sour path
When fine tuning a model using GPT, if I provide prompts with general information about the system I'm trying to model, will GPT remember the information shared when processing new prompts?
For example, if I did a prompt like:
prompt: "with 'A', 'B', 'C', 'D' as valid fields, I want the average of Alpha and Beta and the sum of Charly and Delta"
completion: "(A+B)/2 and C+D)"
Will the system be able to remember that A,B,C,D are valid fields for subsequent prompts?
random wind
When fine tuning a model using GPT, if I provide prompts with general informatio...
Fine-tuning isn't great at remembering information, it tends to hallucinate despite it and it's not even meant for teaching it new information but rather new tasks.
sour path
What would be the best way to go about doing that in terms of subject specific matter then in order to leverage the language capabilities?
random wind
What would be the best way to go about doing that in terms of subject specific m...
Embeddings if you want Q/A and tasks that require semantic search
sour path
We're attempting to generate SQL-like queries based on prompts, would embeddings be useful for that?