File size: 6,010 Bytes
365de9c |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 |
from typing import Any, AsyncGenerator
from app.core.models import LocalLLM, Embedder, Reranker, GeminiLLM, GeminiEmbed, Wrapper
from app.core.processor import DocumentProcessor
from app.core.database import VectorDatabase
import time
import os
from app.settings import settings, BASE_DIR
class RagSystem:
def __init__(self):
self.embedder = (
GeminiEmbed()
if settings.use_gemini
else Embedder(model=settings.models.embedder_model)
)
self.reranker = Reranker(model=settings.models.reranker_model)
self.processor = DocumentProcessor(self.embedder)
self.db = VectorDatabase(embedder=self.embedder)
self.llm = GeminiLLM() if settings.use_gemini else LocalLLM()
self.wrapper = Wrapper()
"""
Provides a prompt with substituted context from chunks
TODO: add template to prompt without docs
"""
def get_general_prompt(self, user_prompt: str, collection_name: str) -> str:
enhanced_prompt = self.enhance_prompt(user_prompt.strip())
relevant_chunks = self.db.search(collection_name, query=enhanced_prompt, top_k=30)
if relevant_chunks is not None and len(relevant_chunks) > 0:
ranks = self.reranker.rank(query=enhanced_prompt, chunks=relevant_chunks)[
: min(5, len(relevant_chunks))
]
relevant_chunks = [relevant_chunks[rank["corpus_id"]] for rank in ranks]
else:
relevant_chunks = []
sources = ""
prompt = ""
for chunk in relevant_chunks:
citation = (
f"[Source: {chunk.filename}, "
f"Page: {chunk.page_number}, "
f"Lines: {chunk.start_line}-{chunk.end_line}, "
f"Start: {chunk.start_index}]\n\n"
)
sources += f"Original text:\n{chunk.get_raw_text()}\nCitation:{citation}"
with open(
os.path.join(BASE_DIR, "app", "prompt_templates", "test2.txt")
) as prompt_file:
prompt = prompt_file.read()
prompt += (
"**QUESTION**: "
f"{enhanced_prompt}\n"
"**CONTEXT DOCUMENTS**:\n"
f"{sources}\n"
)
print(prompt)
return prompt
def enhance_prompt(self, original_prompt: str) -> str:
path_to_wrapping_prompt = os.path.join(BASE_DIR, "app", "prompt_templates", "wrapper.txt")
enhanced_prompt = ""
with open(path_to_wrapping_prompt, "r") as f:
enhanced_prompt = f.read().replace("[USERS_PROMPT]", original_prompt)
return self.wrapper.wrap(enhanced_prompt)
"""
Splits the list of documents into groups with 'split_by' docs (done to avoid qdrant_client connection error handling), loads them,
splits into chunks, and saves to db
"""
def upload_documents(
self,
collection_name: str,
documents: list[str],
split_by: int = 3,
debug_mode: bool = True,
) -> None:
for i in range(0, len(documents), split_by):
if debug_mode:
print(
"<"
+ "-" * 10
+ "New document group is taken into processing"
+ "-" * 10
+ ">"
)
docs = documents[i : i + split_by]
loading_time = 0
chunk_generating_time = 0
db_saving_time = 0
print("Start loading the documents")
start = time.time()
self.processor.load_documents(documents=docs, add_to_unprocessed=True)
loading_time = time.time() - start
print("Start loading chunk generation")
start = time.time()
self.processor.generate_chunks()
chunk_generating_time = time.time() - start
print("Start saving to db")
start = time.time()
self.db.store(collection_name, self.processor.get_and_save_unsaved_chunks())
db_saving_time = time.time() - start
if debug_mode:
print(
f"loading time = {loading_time}, chunk generation time = {chunk_generating_time}, saving time = {db_saving_time}\n"
)
def extract_text(self, response) -> str:
text = ""
try:
text = response.candidates[0].content.parts[0].text
except Exception as e:
print(e)
return text
"""
Produces answer to user's request. First, finds the most relevant chunks, generates prompt with them, and asks llm
"""
async def generate_response(
self, collection_name: str, user_prompt: str, stream: bool = True
) -> str:
general_prompt = self.get_general_prompt(
user_prompt=user_prompt, collection_name=collection_name
)
return self.llm.get_response(prompt=general_prompt)
async def generate_response_stream(
self, collection_name: str, user_prompt: str, stream: bool = True
) -> AsyncGenerator[Any, Any]:
general_prompt = self.get_general_prompt(
user_prompt=user_prompt, collection_name=collection_name
)
async for chunk in self.llm.get_streaming_response(
prompt=general_prompt, stream=True
):
yield self.extract_text(chunk)
"""
Produces the list of the most relevant chunkВs
"""
def get_relevant_chunks(self, collection_name: str, query):
relevant_chunks = self.db.search(collection_name, query=query, top_k=15)
relevant_chunks = [
relevant_chunks[ranked["corpus_id"]]
for ranked in self.reranker.rank(query=query, chunks=relevant_chunks)
]
return relevant_chunks
def create_new_collection(self, collection_name: str) -> None:
self.db.create_collection(collection_name)
def get_collections_names(self) -> list[str]:
return self.db.get_collections()
|