File size: 5,006 Bytes
0632cd1 |
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 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 |
import asyncio
from typing import List, Dict
import faiss
import numpy as np
import pandas as pd
from sqlalchemy.ext.asyncio import AsyncSession
from starlette.websockets import WebSocket
from project.bot.models import MessagePair
from project.config import settings
class SearchBot:
chat_history = []
# is_unknown = False
# unknown_counter = 0
def __init__(self, memory=None):
if memory is None:
memory = []
self.chat_history = memory
async def _summarize_user_intent(self, user_query: str) -> str:
chat_history_str = ''
chat_history = self.chat_history[-self.unknown_counter * 2:]
for i in chat_history:
if i['role'] == 'user':
chat_history_str += f"{i['role']}: {i['content']}\n"
messages = [
{
'role': 'system',
'content': f"{settings.SUMMARIZE_PROMPT}\n"
f"Chat history: ```{chat_history_str}```\n"
f"User query: ```{user_query}```"
}
]
response = await settings.OPENAI_CLIENT.chat.completions.create(
messages=messages,
temperature=0.1,
n=1,
model="gpt-3.5-turbo-0125"
)
user_intent = response.choices[0].message.content
return user_intent
@staticmethod
def _cls_pooling(model_output):
return model_output.last_hidden_state[:, 0]
async def _convert_to_embeddings(self, text_list):
encoded_input = settings.INFO_TOKENIZER(
text_list, padding=True, truncation=True, return_tensors="pt"
)
encoded_input = {k: v.to(settings.device) for k, v in encoded_input.items()}
model_output = settings.INFO_MODEL(**encoded_input)
return self._cls_pooling(model_output).cpu().detach().numpy().astype('float32')
@staticmethod
async def _get_context_data(user_query: list[float]) -> list[dict]:
radius = 30
_, distances, indices = settings.FAISS_INDEX.range_search(user_query, radius)
indices_distances_df = pd.DataFrame({'index': indices, 'distance': distances})
filtered_data_df = settings.products_dataset.iloc[indices].copy()
filtered_data_df.loc[:, 'distance'] = indices_distances_df['distance'].values
sorted_data_df: pd.DataFrame = filtered_data_df.sort_values(by='distance').reset_index(drop=True)
sorted_data_df = sorted_data_df.drop('distance', axis=1)
data = sorted_data_df.head(3).to_dict(orient='records')
return data
@staticmethod
async def create_context_str(context: List[Dict]) -> str:
context_str = ''
for i, chunk in enumerate(context):
context_str += f'{i + 1}) {chunk["chunks"]}'
return context_str
async def _rag(self, context: List[Dict], query: str, session: AsyncSession, country: str):
if context:
context_str = await self.create_context_str(context)
assistant_message = {"role": 'assistant', "content": context_str}
self.chat_history.append(assistant_message)
content = settings.PROMPT
else:
content = settings.EMPTY_PROMPT
user_message = {"role": 'user', "content": query}
self.chat_history.append(user_message)
messages = [
{
'role': 'system',
'content': content
},
]
messages = messages + self.chat_history
stream = await settings.OPENAI_CLIENT.chat.completions.create(
messages=messages,
temperature=0.1,
n=1,
model="gpt-3.5-turbo",
stream=True
)
response = ''
async for chunk in stream:
if chunk.choices[0].delta.content is not None:
chunk_content = chunk.choices[0].delta.content
response += chunk_content
yield response
await asyncio.sleep(0.02)
assistant_message = {"role": 'assistant', "content": response}
self.chat_history.append(assistant_message)
try:
session.add(MessagePair(user_message=query, bot_response=response, country=country))
except Exception as e:
print(e)
async def ask_and_send(self, data: Dict, websocket: WebSocket, session: AsyncSession):
query = data['query']
country = data['country']
transformed_query = await self._convert_to_embeddings(query)
context = await self._get_context_data(transformed_query)
try:
async for chunk in self._rag(context, query, session, country):
await websocket.send_text(chunk)
# await websocket.send_text('finish')
except Exception:
await self.emergency_db_saving(session)
@staticmethod
async def emergency_db_saving(session: AsyncSession):
await session.commit()
await session.close()
|