[rank0]: ValueError: Model architectures ['Qwen2ForCausalLM'] failed to be inspected. Please check the logs for more details.

[magnificent1208]

环境 3090*4卡
执行pip install -r requirment.txt
没出现什么问题

请问是否 不支持3090？

[dhcode-cpp]

支持3090。 请提供更多错误信息

[magnificent1208]

[INFO|image_processing_auto.py:300] 2025-02-13 15:12:53,819 >> Could not locate the image processor configuration file, will try to use the model config instead.
ERROR 02-13 15:12:54 registry.py:299] Error in inspecting model architecture 'Qwen2ForCausalLM'
ERROR 02-13 15:12:54 registry.py:299] Traceback (most recent call last):
ERROR 02-13 15:12:54 registry.py:299]   File "/home/liu/.local/lib/python3.11/site-packages/vllm/model_executor/models/registry.py", line 495, in _run_in_subprocess
ERROR 02-13 15:12:54 registry.py:299]     returned.check_returncode()
ERROR 02-13 15:12:54 registry.py:299]   File "/home/liu/anaconda3/lib/python3.11/subprocess.py", line 502, in check_returncode
ERROR 02-13 15:12:54 registry.py:299]     raise CalledProcessError(self.returncode, self.args, self.stdout,
ERROR 02-13 15:12:54 registry.py:299] subprocess.CalledProcessError: Command '['/home/liu/anaconda3/bin/python', '-m', 'vllm.model_executor.models.registry']' returned non-zero exit status 1.
ERROR 02-13 15:12:54 registry.py:299] 
ERROR 02-13 15:12:54 registry.py:299] The above exception was the direct cause of the following exception:
ERROR 02-13 15:12:54 registry.py:299] 
ERROR 02-13 15:12:54 registry.py:299] Traceback (most recent call last):
ERROR 02-13 15:12:54 registry.py:299]   File "/home/liu/.local/lib/python3.11/site-packages/vllm/model_executor/models/registry.py", line 297, in _try_inspect_model_cls
ERROR 02-13 15:12:54 registry.py:299]     return model.inspect_model_cls()
ERROR 02-13 15:12:54 registry.py:299]            ^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 02-13 15:12:54 registry.py:299]   File "/home/liu/.local/lib/python3.11/site-packages/vllm/model_executor/models/registry.py", line 268, in inspect_model_cls
ERROR 02-13 15:12:54 registry.py:299]     return _run_in_subprocess(
ERROR 02-13 15:12:54 registry.py:299]            ^^^^^^^^^^^^^^^^^^^
ERROR 02-13 15:12:54 registry.py:299]   File "/home/liu/.local/lib/python3.11/site-packages/vllm/model_executor/models/registry.py", line 498, in _run_in_subprocess
ERROR 02-13 15:12:54 registry.py:299]     raise RuntimeError(f"Error raised in subprocess:\n"
ERROR 02-13 15:12:54 registry.py:299] RuntimeError: Error raised in subprocess:
ERROR 02-13 15:12:54 registry.py:299] Error: mkl-service + Intel(R) MKL: MKL_THREADING_LAYER=INTEL is incompatible with libgomp-a34b3233.so.1 library.
ERROR 02-13 15:12:54 registry.py:299] 	Try to import numpy first or set the threading layer accordingly. Set MKL_SERVICE_FORCE_INTEL to force it.
ERROR 02-13 15:12:54 registry.py:299] 
[rank0]: Traceback (most recent call last):
[rank0]:   File "/home/liu/work/X-R1/src/x_r1/grpo.py", line 182, in <module>
[rank0]:     main(script_args, training_args, model_args)
[rank0]:   File "/home/liu/work/X-R1/src/x_r1/grpo.py", line 134, in main
[rank0]:     trainer = XGRPOTrainer(
[rank0]:               ^^^^^^^^^^^^^
[rank0]:   File "/home/liu/.local/lib/python3.11/site-packages/trl/trainer/grpo_trainer.py", line 369, in __init__
[rank0]:     self.llm = LLM(
[rank0]:                ^^^^
[rank0]:   File "/home/liu/.local/lib/python3.11/site-packages/vllm/utils.py", line 1039, in inner
[rank0]:     return fn(*args, **kwargs)
[rank0]:            ^^^^^^^^^^^^^^^^^^^
[rank0]:   File "/home/liu/.local/lib/python3.11/site-packages/vllm/entrypoints/llm.py", line 240, in __init__
[rank0]:     self.llm_engine = self.engine_class.from_engine_args(
[rank0]:                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]:   File "/home/liu/.local/lib/python3.11/site-packages/vllm/engine/llm_engine.py", line 479, in from_engine_args
[rank0]:     engine_config = engine_args.create_engine_config(usage_context)
[rank0]:                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]:   File "/home/liu/.local/lib/python3.11/site-packages/vllm/engine/arg_utils.py", line 1059, in create_engine_config
[rank0]:     model_config = self.create_model_config()
[rank0]:                    ^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]:   File "/home/liu/.local/lib/python3.11/site-packages/vllm/engine/arg_utils.py", line 983, in create_model_config
[rank0]:     return ModelConfig(
[rank0]:            ^^^^^^^^^^^^
[rank0]:   File "/home/liu/.local/lib/python3.11/site-packages/vllm/config.py", line 348, in __init__
[rank0]:     self.multimodal_config = self._init_multimodal_config(
[rank0]:                              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]:   File "/home/liu/.local/lib/python3.11/site-packages/vllm/config.py", line 408, in _init_multimodal_config
[rank0]:     if ModelRegistry.is_multimodal_model(architectures):
[rank0]:        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]:   File "/home/liu/.local/lib/python3.11/site-packages/vllm/model_executor/models/registry.py", line 433, in is_multimodal_model
[rank0]:     model_cls, _ = self.inspect_model_cls(architectures)
[rank0]:                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]:   File "/home/liu/.local/lib/python3.11/site-packages/vllm/model_executor/models/registry.py", line 393, in inspect_model_cls
[rank0]:     return self._raise_for_unsupported(architectures)
[rank0]:            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[rank0]:   File "/home/liu/.local/lib/python3.11/site-packages/vllm/model_executor/models/registry.py", line 350, in _raise_for_unsupported
[rank0]:     raise ValueError(
[rank0]: ValueError: Model architectures ['Qwen2ForCausalLM'] failed to be inspected. Please check the logs for more details.
[rank0]:[W213 15:12:55.655552884 ProcessGroupNCCL.cpp:1250] Warning: WARNING: process group has NOT been destroyed before we destruct ProcessGroupNCCL. On normal program exit, the application should call destroy_process_group to ensure that any pending NCCL operations have finished in this process. In rare cases this process can exit before this point and block the progress of another member of the process group. This constraint has always been present,  but this warning has only been added since PyTorch 2.4 (function operator())
W0213 15:12:56.503000 329696 site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 329799 closing signal SIGTERM
W0213 15:12:56.504000 329696 site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 329800 closing signal SIGTERM
W0213 15:12:56.504000 329696 site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 329801 closing signal SIGTERM
E0213 15:12:56.834000 329696 site-packages/torch/distributed/elastic/multiprocessing/api.py:869] failed (exitcode: 1) local_rank: 0 (pid: 329798) of binary: /home/liu/anaconda3/bin/python
Traceback (most recent call last):
  File "/home/liu/.local/bin/accelerate", line 8, in <module>
    sys.exit(main())
             ^^^^^^
  File "/home/liu/.local/lib/python3.11/site-packages/accelerate/commands/accelerate_cli.py", line 48, in main
    args.func(args)
  File "/home/liu/.local/lib/python3.11/site-packages/accelerate/commands/launch.py", line 1157, in launch_command
    deepspeed_launcher(args)
  File "/home/liu/.local/lib/python3.11/site-packages/accelerate/commands/launch.py", line 845, in deepspeed_launcher
    distrib_run.run(args)
  File "/home/liu/anaconda3/lib/python3.11/site-packages/torch/distributed/run.py", line 910, in run
    elastic_launch(
  File "/home/liu/anaconda3/lib/python3.11/site-packages/torch/distributed/launcher/api.py", line 138, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/liu/anaconda3/lib/python3.11/site-packages/torch/distributed/launcher/api.py", line 269, in launch_agent
    raise ChildFailedError(
torch.distributed.elastic.multiprocessing.errors.ChildFailedError: 
============================================================
src/x_r1/grpo.py FAILED
------------------------------------------------------------
Failures:
  <NO_OTHER_FAILURES>
------------------------------------------------------------
Root Cause (first observed failure):
[0]:
  time      : 2025-02-13_15:12:56
  host      : ubuntu-SYS-4029GP-TRT
  rank      : 0 (local_rank: 0)
  exitcode  : 1 (pid: 329798)
  error_file: <N/A>
  traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
============================================================

[magnificent1208]

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2024 NVIDIA Corporation
Built on Tue_Feb_27_16:19:38_PST_2024
Cuda compilation tools, release 12.4, V12.4.99
Build cuda_12.4.r12.4/compiler.33961263_0

Thu Feb 13 15:16:35 2025       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.144.03             Driver Version: 550.144.03     CUDA Version: 12.4     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |

[magnificent1208]

### yaml config

# Model arguments
model_name_or_path: /nfsdata2/base_model/Qwen2.5-1.5B-Instruct
model_revision: main
torch_dtype: bfloat16

# Data training arguments
dataset_name: AI-MO/NuminaMath-TIR
dataset_configs:
- all
# Num processes is less by 1 as vLLM is using 1 GPU
num_processes: 7

# GRPO trainer config

bf16: true
use_vllm: true
vllm_device: auto
vllm_gpu_memory_utilization: 0.7
do_eval: false
eval_strategy: "no"
eval_steps: 10
gradient_accumulation_steps: 4
gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
hub_strategy: every_save
learning_rate: 3.0e-06
log_level: info
logging_steps: 10
logging_strategy: steps
lr_scheduler_type: cosine
max_prompt_length: 256
num_generations: 8
max_completion_length: 1024
max_steps: -1
num_train_epochs: 3
output_dir: output/X-R1-1.5B
overwrite_output_dir: true
per_device_eval_batch_size: 1
per_device_train_batch_size: 2
push_to_hub: False
report_to:
- wandb
save_strategy: "epoch"
seed: 42

[magnificent1208]

# run

export CUDA_HOME=/usr/local/cuda 
CUDA_VISIBLE_DEVICES=1,2,3,4
ACCELERATE_LOG_LEVEL=info \
accelerate launch \
--config_file recipes/zero3.yaml \
--num_processes=4 \
src/x_r1/grpo.py \
--config recipes/X_R1_zero_1dot5B_config.yaml \
> ./output/test.log 2>&1

[magnificent1208]

支持3090。 请提供更多错误信息

标题所述问题，我通过添加了这行命令，解决了

export MKL_THREADING_LAYER=GNU

以下内容是搜索提供的解释：

根本原因：mkl - service 库的 MKL_THREADING_LAYER=INTEL 设置与 libgomp - a34b3233.so.1 库不兼容。这使得在 vllm.model_executor.models.registry.py 中尝试通过子进程检查 Qwen2ForCausalLM 模型架构时，子进程返回非零退出状态，最终导致 ValueError，阻止了 XGRPOTrainer 的正确初始化。

解决方法：
调整 MKL_THREADING_LAYER 设置：尝试将 MKL_THREADING_LAYER 设置为与 libgomp 兼容的值，如 GNU。在运行脚本前，通过终端设置环境变量：

[dhcode-cpp]

已解决，按照readme安装环境。