Importing the DSPy library is very time consuming, what is this due to?Is it possible to optimize this problem?

[17Reset]

import time

start_time = time.time()
import dspy
print(f"load dspy time: {time.time() - start_time} s")

load dspy time: 5.986269474029541 s

[estsauver]

The long import time is mostly due to DSPy’s design choice to expose many features at the top level. The init.py file imports numerous subpackages and re‑exports their symbols so users can simply write import dspy and immediately access predictors, retrievers, adapters, and evaluation utilities. These subpackages import heavy dependencies—such as litellm, regex, numpy, and pandas—as soon as they’re loaded. This eager loading ensures everything is ready to use but means that importing DSPy pulls in a large dependency graph, making startup slower than a lightweight library.

[chenmoneygithub]

@17Reset Thanks for reporting the issue! We are aware of this, and in an effort of shrinking the memory/time consumption at importing time. In our incoming DSPy 3.0 release, we will ensure this runs much faster.

[17Reset]

Thanks for clearing up the confusion, and look for a better performing DSPy to arrive soon!

[TomeHirata]

Yes, we'll reduce dependencies and packages imported with import dspy in DSPy 3.0.