Readme Signature Declaration

[BrianSemiglia]

I’ve been trying to understand DSPy via the documentation but I’m still a little muddy. I thought it could be useful if the Readme took inspiration from DSPy’s own design by stating the signature of itself as a framework. This is what I gather so far:


The parts of a formula that can vary can be considered a detail. Given a problem, a solution can be achieved in different ways:
Problem -> {Solver} -> Solution

The variability allows for scalability (up and down) and optimization:

Problem -> {Fast Solver} -> Solution
Problem -> {Quality Solver} -> Solution

Declarative formulas ask for the parts that don’t vary and infer the rest.
Problem -> Solution

Prompting is typically done using a formula of:
System Prompt, Query, Example, LLM -> Answer

DSPy is a framework that makes the system prompt a variable that is inferred from examples.
{System Prompt}, Query, Example, Metric, Solver (Module), LLM, Prompt Optimizer -> Answer

Its signature shows the potential for more variability where the ideal llm, optimization and approach to solving the problem are inferred from the query, example and metric:
{System Prompt}, Query, Example, Metric, {Solver (Module)}, {LLM}, {Prompt Optimizer} -> Answer


Am I close?