[Bug] Errors in Optimizer while it seems fully specified according to the docs.

[MaximeRivest]

What happened?

I tought, I specified all lm to used and still some missing.

Why is this not sufficient?

import dspy

lm = dspy.LM("groq/meta-llama/llama-4-scout-17b-16e-instruct")
teacherp = dprogram_optimized.deepcopy()
teacherp.set_lm(lm = dspy.LM("gemini/gemini-2.5-pro"))
dprogram_optimized.set_lm(lm = lm)
optimizer = dspy.MIPROv2(
    metric_dspy, 
    max_bootstrapped_demos = 3, 
    max_labeled_demos = 3, 
    prompt_model=dspy.LM("gemini/gemini-2.5-pro"), 
    task_model=lm,
    auto = "medium",
    num_threads=6)
dprogram_optimized_v2 = optimizer.compile(dprogram_optimized, trainset=trainset, requires_permission_to_run = False, teacher = teacherp)

==> STEP 1: BOOTSTRAP FEWSHOT EXAMPLES <==
2025/08/16 11:13:05 INFO dspy.teleprompt.mipro_optimizer_v2: These will be used as few-shot example candidates for our program and for creating instructions.

2025/08/16 11:13:05 INFO dspy.teleprompt.mipro_optimizer_v2: Bootstrapping N=12 sets of demonstrations...
Bootstrapping set 1/12
Bootstrapping set 2/12
Bootstrapping set 3/12
  0%|        | 0/3 [00:00<?, ?it/s]2025/08/16 11:13:06 
ERROR dspy.teleprompt.bootstrap: Failed to run or to evaluate example Example({'receipt_image': Image(url=data:image/jpeg;base64,<IMAGE_BASE_64_ENCODED(1357136)>), 'receipt_totals': ReceiptTotals(before_tax_total=20.99, after_tax_total=24.13)}) (input_keys={'receipt_image'}) with <function metric_dspy at 0x783f7400e5c0> due to No LM is loaded. Please configure the LM using `dspy.configure(lm=dspy.LM(...))`. e.g, `dspy.configure(lm=dspy.LM('openai/gpt-4o-mini'))`.

I have trust issues with dspy.configure(lm=..). I want to make sure I know which model is used when.

I'm not sure if it's a bug or a design decision?

Steps to reproduce

import dspy

lm = dspy.LM("groq/meta-llama/llama-4-scout-17b-16e-instruct")
teacherp = dprogram_optimized.deepcopy()
teacherp.set_lm(lm = dspy.LM("gemini/gemini-2.5-pro"))
dprogram_optimized.set_lm(lm = lm)
optimizer = dspy.MIPROv2(
    metric_dspy, 
    max_bootstrapped_demos = 3, 
    max_labeled_demos = 3, 
    prompt_model=dspy.LM("gemini/gemini-2.5-pro"), 
    task_model=lm,
    auto = "medium",
    num_threads=6)
dprogram_optimized_v2 = optimizer.compile(dprogram_optimized, trainset=trainset, requires_permission_to_run = False, teacher = teacherp)

DSPy version

3.0.0

[okhat]

This means that MIPRO requires a global LM to be set via dspy.configure. I'm not familiar with the details of where that may be happening.