Spaces:
Running
Running
File size: 3,436 Bytes
ce8b42c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 |
import os
os.system("pip install 'https://modelscope-studios.oss-cn-zhangjiakou.aliyuncs.com/SDK/gradio/gradio-4.44.0-py3-none-any.whl?OSSAccessKeyId=LTAI5tCGZWFdkWKivGKCtvTD&Expires=361727611665&Signature=iynlOFVFiaF3OmxatNMHUBPfb3o%3D'")
os.system("pip install starlette==0.38.6 fastapi==0.112.4")
from typing import List, Tuple, Union
from web_ui import WebUI
import math
from qwen_agent.agents import Assistant
from qwen_agent.tools.base import register_tool
from qwen_agent.tools.doc_parser import Record
from qwen_agent.tools.search_tools.base_search import RefMaterialOutput, BaseSearch
from qwen_agent.log import logger
from qwen_agent.gui.gradio import gr
POSITIVE_INFINITY = math.inf
@register_tool('no_search')
class NoSearch(BaseSearch):
def call(self, params: Union[str, dict], docs: List[Union[Record, str, List[str]]] = None, **kwargs) -> list:
"""The basic search algorithm
Args:
params: The dict parameters.
docs: The list of parsed doc, each doc has unique url.
Returns:
The list of retrieved chunks from each doc.
"""
params = self._verify_json_format_args(params)
# Compatible with the parameter passing of the qwen-agent version <= 0.0.3
max_ref_token = kwargs.get('max_ref_token', self.max_ref_token)
# The query is a string that may contain only the original question,
# or it may be a json string containing the generated keywords and the original question
if not docs:
return []
return self._get_the_front_part(docs, max_ref_token)
@staticmethod
def _get_the_front_part(docs: List[Record], max_ref_token: int) -> list:
all_tokens = 0
_ref_list = []
for doc in docs:
text = []
for page in doc.raw:
text.append(page.content)
all_tokens += page.token
now_ref_list = RefMaterialOutput(url=doc.url, text=text).to_dict()
_ref_list.append(now_ref_list)
logger.info(f'Using tokens: {all_tokens}')
if all_tokens > max_ref_token:
raise gr.Error(f"Your document files (around {all_tokens} tokens) exceed the maximum context length ({max_ref_token} tokens).")
return _ref_list
def sort_by_scores(self,
query: str,
docs: List[Record],
max_ref_token: int,
**kwargs) -> List[Tuple[str, int, float]]:
raise NotImplementedError
def app_gui():
# Define the agent
bot = Assistant(llm={
'model': 'pre-qwen-turbo-1m-1101-blade-opt',
'generate_cfg': {
'max_input_tokens': 1000000,
'max_retries': 10,
}},
name='Qwen-Turbo-1M',
description='Qwen-Turbo natively supports input length of up to 1M tokens. You can upload documents for Q&A, supporting file types: PDF/Word/PPT/TXT/HTML.',
rag_cfg={'max_ref_token': 1000000, 'rag_searchers': ['no_search']},
)
chatbot_config = {
'input.placeholder': "Type \"/clear\" to clear the history",
'verbose': True,
}
WebUI(bot, chatbot_config=chatbot_config).run()
if __name__ == '__main__':
import patching # patch qwen-agent to accelerate 1M processing
app_gui()
|