【模型支持问题】关于不同qwen 30BA3B模型支持的问题

1. 请问slime/scripts/models/qwen3-30B-A3B.sh 脚本所支持的模型是这个吗：https://huggingface.co/Qwen/Qwen3-30B-A3B 。我在将这个模型转换成mcore后运行，发现报错，可能的原因是什么呢：
   Traceback (most recent call last):
   File "/home/jovyan/work/lyl/code/CodeReview/slime/train.py", line 106, in
   train(args)
   File "/home/jovyan/work/lyl/code/CodeReview/slime/train.py", line 27, in train
   actor_model, critic_model = create_training_models(args, pgs, rollout_manager)
   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/home/jovyan/work/lyl/code/CodeReview/slime/slime/ray/placement_group.py", line 143, 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 2972, in get
   values, debugger_breakpoint = worker.get_objects(
   ^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/ray/_private/worker.py", line 1031, in get_objects
   raise value.as_instanceof_cause()
   ray.exceptions.RayTaskError(CheckpointException): ray::MegatronTrainRayActor.init() (pid=3035762, ip=10.72.1.39, actor_id=f1edc61a1301fa5f41971cc702000000, repr=<slime.backends.megatron_utils.actor.MegatronTrainRayActor object at 0x763a5c898440>)
   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/home/jovyan/work/lyl/code/CodeReview/slime/slime/utils/context_utils.py", line 9, in wrapper
   return fn(*args, **kwargs)
   ^^^^^^^^^^^^^^^^^^^
   File "/home/jovyan/work/lyl/code/CodeReview/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 "/home/jovyan/work/lyl/code/CodeReview/slime/slime/backends/megatron_utils/model.py", line 726, in initialize_model_and_optimizer
   iteration, _ = load_checkpoint(
   ^^^^^^^^^^^^^^^^
   File "/home/jovyan/work/lyl/code/CodeReview/slime/slime/backends/megatron_utils/checkpoint.py", line 28, in load_checkpoint
   return _load_checkpoint_megatron(
   ^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/root/Megatron-LM/megatron/training/checkpointing.py", line 1544, in load_checkpoint
   state_dict, checkpoint_name, release, ckpt_type = _load_base_checkpoint(
   ^^^^^^^^^^^^^^^^^^^^^^
   File "/root/Megatron-LM/megatron/training/checkpointing.py", line 1090, in _load_base_checkpoint
   return _load_global_dist_base_checkpoint(
   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/root/Megatron-LM/megatron/training/checkpointing.py", line 984, in _load_global_dist_base_checkpoint
   state_dict = dist_checkpointing.load(sharded_state_dict, checkpoint_name, load_strategy, strict=args.dist_ckpt_strictness)
   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/root/Megatron-LM/megatron/core/dist_checkpointing/serialization.py", line 161, in load
   loaded_state_dict = sharded_strategy.load(sharded_state_dict, checkpoint_dir)
   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/root/Megatron-LM/megatron/core/dist_checkpointing/strategies/torch.py", line 925, in load
   checkpoint.load_state_dict(
   File "/usr/local/lib/python3.12/dist-packages/typing_extensions.py", line 3004, in wrapper
   return arg(*args, **kwargs)
   ^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/distributed/checkpoint/state_dict_loader.py", line 48, in load_state_dict
   return _load_state_dict(
   ^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/distributed/checkpoint/state_dict_loader.py", line 279, in _load_state_dict
   central_plan = distW.reduce_scatter("plan", local_step, global_step)
   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/distributed/checkpoint/utils.py", line 219, in reduce_scatter
   raise result
   torch.distributed.checkpoint.api.CheckpointException: CheckpointException ranks:dict_keys([0, 1, 2, 3, 4, 5, 6, 7])
   ray::MegatronTrainRayActor.init() (pid=3035762, ip=10.72.1.39, actor_id=f1edc61a1301fa5f41971cc702000000, repr=<slime.backends.megatron_utils.actor.MegatronTrainRayActor object at 0x763a5c898440>)
   File "/usr/local/lib/python3.12/dist-packages/torch/distributed/checkpoint/utils.py", line 192, in reduce_scatter
   local_data = map_fun()
   ^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/distributed/checkpoint/logger.py", line 87, in wrapper
   result = func(*args, **kwargs)
   ^^^^^^^^^^^^^^^^^^^^^
   File "/usr/local/lib/python3.12/dist-packages/torch/distributed/checkpoint/state_dict_loader.py", line 266, in local_step
   local_plan = planner.create_local_plan()
   ^^^^^^^^^^^^^^^^^^^^^^^^^^^
   File "/root/Megatron-LM/megatron/core/dist_checkpointing/strategies/torch.py", line 616, in create_local_plan
   self._validate_global_shapes(self.metadata, self.shapes_validation_sharded_tensors)
   File "/root/Megatron-LM/megatron/core/dist_checkpointing/strategies/torch.py", line 569, in _validate_global_shapes
   raise KeyError(
   KeyError: "decoder.layers.0.self_attention.linear_proj.weight from model not in state dict: ['decoder.final_layernorm._extra_state/shard_0_1', 'decoder.final_layernorm.weight', 'decoder.layers.mlp.experts.experts.linear_fc1._extra_state/shard_0.0_48.128', 'decoder.layers.mlp.experts.experts.linear_fc1._extra_state/shard_0.100_48.128', 'decoder.layers.mlp.experts.experts.linear_fc1._extra_state/shard_0.101_48.128', 'decoder.layers.mlp.experts.experts.linear_fc1._extra_state/shard_0.102_48.128', 'decoder.layers.mlp.experts.experts.linear_fc1._extra_state/shard

2. 如果希望跑 qwen 3 30BA3B-Instruct (https://huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507) ，应该怎么设置模型参数文件呢？是可以复用的吗？

3. --loss-mask-type qwen3 这个参数是否有用？对不同的qwen3系列模型（instruct，thinking系列等），应该怎么设置这个参数？

Start with slime/scripts/models/qwen3-30B-A3B.sh and trace the reported failure through train.py, slime/ray/placement_group.py, slime/backends/megatron_utils/actor.py, model.py, and checkpoint.py. Compare the model's checkpoint keys with the expected keys, then verify how the script and --loss-mask-type qwen3 apply to the listed Qwen3 variants. Done means establishing supported models and documenting the required parameter settings or incompatibility.