search

stanfordnlp / dspy

DSPy: The framework for programming—not prompting—foundation models
https://dspy-docs.vercel.app/
MIT License
18.38k stars 1.41k forks source link

Roadmap: Terminology Improvements #1415

Open lebsral opened 2 months ago

lebsral commented 2 months ago

I think one of the great improvements that DSPy has made was going from the term "Teleprompter" to "Optimizer."

That change made what DSPy was doing much clearer. That terminology change made DSPy more accessible to developers coming from different programming traditions and languages.

In that spirit, let's take a hard look at changing some other terms in DSPy.

To get the conversation started let's talk about clean, self-explanatory, easy new names for:

(The not clear at all set, for sure we can do better) COPRO MIPRO forward predict OPTUNA

(The maybe unclear set, maybe we can do better) signature ReAct

okhat commented 2 months ago

Hey, thanks @lebsral ! This is indeed a good time to put together improved naming.

lebsral commented 2 months ago

Personally, I love inside jokes, puns, and clever names. And it will pains me to see some of that going away. For a broader audience, we must get boring.

This is what ChatGPT suggests...

MIPRO (Focused on Minibatch and Prompt Optimization) BatchPromptOptimizer PromptBatcher MiniBatchPrompter IterativePromptTuner BatchwisePromptGen PromptCycleOptimizer PromptMinibatcher BatchTunedPrompter PromptSeriesOptimizer StepwisePromptGen COPRO (Generalized Comprehensive Prompt Optimization) ComprehensivePromptOptimizer PromptSynthesisEngine HolisticPromptRefiner CompletePromptTuner FullCyclePrompter IntegratedPromptOptimizer SignaturePromptCompiler PromptFullOptimizer BroadPromptSynthesis UnifiedPromptOptimizer

I don't like any of those. We might have to use human brains.