Aggregate function to help concatenate prompts

[simonw]

I keep getting errors because I send prompts that were too large for the model.

With GPT-3 I believe it's possible to count tokens in advance. I want an aggregate function to help me assembly a prompt by truncating each text as necessary to fit.

[simonw]

I'm going to ship an alpha with this now so I can deploy and test it a bit before writing the docs.

[simonw]

Example query using the new function:

with texts as (
  select body from blog_entry order by id desc limit 5 
)
select json_object('pre', openai_build_prompt(openai_strip_tags(body), 'Context:
------------
', '
------------
Given the above context, answer the following question: ' || :question, 30, 1000)) from texts

[simonw]

Documentation: https://github.com/simonw/datasette-openai/tree/0.2#openai_build_prompttext-prefix-suffix-completion_tokens-token_limit4000