source ./api/bin/activate pip install -r requirements.txt gcloud auth application-default login python3 api/init.py
cd nextapp npm run dev
curl http://localhost:5000/embeddings \
-H "Content-Type: application/json" \
-d '{
"input": "Your text string goes here",
"model": "text-embedding-ada-002"
}'
- Text embeddings measure how related text strings are to each other. They are commonly used for recommendation, clustering, and classification.
- Embeddings are simple vectors of floating point numbers, with distances between two points indicating their relatedness
Price:
- Model Usage Ada v2 $0.0004 / 1K tokens
The guide https://platform.openai.com/docs/guides/embeddings/use-cases has a bunch of good use cases. Notably, we will try following this guide for learning how to compare text
https://github.com/openai/openai-cookbook/blob/main/text_comparison_examples.md
- Can't really use embeddings because that would be for review to review. Like if you're looking at one review or location and you want to compare to others similar, it makes sense... Otherwise?
- Don't store any maps data, pass onto the model to create the recommendation
- Store the embeddings in a vector engine
- Split text into chunks smaller than token limit
- Embed each chunk
- Store embeddings in vector engine
- OpenAI recommends fine-tuning versus embedding for text classification. When we're going to do tags, maybe we look into that later
- Do not store any google maps data, feed it into the embedding as part of the string input.
- Store the embeddings in a vector engine
- User input for search is used as input to the embedding model, then retrive distance from existing embeddings within the search area
- Some sort of clustering for places.
- docker-compose
- db setup
- Vercel deployment
- Hook up front and back-end