[Bug] train.py num_rollout==0 error

Bug Description

I have done the demo training: example/search-r1/, but i want to do an eval only test.
I found the train.py:

# special case for eval-only
    if args.num_rollout == 0 and args.eval_interval is not None:
        ray.get(rollout_manager.eval.remote(rollout_id=0))

so i set the num_rollout 0, and also set the eval args, but errors happenned.

Anyone who has the similar experience, or how to do a eval only test.

Steps to Reproduce

here is my run args:

ROLLOUT_ARGS=(
   --prompt-data /mnt/workspace/data/al_training/slime/data/train.parquet
   --input-key prompt
   --label-key reward_model
   --apply-chat-template
   --rollout-shuffle 
   --num-rollout 0
   --rollout-batch-size 32
   --n-samples-per-prompt 8
   --rollout-max-response-len 1024
   --rollout-temperature 1
   
   --eval-interval 25
   --eval-prompt-data nq_hotpotqa /mnt/workspace/data/al_training/slime/data/test.parquet
   --eval-input-key prompt
   --eval-label-key reward_model
   --n-samples-per-eval-prompt 1
   
   --global-batch-size 256
   --balance-data
)

Expected Behavior

Do a eval only when i set the right parameters.

Actual Behavior

Error

Environment

i use the slime:test docker image, 20250324 pull from docker source.

Logs

`
Traceback (most recent call last):
  File "/root/slime/train.py", line 106, in <module>
    train(args)
  File "/root/slime/train.py", line 20, in train
    actor_model, critic_model = create_training_models(args, pgs, rollout_manager)
                                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/slime/slime/ray/placement_group.py", line 150, in create_training_models
    start_rollout_ids = ray.get(
                        ^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/ray/_private/auto_init_hook.py", line 22, in auto_init_wrapper
    return fn(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/ray/_private/client_mode_hook.py", line 104, in wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/ray/_private/worker.py", line 2981, in get
    values, debugger_breakpoint = worker.get_objects(
                                  ^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/ray/_private/worker.py", line 1012, in get_objects
    raise value.as_instanceof_cause()
ray.exceptions.RayTaskError(AssertionError): ray::MegatronTrainRayActor.init() (pid=910684, ip=172.17.0.2, actor_id=a39a1f56d660f292ad0b1a8402000000, repr=<slime.backends.megatron_utils.actor.MegatronTrainRayActor object at 0x7f2c1ff54680>)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/slime/slime/utils/timer.py", line 97, in wrapper
    return fn(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^
  File "/root/slime/slime/backends/megatron_utils/actor.py", line 91, in init
    (self.model, self.optimizer, self.opt_param_scheduler, loaded_rollout_id) = initialize_model_and_optimizer(
                                                                                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/slime/slime/backends/megatron_utils/model.py", line 773, in initialize_model_and_optimizer
    model, optimizer, opt_param_scheduler = setup_model_and_optimizer(args, role)
                                            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/slime/slime/backends/megatron_utils/model.py", line 125, in setup_model_and_optimizer
    opt_param_scheduler = get_optimizer_param_scheduler(args, optimizer)
                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/slime/slime/backends/megatron_utils/model.py", line 63, in get_optimizer_param_scheduler
    opt_param_scheduler = OptimizerParamScheduler(
                          ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/root/Megatron-LM/megatron/core/optimizer_param_scheduler.py", line 73, in __init__
    assert self.lr_decay_steps > 0
           ^^^^^^^^^^^^^^^^^^^^^^^
AssertionError
`

Additional Context

No response

Pre-submission Checklist

• I have read the CONTRIBUTING.md and understand the collaboration scope.
• I have read the documentation and my issue is not addressed there.
• I have searched for existing issues and this is not a duplicate.
• I have provided a minimal, reproducible example.

Start with the eval-only branch in train.py, then trace the initialization path through slime/ray/placement_group.py and slime/backends/megatron_utils/model.py. Reproduce the failure with the provided arguments and inspect get_optimizer_param_scheduler alongside Megatron-LM's lr_decay_steps assertion. Done means num-rollout 0 can start evaluation without the reported assertion.