←
DataTalksClub2,702 posts
DataTalksClub avatar

DataTalksClub

@DataTalksClub

Free online community for data builders. Learn data, ML, MLOps, LLMs, RAG, and AI engineering. Join free hands-on courses, events, workshops, and podcasts.

◷ Joined October 2020▧ 981 Photos and videos
DataTalksClub avatar
DataTalksClub@DataTalksClub·25 Jan 2023

Want to talk about data? Data science, career questions, practical machine learning, and more? Join @DataTalksClub! The link is in the profile description. Learn more about our activities in the thread! 🧵

Post attachment
DataTalksClub avatar
⟳ DataTalksClub retweeted
DataTalksClub@DataTalksClub·Nov 17

A data team can spend its week answering one SQL request at a time. Madhu Gurramkonda joins the DataTalks.Club podcast to talk about moving from that queue to AI agents that business teams can actually use.

Post attachment
DataTalksClub avatar
DataTalksClub@DataTalksClub·4h

With an open weight model, you download the weights from Hugging Face and keep them in your own storage. If the license allows it, you host the model yourself, and nobody can pull the plug. 3/4

DataTalksClub avatar
DataTalksClub@DataTalksClub·4h

Alexey Grigorev learned this on his own project. He used DALL-E 2 and 3 for illustrations in a pipeline that generates horror stories for kids. Then the model was discontinued. He had to redo the pipeline. The alternatives were not as good and were more expensive. 2/4

DataTalksClub avatar
DataTalksClub@DataTalksClub·4h

If your pipeline calls a closed model through an API, the provider decides how long that model stays available. 1/4

1:56
DataTalksClub avatar
⟳ DataTalksClub retweeted
DataTalksClub@DataTalksClub·Nov 13

If you added generation to a product this month, what would you start with: images, video, audio, or a mix? Vote in the poll 👇

DataTalksClub avatar
⟳ DataTalksClub retweeted
DataTalksClub@DataTalksClub·Nov 17

Two examples run through the Production AI Agents for Generative Media course: - The Santa Christmas Challenge: a personalized gift card with a Santa voice message - A product catalog: an ad with a voice-over or a short video Here's how they fit together 👇

Post attachment
DataTalksClub avatar
DataTalksClub@DataTalksClub·12h

- Write a specification for the video - Cut it into small pieces - Pick an efficient model for each part - Generate the parts and stick them together - Keep the style consistent, so it looks like one video

DataTalksClub avatar
DataTalksClub@DataTalksClub·12h

How do you generate a video cost-efficiently? In our pre-course Q&A for Production AI Agents for Generative Media, Mikhail Rozhkov from Nebius outlined the approach.

1:37
DataTalksClub avatar
DataTalksClub@DataTalksClub·Nov 17

The conversation covers the foundations that come first, one governed path to data, and a validator that checks answers before anyone sees them.

DataTalksClub avatar
DataTalksClub@DataTalksClub·Nov 17

Once one record works, you scale the Factory to 1,000+ records with a cost line per asset. That's Homework 3, the Factory Scale Challenge: 1,000 records on a public leaderboard. 2/3