Jetson显卡上运行Qwen2-1.5b模型时报错“RuntimeError triu_tril_cuda_template not implemented for ‘BFloat16‘”

Jetson显卡上运行Qwen2-1.5b模型时报错“RuntimeError triu_tril_cuda_template not implemented for ‘BFloat16’”

问题描述：

CUDA_DEVICE="cuda:0"
model_name_or_path = '/qwen2-1.5b-instruct'
Tokenizer = AutoTokenizer.from_pretrained(model_name_or_path,use_fast=False)
Model = AutoModelForCausalLM.from_pretrained(model_name_or_path, device_map=CUDA_DEVICE, torch_dtype="auto")
print