Goals of the week:
This weeks primary goal was to complete the OpenAI interface so that it could be packaged and wrapped in ECL for use by HPCC systems.
What was Accomplished:
- Learned and implemented few shot learning into my chat completions.
- Added a continue conversation function.
- Cleaned up the code through the deletion of unnecessary code and re formatting how a call is made so that the code is more seperated in function between the two existing classes.
What is few-shot learning?
Few-shot learning is the technique in primp writing in which you give the model example prompt completion pairs before asking your request. The model will then use those example pairs to better understand how to answer your request.
Here is an example of a prompt that utilizes the concept of few-shot learning.
What is a continued conversation?
When using OpenAI’s online chatbot, chatgpt, the model automatically keeps track of previous messages and responses so that in a sense it can recall previous information given. However when making a call to OpenAI’s API subsequent requests will have no knowledge of the previous request. To solve this problem I directly took the previous response and appended it to the beginning of my new request so that the information from the previous instance would carry over to the new call.
Immediate Goals moving forward:
The most important goal at the current moment is to develop a community day presentation on my project and its current and potential applications for HPCC systems.
- Start to look into possible data sets to train a potential Graph Neural Network
- Look into guides on applying tensor flow as well.