fix(MIPROv2): zero shot not taking .compile parameters into account before determining if the program was zero shot

[isaacbmiller]

The zeroshot_opt option, which determines if the proposer should use demos or not, was being set before applying the compile time arguments passed into the .compile function

Fixes #8508

[isaacbmiller]

@codex review

[TomeHirata]

We should avoid updating instance attributes as much as possible. Can we update methods that rely on these attributes (e.g., _bootstrap_fewshot_examples) to receive them as arguments?

[isaacbmiller]

Needs testing DNM

[TomeHirata]

LGTM

[TomeHirata]

@isaacbmiller feel free to merge once you run MIPRO with this code to confirm no regression!

[isaacbmiller]

Verified via:

import dspy

lm = dspy.LM("openai/gpt-5-nano", temperature=1.0, max_tokens=16000)

program = dspy.Predict("number_as_string: str -> answer_from_rule: float")
trainset = [
    dspy.Example(number_as_string="123", answer_from_rule=124).with_inputs("number_as_string"),
    dspy.Example(number_as_string="456", answer_from_rule=457).with_inputs("number_as_string"),
    dspy.Example(number_as_string="789", answer_from_rule=790).with_inputs("number_as_string"),
    dspy.Example(number_as_string="101112", answer_from_rule=101113).with_inputs("number_as_string"),
    dspy.Example(number_as_string="131415", answer_from_rule=131416).with_inputs("number_as_string"),
    dspy.Example(number_as_string="161718", answer_from_rule=161719).with_inputs("number_as_string"),
    dspy.Example(number_as_string="192021", answer_from_rule=192022).with_inputs("number_as_string"),
]


def metric_dspy(e, p, t=None):
    return 1 if p.answer_from_rule == e.answer_from_rule else 0

optimizer = dspy.MIPROv2(
    metric_dspy, 
    max_bootstrapped_demos = 1, 
    max_labeled_demos = 1, 
    prompt_model=lm, 
    task_model=lm,
    auto = "medium",
    num_threads=6)

program_optimized_no_demos = optimizer.compile(program, trainset=trainset, requires_permission_to_run = False, max_bootstrapped_demos=0, max_labeled_demos=0)
program_optimized_no_demos.save("program_optimized_no_demos.json")
program_optimized_with_demos = optimizer.compile(program, trainset=trainset, requires_permission_to_run = False)
program_optimized_with_demos.save("program_optimized_with_demos.json")