初始化拥抱脸变压器重新启动程序循环

Question

初始化拥抱面变压器会导致循环重新启动。我创建了简单的循环，它读取文本并回复，但循环在初始化聊天机器人管道时重新启动新线程。下面给出了最小复制示例。

from transformers import pipeline
from transformers.pipelines import Text2TextGenerationPipeline

chatbot_model = pipeline(task="text-generation",model="facebook/blenderbot-400M-distill")
i=0
while(True):
    try:
        if (i%2)==0:
            print(f"{i} - Even")
            i = i+1
            time.sleep(0.5)
        else:
            i = i+1
            continue
    except Exception as e:
        pass

输出循环在打印 40 后聊天机器人初始化后重新启动，并且似乎其他循环从 0 开始。当我在循环中读取文件时也是如此。是由于多线程造成的。运作如何？ 下面提到的输出。

Device set to use cuda:0
0 - Even
2 - Even
4 - Even
6 - Even
8 - Even
10 - Even
12 - Even
14 - Even
16 - Even
18 - Even
20 - Even
22 - Even
24 - Even
26 - Even
28 - Even
30 - Even
32 - Even
34 - Even
36 - Even
38 - Even
40 - Even
The model 'TFBlenderbotForConditionalGeneration' is not supported for text-generation. Supported models are ['TFBertLMHeadModel', 'TFCamembertForCausalLM', 'TFCTRLLMHeadModel', 'TFGPT2LMHeadModel', 'TFGPT2LMHeadModel', 'TFGPTJForCausalLM', 'TFMistralForCausalLM',].
0 - Even
42 - Even
2 - Even
44 - Even
4 - Even
46 - Even

Answer 1

我认为问题出在管道中指定的任务中，请尝试使用

text2text-generation

我认为这应该可以解决问题:)