Flux.2训练lora时无法加载transformer模型

我基于Flux.2的FLUX.2-dev.sh的脚本（简单改动）进行lora两阶段拆分训练，提前将模型通过modelscope下载到本地并确定模型完整。现在sft:data_process阶段能够正常结束并生成cache文件，但在第二阶段加载transformer模型时会报错：
ValueError: Cannot detect the model type. File: /root/DiffSynth-Studio/models/flux2/transformer/diffusion_pytorch_model-00001-of-00007.safetensors. Model hash: 706251fd7d252e9b87616f71b5cf306d
我试过加载单一transformer文件依然报错：ValueError: Cannot detect the model type. File: /root/DiffSynth-Studio/models/flux2/flux2-dev.safetensors. Model hash: 76bded5d9167873fba768a0ed3e52d3d

我确定模型文件没有问题（通过modelscope下载，并通过hash校验），感觉像是bug。

git信息：
commit b4f6a4de6c906763bb68ae074302217c081e872a (HEAD, origin/main, origin/HEAD)

FLUX.2-dev.sh配置：
accelerate launch ../train.py
--max_pixels 1048576
--model_paths '[
[
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00001-of-00010.safetensors",
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00002-of-00010.safetensors",
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00003-of-00010.safetensors",
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00004-of-00010.safetensors",
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00005-of-00010.safetensors",
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00006-of-00010.safetensors",
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00007-of-00010.safetensors",
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00008-of-00010.safetensors",
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00009-of-00010.safetensors",
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00010-of-00010.safetensors"
],
"/root/DiffSynth-Studio/models/flux2/vae/diffusion_pytorch_model.safetensors"
]'
--fp8_models '[
[
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00001-of-00010.safetensors",
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00002-of-00010.safetensors",
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00003-of-00010.safetensors",
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00004-of-00010.safetensors",
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00005-of-00010.safetensors",
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00006-of-00010.safetensors",
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00007-of-00010.safetensors",
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00008-of-00010.safetensors",
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00009-of-00010.safetensors",
"/root/DiffSynth-Studio/models/flux2/text_encoder/model-00010-of-00010.safetensors"
],
"/root/DiffSynth-Studio/models/flux2/vae/diffusion_pytorch_model.safetensors"
]'
--lora_target_modules "to_q,to_k,to_v,add_q_proj,add_k_proj,add_v_proj,to_qkv_mlp_proj,to_out.0,to_add_out,linear_in,linear_out,single_transformer_blocks.0.attn.to_out,single_transformer_blocks.1.attn.to_out,single_transformer_blocks.2.attn.to_out,single_transformer_blocks.3.attn.to_out,single_transformer_blocks.4.attn.to_out,single_transformer_blocks.5.attn.to_out,single_transformer_blocks.6.attn.to_out,single_transformer_blocks.7.attn.to_out,single_transformer_blocks.8.attn.to_out,single_transformer_blocks.9.attn.to_out,single_transformer_blocks.10.attn.to_out,single_transformer_blocks.11.attn.to_out,single_transformer_blocks.12.attn.to_out,single_transformer_blocks.13.attn.to_out,single_transformer_blocks.14.attn.to_out,single_transformer_blocks.15.attn.to_out,single_transformer_blocks.16.attn.to_out,single_transformer_blocks.17.attn.to_out,single_transformer_blocks.18.attn.to_out,single_transformer_blocks.19.attn.to_out,single_transformer_blocks.20.attn.to_out,single_transformer_blocks.21.attn.to_out,single_transformer_blocks.22.attn.to_out,single_transformer_blocks.23.attn.to_out,single_transformer_blocks.24.attn.to_out,single_transformer_blocks.25.attn.to_out,single_transformer_blocks.26.attn.to_out,single_transformer_blocks.27.attn.to_out,single_transformer_blocks.28.attn.to_out,single_transformer_blocks.29.attn.to_out,single_transformer_blocks.30.attn.to_out,single_transformer_blocks.31.attn.to_out,single_transformer_blocks.32.attn.to_out,single_transformer_blocks.33.attn.to_out,single_transformer_blocks.34.attn.to_out,single_transformer_blocks.35.attn.to_out,single_transformer_blocks.36.attn.to_out,single_transformer_blocks.37.attn.to_out,single_transformer_blocks.38.attn.to_out,single_transformer_blocks.39.attn.to_out,single_transformer_blocks.40.attn.to_out,single_transformer_blocks.41.attn.to_out,single_transformer_blocks.42.attn.to_out,single_transformer_blocks.43.attn.to_out,single_transformer_blocks.44.attn.to_out,single_transformer_blocks.45.attn.to_out,single_transformer_blocks.46.attn.to_out,single_transformer_blocks.47.attn.to_out"
--tokenizer_path /root/DiffSynth-Studio/models/flux2/tokenizer
--remove_prefix_in_ckpt "pipe.dit."
--lora_base_model "dit"
--dataset_base_path ./training/20260203/dataset
--dataset_metadata_path ./training/20260203/dataset/metadata.csv
--dataset_repeat 1
--dataset_num_workers 8
--output_path ./training/20260203/output_fp8_r32_epo30/cache
--learning_rate 1e-4
--num_epochs 30
--lora_rank 32
--find_unused_parameters
--use_gradient_checkpointing
--use_gradient_checkpointing_offload
--task "sft:data_process"

accelerate launch ../train.py
--max_pixels 1048576
--model_paths '[
"/root/DiffSynth-Studio/models/flux2/flux2-dev.safetensors"
]'
--fp8_models '[
"/root/DiffSynth-Studio/models/flux2/flux2-dev.safetensors"
]'
--lora_target_modules "to_q,to_k,to_v,add_q_proj,add_k_proj,add_v_proj,to_qkv_mlp_proj,to_out.0,to_add_out,linear_in,linear_out,single_transformer_blocks.0.attn.to_out,single_transformer_blocks.1.attn.to_out,single_transformer_blocks.2.attn.to_out,single_transformer_blocks.3.attn.to_out,single_transformer_blocks.4.attn.to_out,single_transformer_blocks.5.attn.to_out,single_transformer_blocks.6.attn.to_out,single_transformer_blocks.7.attn.to_out,single_transformer_blocks.8.attn.to_out,single_transformer_blocks.9.attn.to_out,single_transformer_blocks.10.attn.to_out,single_transformer_blocks.11.attn.to_out,single_transformer_blocks.12.attn.to_out,single_transformer_blocks.13.attn.to_out,single_transformer_blocks.14.attn.to_out,single_transformer_blocks.15.attn.to_out,single_transformer_blocks.16.attn.to_out,single_transformer_blocks.17.attn.to_out,single_transformer_blocks.18.attn.to_out,single_transformer_blocks.19.attn.to_out,single_transformer_blocks.20.attn.to_out,single_transformer_blocks.21.attn.to_out,single_transformer_blocks.22.attn.to_out,single_transformer_blocks.23.attn.to_out,single_transformer_blocks.24.attn.to_out,single_transformer_blocks.25.attn.to_out,single_transformer_blocks.26.attn.to_out,single_transformer_blocks.27.attn.to_out,single_transformer_blocks.28.attn.to_out,single_transformer_blocks.29.attn.to_out,single_transformer_blocks.30.attn.to_out,single_transformer_blocks.31.attn.to_out,single_transformer_blocks.32.attn.to_out,single_transformer_blocks.33.attn.to_out,single_transformer_blocks.34.attn.to_out,single_transformer_blocks.35.attn.to_out,single_transformer_blocks.36.attn.to_out,single_transformer_blocks.37.attn.to_out,single_transformer_blocks.38.attn.to_out,single_transformer_blocks.39.attn.to_out,single_transformer_blocks.40.attn.to_out,single_transformer_blocks.41.attn.to_out,single_transformer_blocks.42.attn.to_out,single_transformer_blocks.43.attn.to_out,single_transformer_blocks.44.attn.to_out,single_transformer_blocks.45.attn.to_out,single_transformer_blocks.46.attn.to_out,single_transformer_blocks.47.attn.to_out"
--tokenizer_path /root/DiffSynth-Studio/models/flux2/tokenizer
--remove_prefix_in_ckpt "pipe.dit."
--lora_base_model "dit"
--dataset_base_path ./training/20260203/output_fp8_r32_epo30/cache
--dataset_repeat 5
--dataset_num_workers 8
--output_path ./training/20260203/output_fp8_r32_epo30
--learning_rate 1e-4
--num_epochs 30
--lora_rank 32
--find_unused_parameters
--use_gradient_checkpointing
--use_gradient_checkpointing_offload
--task "sft:train"

Start with the second accelerate launch of ../train.py in FLUX.2-dev.sh, using the reported flux2-dev.safetensors path, and reproduce the model-type detection error. Trace the transformer loading path from train.py and compare it with the first data_process command. Done means the local FLUX.2 transformer loads successfully and the sft:train stage can proceed.