deepseek-v3微调报错

[elimsjxr]

Describe the bug
What the bug is, and how to reproduce, better with screenshots(描述bug以及复现过程，最好有截图)

升腾npu跑deepseek-v3 lora微调，用的v3.0.1分支代码：
命令：
swift sft
--model /work/share/weights/DeepSeek-V3
--dataset AI-ModelScope/alpaca-gpt4-data-en
--train_type lora
--output_dir output \

报错：
[INFO:swift] Loading the model using model_dir: /work/share/weights/DeepSeek-V3
[WARNING:swift] torch_dtype: torch.bfloat16, but support_bf16: False.
[INFO:swift] model_kwargs: {'device_map': 'npu:0'}
Traceback (most recent call last):
File "/work/share/ms-swift-3.0.1/swift/cli/sft.py", line 5, in
sft_main()
File "/work/share/ms-swift-3.0.1/swift/llm/train/sft.py", line 272, in sft_main
return SwiftSft(args).main()
File "/work/share/ms-swift-3.0.1/swift/llm/train/sft.py", line 30, in init
self._prepare_model_tokenizer()
File "/work/share/ms-swift-3.0.1/swift/llm/train/sft.py", line 62, in _prepare_model_tokenizer
self.model, self.processor = args.get_model_processor()
File "/work/share/ms-swift-3.0.1/swift/llm/argument/base_args/base_args.py", line 255, in get_model_processor
model, processor = get_model_tokenizer(**kwargs)
File "/work/share/ms-swift-3.0.1/swift/llm/model/register.py", line 503, in get_model_tokenizer
model, processor = get_function(model_dir, model_info, model_kwargs, load_model, **kwargs)
File "/work/share/jxr/ms-swift-3.0.1/swift/llm/model/model/deepseek.py", line 53, in get_model_tokenizer_deepseek_moe
model, tokenizer = get_model_tokenizer_with_flash_attn(model_dir, model_info, model_kwargs, load_model, kwargs)
File "/work/share/ms-swift-3.0.1/swift/llm/model/register.py", line 265, in get_model_tokenizer_with_flash_attn
return get_model_tokenizer_from_local(model_dir, model_info, model_kwargs, load_model, kwargs)
File "/work/share/ms-swift-3.0.1/swift/llm/model/register.py", line 188, in get_model_tokenizer_from_local
model = automodel_class.from_pretrained(
File "/usr/local/lib/python3.10/dist-packages/transformers/models/auto/auto_factory.py", line 559, in from_pretrained
return model_class.from_pretrained(
File "/usr/local/lib/python3.10/dist-packages/transformers/modeling_utils.py", line 3442, in from_pretrained
config.quantization_config = AutoHfQuantizer.merge_quantization_configs(
File "/usr/local/lib/python3.10/dist-packages/transformers/quantizers/auto.py", line 169, in merge_quantization_configs
quantization_config = AutoQuantizationConfig.from_dict(quantization_config)
File "/usr/local/lib/python3.10/dist-packages/transformers/quantizers/auto.py", line 93, in from_dict
raise ValueError(
ValueError: Unknown quantization type, got fp8 - supported types are: ['awq', 'bitsandbytes_4bit', 'bitsandbytes_8bit', 'gptq', 'aqlm', 'quanto', 'eetq', 'hqq', 'compressed-tensors', 'fbgemm_fp8', 'torchao']
[ERROR] 2025-01-02-17:23:14 (PID:6941, Device:0, RankID:-1) ERR99999 UNKNOWN application exceptionYour hardware and system info

Write your system info like CUDA version/system/GPU/torch version here(在这里给出硬件信息和系统信息，如CUDA版本，系统，GPU型号和torch版本等)
python 3.10
torch 2.1.0
ms-swift 3.0.1

[Jintao-Huang]

需要将fp8 -> bf16 才能训练的

[elimsjxr]

需要将fp8 -> bf16 才能训练的

这是需要修改模型配置文件吗

[Jintao-Huang]

https://modelscope.cn/models/deepseek-ai/DeepSeek-V3

这里有转的例子