NyanNyanovich / nyan

Automatic news aggregator in Telegram / Автоматический агрегатор новостей в Телеграме
https://t.me/nyannews
Apache License 2.0
198 stars 30 forks source link

8 hours summary using GPT not present #15

Open b0tm1nd opened 1 year ago

b0tm1nd commented 1 year ago

After setting up, everything is working except 8 hours summary. I see files summary.py, openai.py, topics.py are present but don't seem to be included anywhere. Maybe some bash scripts missing? Also I've read that for openai you need to put your keys into .env which might also need to be mentioned in the setup instruction.

NyanNyanovich commented 1 year ago

It is not a part of core functions.

I have this command in cron, with three different launch times:

python3 -m nyan.topics --mongo-config-path configs/mongo_config.json --client-config-path configs/client_config.json --renderer-config-path configs/renderer_config.json --prompt-path nyan/prompts/topics.txt --model-name gpt-4 --duration-hours 8 --auto --template-path nyan/templates/topics.html

And OPENAI_API_KEY in env variables.

b0tm1nd commented 1 year ago

Thanks. What else functions might need to be configured with cron? Probably, archiving? In the whitepaper it says that there is a summary around 9pm every day, I guess that relates to summary.py but you're not currently using it in your channel?

NyanNyanovich commented 1 year ago

Nothing. Archiving is manual using this script. Nope, it was about ratings.py, but having both topics and ratings seems too much for me.

b0tm1nd commented 1 year ago

How often do you run archive?

NyanNyanovich commented 1 year ago

Rarely. You can find all runs here: https://github.com/NyanNyanovich/nyan/releases

sellerbto commented 9 months ago

Добрый день! Интересный проект, последнее время думал сделать для себя агрегатор, чтобы посмотреть основные новости один раз за день и не чекать постоянно каналы.

Есть вопрос: А почему бы не запустить какую-нибудь доступную модель локально, чтобы делать выжимку?

NyanNyanovich commented 9 months ago

Нет таких моделей, по крайней мере сравнимых с gpt-4. Я сохраняю все выжимки, когда-нибудь на них можно будет дообучить модель. Для preview версии это выходит примерно в 50 сентов в день всё равно.