镜像：image.sourcefind.cn:5000/dcu/admin/base/jupyterlab-llamafactory:pytorch2.4.1-py3.10-dtk25.04.1-model 安装了：bitsandbytes-0.42.0+das.opt1.dtk24043-py3-none-any.whl 开始微调，报错，提示没有triton包 然后安装：triton-3.0.0+das.opt1.dtk25041-cp310-cp310-manylinux_2_28_x86_64.whl 开始微调，命令 llamafactory-cli train --stage sft --do_train True --model_name_or_path /data/Hugging-Face/hub/models--Qwen--Qwen2.5-32B-Instruct/snapshots/5ede1c97bbab6ce5cda5812749b4c0bdf79b18dd --preprocessing_num_workers 16 --finetuning_type lora --template qwen --flash_attn auto --dataset_dir data --dataset output_B,output_A,jidiqa_dataset,xiaoliu,baiwenbaida --cutoff_len 2048 --learning_rate 5e-05 --num_train_epochs 10.0 --max_samples 100000 --per_device_train_batch_size 2 --gradient_accumulation_steps 8 --lr_scheduler_type cosine --max_grad_norm 1.0 --logging_steps 1 --save_steps 100 --warmup_steps 0 --packing False --enable_thinking True --report_to none --output_dir saves/Qwen2.5-32B-Instruct/lora/train_2025-12-15-09-14-12 --bf16 True --plot_loss True --trust_remote_code True --ddp_timeout 180000000 --include_num_input_tokens_seen True --optim adamw_torch --quantization_bit 4 --quantization_method bnb --double_quantization True --lora_rank 8 --lora_alpha 16 --lora_dropout 0 --lora_target all 报错：[rank2]: File "/opt/conda/lib/python3.10/site-packages/accelerate/big_modeling.py", line 502, in dispatch_model [rank2]: model.to(device) [rank2]: File "/opt/conda/lib/python3.10/site-packages/transformers/modeling_utils.py", line 3688, in to [rank2]: raise ValueError( [rank2]: ValueError: Calling `to()` is not supported for `4-bit` quantized models with the installed version of bitsandbytes. The current device is `cuda:2`. If you intended to move the model, please install bitsandbytes >= 0.43.2. 官网目前最新版本的bitsandbytes是0.42 ps.这个机器我最早五六月份的时候，用过llama-factory int4量化成功微调过模型。 现在完全不行了，不知道是哪里有问题？