#nice - if it helps any, here's my prompt

1 messages · Page 1 of 1 (latest)

short grove
#

Ok I think I need to fully switch to this approach. Currently I'm passing the module source for the module I want to generate examples for and it's too confusing to the llm to try to understand what that module would look like in the dagger API. It keeps trying to use the struct from the module

#

I'm looking around in your module... do you end up sending through Module.Objects() somewhere?

frosty nexus
#

it learns from the GraphQL schema instead, which is quite a bit different

short grove
#

ah the Introspection query

frosty nexus
#

yea

#

I guess you could do that too - call Module.serve and then introspect?

#

but, a warning, that introspection is expensive token-wise haha. It might be better to come up with a more succinct format for LLM

#

the caching I added for Anthropic should help I guess

short grove
#

so far this demo passes like 100x more context than any of the other demos and gemini seems to handle that really well compared to the models I was using on ollama. Probably some tuning I could do on ollama to handle it better, but still impressed with gemini

frosty nexus
#

nice, been wanting to try it out