accelerate                               1.12.0
mindnlp                                  0.4.1
mindspore                                2.4.10
mindspore_lite                           2.4.10
modelarts-mindspore-model-server         1.0.6
tokenizers                               0.19.1
transformers                             4.40.0

# 华为云 环境变量设置
import env


import mindspore
import mindspore as ms
import mindnlp
from transformers import pipeline
import os
# 
# 指定 Ascend NPU
ms.set_context(device_target="Ascend", mode=ms.GRAPH_MODE, device_id=0)

pipe = pipeline(
    "text-generation",
    model="Qwen/Qwen1.5-1.8B",
    # ms_dtype=mindspore.bfloat16,   # 大模型给的写法，但版本对应不上
    # mirror='modelscope',
    device_map="auto"
)

messages = [{"role": "user", "content": "那么打个招呼吧，你好, 介绍一下成龙。"}]
import time
start = time.time()
print(pipe(messages, max_new_tokens=100)[0]["generated_text"][-1]["content"])
print(time.time()  - start)