Can algorithmically optimizing prompts outperform prompt engineering?
Resources:
DSPy Breakdown: https://github.com/ALucek/dspy-breakdown/blob/main/dspy_breakdown.ipynb
DSPy Documentation: https://dspy.ai/
DSPy Repo: https://github.com/stanfordnlp/dspy
Chapters:
00:00 - What is DSPy?
02:51 - Setup
04:12 - Signatures Overview
10:41 - Modules Overview
12:01 - Modules: Chain of Thought
13:04 - Modules: Program of Thought
14:53 - Modules: ReAct
17:47 - Modules: Others
19:22 - Why We Need Metrics
20:49 - Metrics: Example Data Type & Overview
23:07 - Metrics: Simple
24:46 - Metrics: Intermediate
27:21 - Metrics: Advanced
30:01 - Optimizers Overview
33:20 - Optimizer: Labeled Few Shot
34:45 - Optimizer: Bootstrap Few Shot
36:48 - Optimizer: Bootstrap Few Shot With Random Search
38:49 - Optimizer: KNN Fewshot
40:07 - Optimizer: COPRO
43:01 - Optimizer: MIPROv2
45:28 - Fine Tuning Based Optimizers
46:10 - Optimizer: Bootstrap Fine Tune
49:06 - Choosing Optimizers
50:05 - Optimizing Optimizers
51:17 - Is DSPy Worth it?
#ai #DSPy #datascience