Settings

Theme

LlamaAcademy: Teach GPTs to understand API documentation with LoRA

github.com

104 points by danicgross 3 years ago · 14 comments

Reader

joshka 3 years ago

It seems weird to me that you'd want a model per API to be trained rather than a model that understands many APIs. There's likely more similarities than differences in that the differences are really your domain model which infects the nouns of the but less so the verbs and structure.

  • haxton 3 years ago

    I played with building out a graphql mesh [0] of a few different APIs as I was curious to see if I could build one schema (and subsets of it) and have GPT interface over that. Turns out, it did a pretty good job if you can provide it the right portions of the schema it needs.

    It also helped out when I was struggling to reconcile with how large of JSON payloads I was getting. The REST endpoints are just killing the prompt size, but having the model choose the fields it needed from GraphQL really helped out there.

    Put it down for a while until I can get access to the plugin fine-tuned version of chatgpt and see if there's still a need or if it is additive still.

    [0] https://github.com/Urigo/graphql-mesh/

  • toxicFork 3 years ago

    Yeah we just need to keep something trained for the latest APIs

    Or at least latest popular x versions of APIs just in case

  • ttymck 3 years ago

    Won't structure also be correlated with domain model?

frozencell 3 years ago

How to make beautiful schemas like those (what tool)

(1) https://github.com/danielgross/LlamaAcademy/raw/main/assets/...

  • riceart 3 years ago

    With literally just about any diagramming tool - Vizio, dia, xfig, PowerPoint, etc. Frankly those are not particularly consistent or careful - they’re not that good. The text positioning and alignment is inconsistent and careless, bleeding into the boxes in places, the arrows don’t even meet the boxes in a consistent way. I’m not trying to shit on it (I’ve done my fair share of mediocre diagrams) - but it’s not professional level or particularly careful work (it’s a jpeg to boot) and looks done by hand (without even bothering to use alignment affordances.)

  • c_s_guy 3 years ago
nextworddev 3 years ago

Most of the heavy lifting is done by GPT4 so a bit weird to call it LlamaAcademy

  • hex4def6 3 years ago

    Given that GPT is a trademark of OpenAI, I'd be careful about naming any product or service with "GPT" in the name, even if it depends on it.

rasengan 3 years ago

Cool but I would be careful also as I’m pretty sure this is in breach of the OpenAI TOS.

Keyboard Shortcuts

j
Next item
k
Previous item
o / Enter
Open selected item
?
Show this help
Esc
Close modal / clear selection