[Bug] assert_transform_module not present in dspy 2.6.0

[RahulBalan]

What happened?

We were using dspy assertion assert_transform_module. But when we updated dspy to 2.6.0, the assertions.py is fully commented.
what is the new location?

Steps to reproduce

from dspy.primitives.assertions import assert_transform_module, backtrack_handler
ImportError: cannot import name 'assert_transform_module' from 'dspy.primitives.assertions' (/Users/rvbalan/Disk/python/cpqcodeassist/cpqcodeassist/.venv/lib/python3.12/site-packages/dspy/primitives/assertions.py)

DSPy version

2.6.0

[bongiovimatthew]

We are also seeing this issue, which is caused by the fact that the assertions code has been fully removed in this PR, without replacement. There are a number of questions about this on the discord server.

@okhat, is there a plan or roadmap for when we can expect assertions v2? Would it be possible to put assertions v1 back in place while v2 is finalized? Our team is now caught between versions, as there are things in 2.6.0 that we would like to leverage, but we require assertions as well.

[imJarrad]

Same here. Bit stuck until assertions v2. Keen to hear when that might be.

[okhat]

While we prepare a short tutorial on this: Folks looking to migrate from 2.5-style Assertions, you can now use dspy.BestOfN or dspy.Refine which replace the assertions functionality with streamlined modules instead.

module = dspy.ChainOfThought(...) # or a complex multi-step dspy.Module
module = dspy.BestOfN(module, N=5, reward_fn=reward_fn, threshold=1.0)

module(...) # at most 5 retries, picking best reward, but stopping if `threshold` is reached

where, reward functions can return scalar values like float or bool, e.g.

def reward_fn(input_kwargs, prediction):
    return len(prediction.field1) == len(prediction.field1)

[christeefy]

Thanks for the update @okhat!

With dspy.BestOfN and dspy.Refine, is there a way to provide (dynamic) feedback on failures similar to dspy.Assert(<predicate>, <feedback>)?

[okhat]

Coming likely this week! It will be via return value of the metric.

[christeefy]

@okhat, just following up on this — I don’t see it implemented in main. Is there an open ticket or PR I can track for updates on this issue? Thanks!