[Bug]: KeyError: 'language_model.layers.0.self_attn.qkv_proj.weight' #19149
Description
Your current environment
My Environment
torch 2.6.0
torchaudio 2.6.0
torchvision 0.21.0
transformers 4.52.4
transformers-stream-generator 0.0.5
vllm 0.8.5
🐛 Describe the bug
I use ms-swift to GRPO Qwen2.5-VL-3B model.
The error message show as follow:
Traceback (most recent call last):
File "/mdata/llh/Projects/ms-swift/swift/cli/rlhf.py", line 5, in <module>
rlhf_main()
File "/mdata/llh/Projects/ms-swift/swift/llm/train/rlhf.py", line 169, in rlhf_main
return SwiftRLHF(args).main()
^^^^^^^^^^^^^^^^^^^^^^
File "/mdata/llh/Projects/ms-swift/swift/llm/base.py", line 49, in main
result = self.run()
^^^^^^^^^^
File "/mdata/llh/Projects/ms-swift/swift/llm/train/sft.py", line 123, in run
return self.train(trainer)
^^^^^^^^^^^^^^^^^^^
File "/mdata/llh/Projects/ms-swift/swift/llm/train/sft.py", line 184, in train
trainer.train(trainer.args.resume_from_checkpoint)
File "/mdata/llh/Projects/ms-swift/swift/trainers/mixin.py", line 369, in train
res = super().train(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/mdata/llh/miniconda3/envs/swift/lib/python3.11/site-packages/transformers/trainer.py", line 2240, in train
return inner_training_loop(
^^^^^^^^^^^^^^^^^^^^
File "/mdata/llh/miniconda3/envs/swift/lib/python3.11/site-packages/transformers/trainer.py", line 2555, in _inner_training_loop
tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/mdata/llh/Projects/ms-swift/swift/trainers/rlhf_trainer/grpo_trainer.py", line 1330, in training_step
return super().training_step(model, inputs, num_items_in_batch)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/mdata/llh/miniconda3/envs/swift/lib/python3.11/site-packages/transformers/trainer.py", line 3739, in training_step
inputs = self._prepare_inputs(inputs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/mdata/llh/miniconda3/envs/swift/lib/python3.11/site-packages/trl/extras/profiling.py", line 96, in wrapper
return func(self, *args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/mdata/llh/Projects/ms-swift/swift/trainers/rlhf_trainer/grpo_trainer.py", line 322, in _prepare_inputs
generation_batch = self._generate_and_score_completions(generation_batch)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/mdata/llh/Projects/ms-swift/swift/trainers/rlhf_trainer/grpo_trainer.py", line 812, in _generate_completions
inputs, outputs = self._fast_infer(inputs)
^^^^^^^^^^^^^^^^^^^^^^^^
File "/mdata/llh/Projects/ms-swift/swift/trainers/rlhf_trainer/grpo_trainer.py", line 767, in _fast_infer
self._move_model_to_vllm()
File "/mdata/llh/miniconda3/envs/swift/lib/python3.11/site-packages/trl/extras/profiling.py", line 96, in wrapper
return func(self, *args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/mdata/llh/Projects/ms-swift/swift/trainers/rlhf_trainer/grpo_trainer.py", line 506, in _move_model_to_vllm
llm_model.load_weights(state_dict.items())
File "/mdata/llh/miniconda3/envs/swift/lib/python3.11/site-packages/vllm/model_executor/models/qwen2_5_vl.py", line 1126, in load_weights
return loader.load_weights(weights, mapper=self.hf_to_vllm_mapper)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/mdata/llh/miniconda3/envs/swift/lib/python3.11/site-packages/vllm/model_executor/models/utils.py", line 261, in load_weights
autoloaded_weights = set(self._load_module("", self.module, weights))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/mdata/llh/miniconda3/envs/swift/lib/python3.11/site-packages/vllm/model_executor/models/utils.py", line 222, in _load_module
yield from self._load_module(prefix,
File "/mdata/llh/miniconda3/envs/swift/lib/python3.11/site-packages/vllm/model_executor/models/utils.py", line 195, in _load_module
loaded_params = module_load_weights(weights)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/mdata/llh/miniconda3/envs/swift/lib/python3.11/site-packages/vllm/model_executor/models/qwen2.py", line 486, in load_weights
return loader.load_weights(weights)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/mdata/llh/miniconda3/envs/swift/lib/python3.11/site-packages/vllm/model_executor/models/utils.py", line 261, in load_weights
autoloaded_weights = set(self._load_module("", self.module, weights))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/mdata/llh/miniconda3/envs/swift/lib/python3.11/site-packages/vllm/model_executor/models/utils.py", line 222, in _load_module
yield from self._load_module(prefix,
File "/mdata/llh/miniconda3/envs/swift/lib/python3.11/site-packages/vllm/model_executor/models/utils.py", line 195, in _load_module
loaded_params = module_load_weights(weights)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/mdata/llh/miniconda3/envs/swift/lib/python3.11/site-packages/vllm/model_executor/models/qwen2.py", line 391, in load_weights
param = params_dict[name]
~~~~~~~~~~~^^^^^^
KeyError: 'language_model.layers.0.self_attn.qkv_proj.weight'
Before submitting a new issue...
- Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.