Firebase After Hours #3 - Genkit: More than Meets the AI!
HTML-код
- Опубликовано: 16 июн 2024
- Hang out with Nohe and Peter to learn more about Firebase Genkit, Google's open source AI integration framework. With special guest Pavel Jbanov, lead engineer on the Genkit team.
Hosts:
Alexander Nohe
Peter Friese
Special guest: Pavel Jbanov
Catch the latest from Firebase → goo.gle/firebase-latest
Watch more Firebase After Hours → goo.gle/FirebaseAfterHours
Subscribe to Firebase → goo.gle/Firebase
#FirebaseAfterHours 
Наука
Thanks everyone for joining this livestream. We've compiled a list of the questions that were asked, including the answers we gave in this discussions thread: github.com/firebase/genkit/discussions/424 - if you'd like to ask more questions, feel free to use the Genkit discussions forums (github.com/firebase/genkit/discussions) - the team is monitoring these on a daily basis.
Very explanatory video and a great product, congrats!!
My personal notes:
19:24 Stream to return the data partially, interesting for long lists for examples.
21:00 A good example with schema
38:00 Scroll to see the upper part of the code (I don't know if that's available somewhere)
40:30 As cloud function (so you can call from your API)
43:43 Open Telemetry, great tool to know
50:00 He uses a pdf as context to the next prompt
1:16:00 Enforce app check
1:19:35 Setup Genkit in an already existent Express project
Thanks a lot! ✌️ Very interesting presentation. Are the examples presented available on a repository somewhere?
github.com/FirebaseExtended/compass-travel-planning-sample
When performing the RAG what is the recommended limit to retrieve documents from firebase?
It's generally good to start somewhere around 3-5, but there's no universal rule, really depends on your application -- your chunk size, how the original document is structured, how dense or spread our relevant information typically is in your documents.
I created a searched based on the RAG technique. It's weird that the same results I pass to the LLM I don't get any results which the error is:
FAILED_PRECONDITION: Generation resulted in no candidates matching provided output schema.
Then I run the same query and results appear
guys you should at least check the mics and volumen levels before, the guy in yellow is talking under a mountain
Yeah, that is my bad. I unfortunately had a leaf blower outside my window as soon as we started recording so I turned down my mic and was unable to recalibrate it once the stream started.
Gotta say that this, as a tutorial, is a huge stinker of a tutorial. I know y'all are excited about these features but this did nothing to help me use genkit as a javascript dev. Trying to go through the quickstart -- which forces me to install a typscript file! -- is colossally confusing.
- no mention of how to handle the different environment deployments (nextjs, cloud, firebase)
- no mention of how to install this into a typical node app with a "hello world" function (instead, y'all are spinning up banana restaurants)