file_rag/config.py

import argparse
from transformers import AutoTokenizer, AutoModelForSequenceClassification
from FlagEmbedding import BGEM3FlagModel
from fastapi import WebSocket
from util import get_cus_logger
from marker.converters.pdf import PdfConverter
from marker.models import create_model_dict
def load_parse_args():
    # 创建 ArgumentParser 对象
    parser = argparse.ArgumentParser(description="Process some integers.")
    parser.add_argument('--base_url', type=str, default="https://dashscope.aliyuncs.com/compatible-mode/v1",help='llm base_url')
    parser.add_argument('--api_key', type=str, default="sk-04b63960983445f980d85ff185a17876",help='llm api_key')
    parser.add_argument('--model', type=str, choices=['qwen-max', 'gpt-4'], default='qwen-max', help='The model to use')
    parser.add_argument('--static_dir', type=str, default="/workspace", help='the directory for the code to work')

    # 解析命令行参数
    args = parser.parse_args()
    return args

nltk_path = '/root/nltk_data/tokenizers'
args = load_parse_args()
static_dir = '/workspace'
llm_config={
    "config_list": [
        {
            "model": args.model, # Same as in vLLM command
            "api_key": args.api_key, # Not needed
            "base_url": args.base_url  # Your vLLM URL, with '/v1' added
        }
    ],
    "cache_seed": None, # Turns off caching, useful for testing different models
    "temperature": 0.5,
}
llm_config_ds={
    "config_list": [
        {
            "model": 'deepseek-r1', # Same as in vLLM command
            "api_key": args.api_key, # Not needed
            "base_url": args.base_url  # Your vLLM URL, with '/v1' added
        }
    ],
    "cache_seed": None, # Turns off caching, useful for testing different models
    "temperature": 0.5,
}

milvus_url = "http://10.41.1.57:19530"

bge_model_path = '/model/bge-m3'
bge_rerank_path = '/model/bge-reranker-v2-m3'
BASE_UPLOAD_DIRECTORY = '/workspace'
upload_path = """/workspace/{client_id}/"""
converter = PdfConverter(
    artifact_dict=create_model_dict(),
)

bge_model = BGEM3FlagModel(bge_model_path, use_fp16=True, device='cuda:0') # Setting use_fp16 to True speeds up computation with a slight performance degradation
bge_rerank_tokenizer = AutoTokenizer.from_pretrained(bge_rerank_path)
bge_rerank_model = AutoModelForSequenceClassification.from_pretrained(bge_rerank_path)
bge_rerank_model.to('cuda:0')
bge_rerank_model.eval()