QLoRa微调模型报错:AssertionError: Both operands must be same dtype. Got fp16 and bf16

环境：

Ubuntu20.04

Llama factory

Qwen2.5-7B-Instruct

llama.cpp

H20 95G

问题描述：

QLoRa微调模型报错:AssertionError: Both operands must be same dtype. Got fp16 and bf16

[INFO|trainer.py:2423] 2025-05-08 00:06:20,829 >>   Number of trainable parameters = 105,512,960
  0%|