Bug Description

我编写了一个没有前端界面的聊天机器人(提问的方式是将问题写入prompt.txt文件，程序逐行返回答案),在使用Ollama本地加载llama2-7b后，我使用LlamaIndex(RAG)调整了预训练模型。在转换阶段，我对Ingest Pipeline执行进行了并行化。使用串行执行不会引起问题，但在使用并行执行时，会出现一系列关于超时的异常。以下是我的代码：

import torch
import chromadb
import time
import os
import cProfile
import pstats
from pstats import SortKey
from llama_index.llms.ollama import Ollama
from llama_index.llms.openai import OpenAI
from llama_index.core import Settings
from llama_index.core import SimpleDirectoryReader, VectorStoreIndex
from llama_index.core.node_parser import SentenceSplitter
from llama_index.core.node_parser import SimpleNodeParser
from llama_index.core import ServiceContext
from llama_index.vector_stores.chroma import ChromaVectorStore
from llama_index.core import StorageContext
from llama_index.embeddings.ollama import OllamaEmbedding
from llama_index.core.extractors import TitleExtractor
from llama_index.core.postprocessor import LongContextReorder
from llama_index.core.ingestion import IngestionPipeline
from llama_index.core.evaluation import (FaithfulnessEvaluator,
 QueryResponseEvaluator,
 DatasetGenerator,
 RelevancyEvaluator)
import asyncio
import nest_asyncio
nest_asyncio.apply()

# load for the uploaded files

def data_loading():
 reader = SimpleDirectoryReader(input_dir="data/", recursive=True)
 documents = reader.load_data(num_workers=4)
 print("finish reading files
")
 return documents

# RAG

def building_reg_pipeline():
 # LLM, Llama2
 Settings.llm = Ollama(model="llama2",
 request_timeout=300.0,
 device_map="cpu")

# query from the LLM via query engine

def querying(prompt, query_engine):
 start_time = time.time()
 response = query_engine.query(prompt)
 query_time = time.time() - start_time
 print("Total query time : ", query_time, "
")
 return response

# load for prompts(our questions)

def prompt_loading():
 with open("prompt/prompt.txt", "r") as prompt_file:
 lines = prompt_file.readlines()
 for line in lines:
 yield line.rstrip()

# load for auto generation prompts(questions), not finished yet

def auto_generator_prompt(documents):
 data_generator = DatasetGenerator.from_documents(documents)
 eval_questions = data_generator.generate_questions_from_nodes()
 yield eval_questions
def main():
 # load for query_engine
 query_engine = building_reg_pipeline()
 # performance evaluation of data loading
 p = pstats.Stats("newstas")
 p.strip_dirs().sort_stats(SortKey.CUMULATIVE).print_stats(15)
if __name__ == '__main__':
 main()