Allow customizing num_init_trials in MBM benchmark method

ax/benchmark/methods/modular_botorch.py

@@ -14,7 +14,6 @@
 from ax.modelbridge.generation_strategy import GenerationStep, GenerationStrategy
 from ax.modelbridge.registry import Models
 from ax.models.torch.botorch_modular.model import SurrogateSpec
-
 from ax.service.scheduler import SchedulerOptions
 from botorch.acquisition.acquisition import AcquisitionFunction
 from botorch.acquisition.analytic import LogExpectedImprovement
@@ -59,13 +58,15 @@ def get_sobol_botorch_modular_acquisition(
     acqf_name = acqf_name_abbreviations.get(
         acquisition_cls.__name__, acquisition_cls.__name__
     )
-    name = f"MBM::{model_name}_{acqf_name}"
+    name = name or f"MBM::{model_name}_{acqf_name}"
 
     generation_strategy = GenerationStrategy(
         name=name,
         steps=[
             GenerationStep(
-                model=Models.SOBOL, num_trials=num_sobol_trials, min_trials_observed=5
+                model=Models.SOBOL,
+                num_trials=num_sobol_trials,
+                min_trials_observed=num_sobol_trials,
             ),
             GenerationStep(
                 model=Models.BOTORCH_MODULAR,

ax/benchmark/tests/methods/test_methods.py

@@ -48,9 +48,13 @@ def test_benchmark_replication_runs(self) -> None:
             model_cls=SingleTaskGP,
             scheduler_options=get_sequential_optimization_scheduler_options(),
             acquisition_cls=LogExpectedImprovement,
+            num_sobol_trials=2,
+            name="test",
             distribute_replications=False,
         )
         n_sobol_trials = method.generation_strategy._steps[0].num_trials
+        self.assertEqual(n_sobol_trials, 2)
+        self.assertEqual(method.name, "test")
         # Only run one non-Sobol trial
         problem = get_problem(problem_name="ackley4", num_trials=n_sobol_trials + 1)
         result = benchmark_replication(problem=problem, method=method, seed=0)