Jason Antic
Jason Antic
@bitplane That all sounds fantastic. Yeah as far as performance goes- I've had to do a lot of optimization of our more advanced video stuff in our commercial work with...
> Another approach could be to run history through an embeddings API, save the embeddings to a Vector DB, then do a lookup for relevant _memories_ on each step. I...
I seem to be getting much better json generation behavior after I change this in prompt.txt: From this: > You should only respond in JSON format as described below. To...
I do also have this in there but it didn't seem to make a difference: > Review your own json output, step by step.
> Yep, this has majorly improved the consistency of my replies. I was getting the failures before extremely frequently, I haven't seen any so far since including this. Thanks for...
> I had to reduce the prompt to 2000 words and told it to accept user input. I've been wondering if the core problem is simply asking the model to...
> Also, we could also use regex to retrieve the info from the JSON, since often what is missing in the answer is often minor: eg some columns, or comas,...
> Oops sorry I forgot the re-downloading check. Hold fire, I'll deal with that ASAP. There hasn't been any activity after this. Want to go ahead with this or do...
Thanks for digging into this and reporting it so thoroughly!