search

monarch-initiative / ontogpt

LLM-based ontological extraction tools, including SPIRES
https://monarch-initiative.github.io/ontogpt/
BSD 3-Clause "New" or "Revised" License
603 stars 75 forks source link

Relax truncation with `pubmed-annotate` #281

Open caufieldjh opened 11 months ago

caufieldjh commented 11 months ago

The pubmed-annotate command currently truncates inputs to ~3K characters. That's not necessary with the larger context models, so it can be relaxed and/or added as a CLI parameter.

caufieldjh commented 11 months ago

The max_tokens parameter passed to the OpenAI API just defaults to 3000, though that value is updated for some functions. So in general, the default will work for all available models but should be updated on a per-model basis.