Type error of mm_processor_cache_gb

[unlabeledData]

I try to reproduce the CHORD method with Qwen2.5-1.5B-Instruct : ms-swift/examples/train/grpo/internal/chord.sh

I found that when vllm==0.10.2., a model config args error will occor during the loading of vllm engine:

pydantic_core._pydantic_core.ValidationError: 1 validation error for ModelConfig
[rank1]: mm_processor_cache_gb
[rank1]: Input should be a valid number [type=float_type, input_value=None, input_type=NoneType]
[rank1]: For further information visit https://errors.pydantic.dev/2.11/v/float_type

mm_processor_cache_gb=None is not allowed.

[hjh0119]

Can you provide the complete error stack and script?

btw, you can set --vllm_mm_processor_cache_gb 4 to bypass this error.

[unlabeledData]

--vllm_mm_processor_cache_gb 4 works.
Thank you for your help.

[singing4you]

遇到了同样的问题
ASCEND_RT_VISIBLE_DEVICES=7 swift rollout --model /data/models/Qwen2_5-0_5B-Instruct --model_type qwen2_5 --use_async_engine true --tensor_parallel_size 1 --max_turns 3 --vllm_enforce_eager True --multi_turn_scheduler thinking_tips_scheduler

报错

WARNING 10-27 15:36:46 [registry.py:458] Model architecture Qwen3MoeForCausalLM is already registered, and will be overwritten by the new model class vllm_ascend.models.qwen3_moe:CustomQwen3MoeForCausalLM.
WARNING 10-27 15:36:46 [registry.py:458] Model architecture Qwen3ForCausalLM is already registered, and will be overwritten by the new model class vllm_ascend.models.qwen3:CustomQwen3ForCausalLM.
Process Process-1:
Traceback (most recent call last):
  File "/home/ma-user/anaconda3/envs/swift/lib/python3.10/multiprocessing/process.py", line 314, in _bootstrap
    self.run()
  File "/home/ma-user/anaconda3/envs/swift/lib/python3.10/multiprocessing/process.py", line 108, in run
    self._target(*self._args, **self._kwargs)
  File "/data/lab/work_dir/ms-swift-3.9.2/swift/llm/infer/rollout.py", line 158, in llm_worker_entry
    asyncio.run(async_llm_worker(*args, **kwargs))
  File "/home/ma-user/anaconda3/envs/swift/lib/python3.10/asyncio/runners.py", line 44, in run
    return loop.run_until_complete(main)
  File "/home/ma-user/anaconda3/envs/swift/lib/python3.10/asyncio/base_events.py", line 649, in run_until_complete
    return future.result()
  File "/data/lab/work_dir/ms-swift-3.9.2/swift/llm/infer/rollout.py", line 124, in async_llm_worker
    engine = SwiftRolloutDeploy.get_infer_engine(args, template=args.get_template(None))
  File "/data/lab/work_dir/ms-swift-3.9.2/swift/llm/infer/rollout.py", line 242, in get_infer_engine
    return GRPOVllmEngine(**kwargs)
  File "/data/lab/work_dir/ms-swift-3.9.2/swift/llm/infer/infer_engine/grpo_vllm_engine.py", line 60, in __init__
    super().__init__(
  File "/data/lab/work_dir/ms-swift-3.9.2/swift/llm/infer/infer_engine/vllm_engine.py", line 140, in __init__
    self._prepare_engine()
  File "/data/lab/work_dir/ms-swift-3.9.2/swift/llm/infer/infer_engine/vllm_engine.py", line 150, in _prepare_engine
    engine = llm_engine_cls.from_engine_args(self.engine_args)
  File "/home/ma-user/anaconda3/envs/vllm/vllm/engine/async_llm_engine.py", line 572, in from_engine_args
    vllm_config = engine_args.create_engine_config(usage_context)
  File "/home/ma-user/anaconda3/envs/vllm/vllm/engine/arg_utils.py", line 1057, in create_engine_config
    model_config = self.create_model_config()
  File "/home/ma-user/anaconda3/envs/vllm/vllm/engine/arg_utils.py", line 904, in create_model_config
    return ModelConfig(
  File "/home/ma-user/anaconda3/envs/swift/lib/python3.10/site-packages/pydantic/_internal/_dataclasses.py", line 123, in __init__
    s.__pydantic_validator__.validate_python(ArgsKwargs(args, kwargs), self_instance=s)
pydantic_core._pydantic_core.ValidationError: 1 validation error for ModelConfig
mm_processor_cache_gb
  Input should be a valid integer [type=int_type, input_value=None, input_type=NoneType]
    For further information visit https://errors.pydantic.dev/2.11/v/int_type
ERROR:    Traceback (most recent call last):
  File "/home/ma-user/anaconda3/envs/swift/lib/python3.10/site-packages/starlette/routing.py", line 692, in lifespan
    async with self.lifespan_context(app) as maybe_state:
  File "/home/ma-user/anaconda3/envs/swift/lib/python3.10/contextlib.py", line 199, in __aenter__
    return await anext(self.gen)
  File "/data/lab/work_dir/ms-swift-3.9.2/swift/llm/infer/rollout.py", line 204, in lifespan
    msg = connection.recv()
  File "/home/ma-user/anaconda3/envs/swift/lib/python3.10/multiprocessing/connection.py", line 250, in recv
    buf = self._recv_bytes()
  File "/home/ma-user/anaconda3/envs/swift/lib/python3.10/multiprocessing/connection.py", line 414, in _recv_bytes
    buf = self._recv(4)
  File "/home/ma-user/anaconda3/envs/swift/lib/python3.10/multiprocessing/connection.py", line 383, in _recv
    raise EOFError
EOFError

ERROR:    Application startup failed. Exiting.

传入 --vllm_mm_processor_cache_gb 0 确实解决了

[unlabeledData]

I also want to know how to evaluate the training performance of the chord method?
I use Qwen2.5-1.5B-Instruct as the model, lm_eval as the evaluation tool, task is AIME.
I found that before and after training, the lm eval results on AIME are both 0.

Could someone give me some suggestions?