Model Information
- Weather Forecast-Based Personalized Recommendation System with Gemma Model.
- Language: Korean
Description
- This project aims to fine-tune the Gemma2 2B model using weather data from the Korea Meteorological Administration (KMA). By utilizing the KMA Short-term Forecast API, we will fetch real-time weather data, which will be used to provide personalized recommendations for clothing, activities, and weather-related precautions. For example, on a rainy day, the system will notify users to bring an umbrella, and on cold days, it will suggest wearing appropriate warm clothing. The model will be trained using historical weather data from Korea to ensure localized and accurate recommendations. Once the model is fine-tuned, it will be deployed on Hugging Face, making it accessible via an API.
Base Model
- google/gemma-2-2b-it
- https://huggingface.co/google/gemma-2-2b-it
Datasets
- OBS_AWS_TIM(KMA)2021~2023
- Temperature, precipitation, wind, humidity, pressure, etc.
- https://data.kma.go.kr/data/grnd/selectAwsRltmList.do?pgmNo=56
- target: season
length of a season?
What is the length of a season?
- Seasons are fundamental climate phenomena that greatly impact the way humans, animals, and plants live. The start date and length of each season are crucial for making timely plans in agriculture, as well as in various social and economic activities.
- The Korea Meteorological Administration (KMA) provides past and future season start dates and duration information based on the natural season classification criteria, which uses daily average temperatures.
[Criteria for Determining the Start of Each Season]
- Spring: The first day when the daily average temperature rises above 5°C and doesn’t fall below this threshold again.
- Summer: The first day when the daily average temperature rises above 20°C and stays above it.
- Autumn: The first day when the daily average temperature drops below 20°C and doesn’t rise back above it.
- Winter: The first day when the daily average temperature falls below 5°C and doesn’t rise above this level again.
- [Changes in the Length of Seasons in Korea]
- Over the past 109 years (1912–2020), the average data from six key locations in Korea show that summer has become longer and winter shorter, while spring and autumn have seen little change.
- Compared to the previous 30 years, in the recent 30 years, the start of spring and summer has advanced by 17 days and 11 days, respectively. Meanwhile, the start of autumn and winter has been delayed by 9 days and 5 days, respectively.
- As a result, summer has lengthened by 20 days, and winter has shortened by 22 days.
- In the last 30 years, summer has lasted for about 118 days, making it the longest season, spanning roughly four months.
Fine-tuning a Gemma model
[1. Access Gemma]
- 1-1. Navigate to the gemma-2b repository that you wish to use.
- 1-2. Click on "Acknowledge license" in the "Access Gemma on Hugging Face" section.
- 1-3. Click on "Authorize," check the required fields, and then click "Accept."
- 1-4. If you see the message "You have been granted access to this model" [Gated model], this means you now have access to the Gemma model.
[2. Access Tokens]
- 2-1. Click on your profile in the upper-right corner and select "Settings."
- 2-2. In the menu, click on "Access Tokens," then click on [+ Create new token].
- 2-3. Under [Token type], choose ['Read'] from the options: 'Fine-grained', 'Read', 'Write'.
- 2-4. Enter a name for the token, then click "Create token."
- 2-5. Save the generated Access Token, and use it for Hugging Face authentication, such as in notebook_login.
from huggingface_hub import notebook_login
notebook_login()
[3. Gemma-2-2b-it(instruction-tuned)]
[4. Data Preprocessing]
[5. Fine-Tuning]
[6. Prompt Engineering]