฑ๊ธ€ ์ ์ˆ˜ ํ‰๊ท : 6.29, ๋ฉ€ํ‹ฐ ์ ์ˆ˜ ํ‰๊ท : 3.86
์ „์ฒด ์‹ฑ๊ธ€ ์ ์ˆ˜ ํ‰๊ท : 7.71
์ „์ฒด ๋ฉ€ํ‹ฐ ์ ์ˆ˜ ํ‰๊ท : 7.19
์ „์ฒด ์ ์ˆ˜: 7.45

English

AlpacaEval

                            length_controlled_winrate  win_rate  standard_error  n_total  avg_length
gpt-4o-2024-05-13                               57.46     51.33            1.47      805        1873
gpt-4-turbo-2024-04-09                          55.02     46.12            1.47      805        1802
gpt4_1106_preview                               50.00     50.00            0.00      805        2049
claude-3-opus-20240229                          40.51     29.11            1.39      805        1388
claude-3-sonnet-20240229                        34.87     25.56            1.34      805        1420
Meta-Llama-3-70B-Instruct                       34.42     33.18            1.39      805        1919
gemini-pro                                      24.38     18.18            1.16      805        1456
Mixtral-8x7B-Instruct-v0.1                      23.69     18.26            1.19      805        1465
Meta-Llama-3-8B-Instruct                        22.92     22.57            1.26      805        1899
**heegyu/mandoo-9b-2407-sft**  <---             19.82     18.18            1.13      805        1847
Mistral-7B-Instruct-v0.2                        17.11     14.72            1.08      805        1676
alpaca-7b                                        5.88      2.59            0.49      805         396

IFEval

Model ์‹ฑ๊ธ€ํ„ด
gemma-2-9b-it 76.95
mandoo-9b-2407-sft 59.19
Strict Accuracy Scores: Avg 0.59191279139
prompt-level: 0.5471349353049908                  
instruction-level: 0.6366906474820144 

Loose Accuracy Scores:
prompt-level: 0.589648798521257                   
instruction-level: 0.6774580335731415