분석시각화 대회 코드 공유 게시물은
내용 확인 후
좋아요(투표) 가능합니다.
새로운 데이터셋 만들기 Based on Mistral-7B
개인적으로 캐글을 참가하면서 인상 깊은 자료가 있었습니다.
해당 자료는 다음과 같습니다.
https://www.kaggle.com/code/defdet/mistral-essay-generation
해당 노트북은 `mistralai/Mistral-7B-v0.1`를 풀로드하여 특정 데이터를 만드는 코드입니다.
토픽이나 템퍼러쳐에 대한 부분은 다음 코드에서 이용하실 수 있습니다.
```
TOPICS = ["Car-free cities", "Does the electoral college work?", "Exploring Venus", "The Face on Mars", "Facial action coding system", "Seeking multiple opinions", "Phones and driving"] top_k = 6
```
위에서 top_k의 인덱스가 좀 과하게 설정되었는데, 2부터 시작해서 점차 늘려가시기 바랍니다.
인스트럭션에 대한 부분은 다음과 같습니다.
```
<s>[INST] Write an essay based on the topic provided as if you were a student. Your essay needs to be unique and convincing and not very long. Output nothing but the essay. Topic: {TOPICS[2]}[/INST]
```
해당 노트북 하단에 위치해 있으며, 해당 INST 밑 부분을 수정하시면, 해당 내용으로 프롬프트가 진행됩니다.
해당 구조로 In-Context Learning 같은 기법을 이용하실 수도 있을겁니다.
DACON Co.,Ltd | CEO Kookjin Kim | 699-81-01021
Mail-order-sales Registration Number: 2021-서울영등포-1704
Business Providing Employment Information Number: J1204020250004
#901, Eunhaeng-ro 3, Yeongdeungpo-gu, Seoul 07237
E-mail dacon@dacon.io |
Tel. 070-4102-0545
Copyright ⓒ DACON Inc. All rights reserved