Project

General

Profile

Feature #3642

Task #3653: Launch with first chapter support - Phase 1 - Practice for Decent Learners (Top 5 Themes, No DetailedQNAs, No Detailed Answers)

Changes to call_openai_api to handle context limitations

Added by Ram Kordale 5 months ago. Updated 3 months ago.

Status:
Closed
Priority:
Normal
Assignee:
Start date:
07/01/2024
Due date:
% Done:

100%

Estimated time:

Description

Add an additional argument num_tokens. Add an additional output num_tokens.

Clients interested in handling the context limitation must send num_tokens received in the previous call in the thread as argument. First time, the value is 0.

As always, get assistant, instructions, prompt_prefix and prompt_suffix based on model and index. In addition, also get periodic_prompt from DS.

So far, prompt used is concat(prompt_prefix, <prompt>, prompt_suffix). If num_tokens is not zero, continue this. Otherwise prompt should be concat(periodic_prompt, prompt_prefix, <prompt>, prompt_suffix).

One more change: add response['usage']['total_tokens'] (tokens used in this api call) to num_tokens and return it.

Also available in: Atom PDF