测试使用原始精度模型微调的时候没问题，但是相同配置，使用低精度Int4微调的时候出错了？

[thinkbig]

微调低精度模型的时候需要额外配置什么参数吗？
核心错误是：
TypeError: output tensor must have the same type as input tensor

当前配置如下:
nproc_per_node=2

CUDA_VISIBLE_DEVICES=0,1
NPROC_PER_NODE=$nproc_per_node
swift sft
--model './models/DeepSeek-R1-Distill-Qwen-32B-Int4-W4A16'
--model_type deepseek_r1_distill
--train_type lora
--dataset './dataset/dftest/'
--num_train_epochs 1
--per_device_train_batch_size 1
--gradient_accumulation_steps 4
--learning_rate 1e-4
--max_grad_norm 0.5
--lora_rank 8
--lora_alpha 32
--target_modules all-linear
--eval_steps 100
--save_steps 100
--save_total_limit 5
--logging_steps 5
--max_length 1024
--packing True
--gradient_checkpointing True
--output_dir output
--system 'You are a helpful assistant.'
--warmup_ratio 0.05
--dataloader_num_workers 4
--dataset_num_proc 4
--lr_scheduler_type cosine
--max_grad_norm 0.5
--optim adamw_bnb_8bit
--report_to none
--deepspeed zero3

完整错误栈：

[rank0]: Traceback (most recent call last):
[rank0]: File "/home/aicarmap/tools/swift/swift/cli/sft.py", line 5, in
[rank0]: sft_main()
[rank0]: File "/home/aicarmap/tools/swift/swift/llm/train/sft.py", line 257, in sft_main
[rank0]: return SwiftSft(args).main()
[rank0]: File "/home/aicarmap/tools/swift/swift/llm/base.py", line 46, in main
[rank0]: result = self.run()
[rank0]: File "/home/aicarmap/tools/swift/swift/llm/train/sft.py", line 137, in run
[rank0]: return self.train(trainer)
[rank0]: File "/home/aicarmap/tools/swift/swift/llm/train/sft.py", line 196, in train
[rank0]: trainer.train(trainer.args.resume_from_checkpoint)
[rank0]: File "/home/aicarmap/tools/swift/swift/trainers/mixin.py", line 262, in train
[rank0]: res = super().train(*args, **kwargs)
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/transformers/trainer.py", line 2241, in train
[rank0]: return inner_training_loop(
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/transformers/trainer.py", line 2548, in _inner_training_loop
[rank0]: tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/transformers/trainer.py", line 3740, in training_step
[rank0]: self.accelerator.backward(loss, **kwargs)
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/accelerate/accelerator.py", line 2321, in backward
[rank0]: self.deepspeed_engine_wrapped.backward(loss, **kwargs)
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/accelerate/utils/deepspeed.py", line 275, in backward
[rank0]: self.engine.step()
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/runtime/engine.py", line 2322, in step
[rank0]: self._take_model_step(lr_kwargs)
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/runtime/engine.py", line 2225, in _take_model_step
[rank0]: self.optimizer.step()
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/utils/nvtx.py", line 18, in wrapped_fn
[rank0]: ret_val = func(*args, **kwargs)
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/runtime/zero/stage3.py", line 2130, in step
[rank0]: self._post_step(timer_names)
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/utils/nvtx.py", line 18, in wrapped_fn
[rank0]: ret_val = func(*args, **kwargs)
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/runtime/zero/stage3.py", line 2056, in _post_step
[rank0]: self.persistent_parameters[0].all_gather(self.persistent_parameters)
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/runtime/zero/partition_parameters.py", line 1162, in all_gather
[rank0]: return self._all_gather(param_list, async_op=async_op, hierarchy=hierarchy)
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/utils/nvtx.py", line 18, in wrapped_fn
[rank0]: ret_val = func(*args, **kwargs)
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/runtime/zero/partition_parameters.py", line 1530, in _all_gather
[rank0]: self._allgather_params_coalesced(all_gather_nonquantize_list, hierarchy, quantize=False)
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/runtime/zero/partition_parameters.py", line 1840, in _allgather_params_coalesced
[rank0]: h = dist.all_gather_into_tensor(allgather_params[param_idx],
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/comm/comm.py", line 117, in log_wrapper
[rank0]: return func(*args, **kwargs)
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/comm/comm.py", line 311, in all_gather_into_tensor
[rank0]: return cdb.all_gather_into_tensor(output_tensor=output_tensor, input_tensor=tensor, group=group, async_op=async_op)
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/comm/torch.py", line 224, in all_gather_into_tensor
[rank0]: return self.all_gather_function(output_tensor=output_tensor,
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/torch/distributed/c10d_logger.py", line 83, in wrapper
[rank0]: return func(*args, **kwargs)
[rank0]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py", line 3435, in all_gather_into_tensor
[rank0]: work = group._allgather_base(output_tensor, input_tensor, opts)
[rank0]: TypeError: output tensor must have the same type as input tensor
[rank1]: Traceback (most recent call last):
[rank1]: File "/home/aicarmap/tools/swift/swift/cli/sft.py", line 5, in
[rank1]: sft_main()
[rank1]: File "/home/aicarmap/tools/swift/swift/llm/train/sft.py", line 257, in sft_main
[rank1]: return SwiftSft(args).main()
[rank1]: File "/home/aicarmap/tools/swift/swift/llm/base.py", line 46, in main
[rank1]: result = self.run()
[rank1]: File "/home/aicarmap/tools/swift/swift/llm/train/sft.py", line 137, in run
[rank1]: return self.train(trainer)
[rank1]: File "/home/aicarmap/tools/swift/swift/llm/train/sft.py", line 196, in train
[rank1]: trainer.train(trainer.args.resume_from_checkpoint)
[rank1]: File "/home/aicarmap/tools/swift/swift/trainers/mixin.py", line 262, in train
[rank1]: res = super().train(*args, **kwargs)
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/transformers/trainer.py", line 2241, in train
[rank1]: return inner_training_loop(
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/transformers/trainer.py", line 2548, in _inner_training_loop
[rank1]: tr_loss_step = self.training_step(model, inputs, num_items_in_batch)
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/transformers/trainer.py", line 3740, in training_step
[rank1]: self.accelerator.backward(loss, **kwargs)
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/accelerate/accelerator.py", line 2321, in backward
[rank1]: self.deepspeed_engine_wrapped.backward(loss, **kwargs)
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/accelerate/utils/deepspeed.py", line 275, in backward
[rank1]: self.engine.step()
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/runtime/engine.py", line 2322, in step
[rank1]: self._take_model_step(lr_kwargs)
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/runtime/engine.py", line 2225, in _take_model_step
[rank1]: self.optimizer.step()
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/utils/nvtx.py", line 18, in wrapped_fn
[rank1]: ret_val = func(*args, **kwargs)
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/runtime/zero/stage3.py", line 2130, in step
[rank1]: self._post_step(timer_names)
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/utils/nvtx.py", line 18, in wrapped_fn
[rank1]: ret_val = func(*args, **kwargs)
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/runtime/zero/stage3.py", line 2056, in _post_step
[rank1]: self.persistent_parameters[0].all_gather(self.persistent_parameters)
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/runtime/zero/partition_parameters.py", line 1162, in all_gather
[rank1]: return self._all_gather(param_list, async_op=async_op, hierarchy=hierarchy)
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/utils/nvtx.py", line 18, in wrapped_fn
[rank1]: ret_val = func(*args, **kwargs)
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/runtime/zero/partition_parameters.py", line 1530, in _all_gather
[rank1]: self._allgather_params_coalesced(all_gather_nonquantize_list, hierarchy, quantize=False)
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/runtime/zero/partition_parameters.py", line 1840, in _allgather_params_coalesced
[rank1]: h = dist.all_gather_into_tensor(allgather_params[param_idx],
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/comm/comm.py", line 117, in log_wrapper
[rank1]: return func(*args, **kwargs)
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/comm/comm.py", line 311, in all_gather_into_tensor
[rank1]: return cdb.all_gather_into_tensor(output_tensor=output_tensor, input_tensor=tensor, group=group, async_op=async_op)
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/deepspeed/comm/torch.py", line 224, in all_gather_into_tensor
[rank1]: return self.all_gather_function(output_tensor=output_tensor,
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/torch/distributed/c10d_logger.py", line 83, in wrapper
[rank1]: return func(*args, **kwargs)
[rank1]: File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/torch/distributed/distributed_c10d.py", line 3435, in all_gather_into_tensor
[rank1]: work = group._allgather_base(output_tensor, input_tensor, opts)
[rank1]: TypeError: output tensor must have the same type as input tensor
Train: 0%| | 0/1 [02:12<?, ?it/s]
W0224 02:11:09.274000 1721624 site-packages/torch/distributed/elastic/multiprocessing/api.py:897] Sending process 1721696 closing signal SIGTERM
E0224 02:11:09.539000 1721624 site-packages/torch/distributed/elastic/multiprocessing/api.py:869] failed (exitcode: 1) local_rank: 0 (pid: 1721695) of binary: /home/aicarmap/anaconda3/envs/msswift/bin/python
Traceback (most recent call last):
File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/runpy.py", line 196, in _run_module_as_main
return _run_code(code, main_globals, None,
File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/runpy.py", line 86, in _run_code
exec(code, run_globals)
File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/torch/distributed/run.py", line 923, in
main()
File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/init.py", line 355, in wrapper
return f(*args, **kwargs)
File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/torch/distributed/run.py", line 919, in main
run(args)
File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/torch/distributed/run.py", line 910, in run
elastic_launch(
File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 138, in call
return launch_agent(self._config, self._entrypoint, list(args))
File "/home/aicarmap/anaconda3/envs/msswift/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 269, in launch_agent
raise ChildFailedError(
torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
/home/aicarmap/tools/swift/swift/cli/sft.py FAILED