MiproV2 prompt and task model are not using the global setting

[chenmoneygithub]

According to the comment if users are not specifying prompt_model and task_model, the globally configured LM will be used. However, it seems not being the case while executing the code:

import dspy
from dspy.datasets.gsm8k import GSM8K, gsm8k_metric

# Set up the LM.
lm = dspy.OpenAI(model="gpt-4o-mini", max_tokens=250)

dspy.settings.configure(lm=lm)

# Load math questions from the GSM8K dataset.
gsm8k = GSM8K()
gsm8k_trainset, gsm8k_devset = gsm8k.train[:10], gsm8k.dev[:10]

class CoT(dspy.Module):
    def __init__(self):
        super().__init__()
        self.prog = dspy.ChainOfThought("question -> answer")

    def forward(self, question):
        return self.prog(question=question)

from dspy.teleprompt import MIPROv2

# Set up the optimizer: we want to "bootstrap" (i.e., self-generate) 4-shot examples of our CoT program.
config = dict(max_bootstrapped_demos=4, max_labeled_demos=4)

# Optimize! Use the `gsm8k_metric` here. In general, the metric is going to tell the optimizer how well it's doing.
optimizer = MIPROv2(metric=gsm8k_metric, num_candidates=3)
optimized_cot = optimizer.compile(CoT(), trainset=gsm8k_trainset)

It throws error:

Do you wish to continue? (y/n): y
Traceback (most recent call last):
  File "/Users/chen.qian/Documents/mlflow_team/dspy/tmp.py", line 32, in <module>
    optimized_cot = optimizer.compile(CoT(), trainset=gsm8k_trainset)
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/chen.qian/Documents/mlflow_team/dspy/dspy/teleprompt/mipro_optimizer_v2.py", line 222, in compile
    proposer = GroundedProposer(
               ^^^^^^^^^^^^^^^^^
  File "/Users/chen.qian/Documents/mlflow_team/dspy/dspy/propose/grounded_proposer.py", line 260, in __init__
    self.data_summary = create_dataset_summary(
                        ^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/chen.qian/Documents/mlflow_team/dspy/dspy/propose/dataset_summary_generator.py", line 51, in create_dataset_summary
    observation = dspy.Predict(DatasetDescriptor, n=1, temperature=1.0)(examples=order_input_keys_in_string(trainset[0:upper_lim].__repr__()))
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/chen.qian/Documents/mlflow_team/dspy/dspy/predict/predict.py", line 91, in __call__
    return self.forward(**kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/chen.qian/Documents/mlflow_team/dspy/dspy/predict/predict.py", line 104, in forward
    assert lm is not None, "No LM is loaded."
           ^^^^^^^^^^^^^^
AssertionError: No LM is loaded.

Checking out the code the self.prompt_model assignment is not reading from the global setting by default, is that a bug or I am missing something? @omkar-sh

[okhat]

cc @XenonMolecule