Spaces:
Paused
Paused
File size: 12,852 Bytes
869eb7d 0c3c7ed 869eb7d 593b823 869eb7d 3c35194 869eb7d 593b823 869eb7d 0c3c7ed 869eb7d 593b823 0c3c7ed 593b823 869eb7d 593b823 869eb7d 593b823 869eb7d 0c3c7ed 869eb7d 0c3c7ed 869eb7d 0c3c7ed 869eb7d |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 |
import datetime
import os
import traceback
from typing import Any, Coroutine
from dotenv import load_dotenv
from langchain.chains import LLMChain
from langchain.chains.combine_documents import create_stuff_documents_chain
from langchain.chains.retrieval import create_retrieval_chain
from langchain.retrievers import MultiQueryRetriever, MergerRetriever, ContextualCompressionRetriever, EnsembleRetriever
from langchain_cohere import CohereRerank
from langchain_core.documents import Document
from langchain_core.prompts import PromptTemplate, BasePromptTemplate
from agent.Agent import Agent
from agent.agents import chat_openai_llm, deepinfra_chat
from conversation.conversation_store import ConversationStore
from prompt.prompt_store import PromptStore
from retrieval import retrieve_with_rerank
load_dotenv()
conversation_store = ConversationStore()
prompt_store = PromptStore()
grammar_check_1 = prompt_store.get_by_name("gramar_check_1").text
rewrite_hyde_1 = prompt_store.get_by_name("rewrite_hyde_1").text
rewrite_hyde_2 = prompt_store.get_by_name("rewrite_hyde_2").text
rewrite_1 = prompt_store.get_by_name("rewrite_1").text
rewrite_2 = prompt_store.get_by_name("rewrite_2").text
rewrite_hyde = prompt_store.get_by_name("rewrite_hyde").text
def replace_nl(input: str) -> str:
return input.replace('\r\n', '<br>').replace('\n', '<br>').replace('\r', '<br>')
def rewrite(agent: Agent, q: str, prompt: str) -> list[str]:
prompt_template = PromptTemplate(
input_variables=["question"],
template=prompt
)
llm_chain = LLMChain(
llm=agent.llm,
prompt=prompt_template,
verbose=False
)
questions = llm_chain.invoke(
input={"question": q}
)["text"].splitlines()
return [x for x in questions if ("##" not in x and len(str(x).strip()) > 0)]
def rag_with_rerank_check_rewrite_hyde(agent: Agent, q: str, retrieve_document_count: int, prompt: str,
check_prompt: str,
rewrite_prompt: str):
rewritten_list: list[str] = rewrite(agent, q, rewrite_prompt)
if len(rewritten_list) == 0:
return "Neviem, nemám podklady!", "", ""
context_doc = retrieve_subqueries_hyde(agent, retrieve_document_count, rewritten_list)
if len(context_doc) == 0:
return "Neviem, nemám kontext!", "", ""
result = answer_pipeline(agent, context_doc, prompt, q)
answer = result["text"]
check_result = check_pipeline(answer, check_prompt, context_doc, q)
return answer, check_result, context_doc
def rag_with_rerank_check_multi_query_retriever(agent: Agent, q: str, retrieve_document_count: int, prompt: str,
check_prompt: str):
context_doc = hyde_retrieval(agent, retrieve_document_count).invoke(
input=q,
kwargs={"k": retrieve_document_count}
)
if len(context_doc) == 0:
return "Neviem, nemám kontext!", "", ""
result = answer_pipeline(agent, context_doc, prompt, q)
answer = result["text"]
check_result = check_pipeline(answer, check_prompt, context_doc, q)
return answer, check_result, context_doc
async def rag_chain(agent: Agent, q: str, retrieve_document_count: int, prompt: str,
check_prompt: str):
result = await create_retrieval_chain(
retriever=hyde_2_retrieval(agent, retrieve_document_count),
combine_docs_chain=create_stuff_documents_chain(
llm=agent.llm,
prompt=PromptTemplate(
input_variables=["context", "question", "actual_date"],
template=prompt
),
document_prompt=PromptTemplate(input_variables=[], template="page_content")
)
).ainvoke(
input={
"question": q,
"input": q,
"actual_date": datetime.date.today().isoformat()
}
)
print(result)
check_result = check_pipeline(result["answer"], check_prompt, result["context"], q)
print(check_result)
return result["answer"], check_result, result["context"]
def vanilla_rag_chain(agent: Agent, q: str, retrieve_document_count: int, prompt: str,
check_prompt: str):
retriever = ContextualCompressionRetriever(
base_compressor=(CohereRerank(
model="rerank-multilingual-v3.0",
top_n=retrieve_document_count
)),
base_retriever=(agent.embedding.get_vector_store().as_retriever(
search_type="similarity",
search_kwargs={"k": min(retrieve_document_count * 10, 500)},
))
)
result = create_retrieval_chain(
retriever=retriever,
combine_docs_chain=create_stuff_documents_chain(
llm=agent.llm,
prompt=PromptTemplate(
input_variables=["context", "question", "actual_date"],
template=prompt
),
document_prompt=PromptTemplate(input_variables=[], template="page_content")
)
).invoke(
input={
"question": q,
"input": q,
"actual_date": datetime.date.today().isoformat()
}
)
print(result)
check_result = check_pipeline(result["answer"], check_prompt, result["context"], q)
print(check_result)
return result["answer"], check_result, result["context"]
def hyde_retrieval(agent, retrieve_document_count):
retriever_1 = MultiQueryRetriever.from_llm(
llm=agent.llm,
retriever=agent.embedding.get_vector_store().as_retriever(
search_type="similarity",
search_kwargs={"k": retrieve_document_count}
),
prompt=PromptTemplate(
input_variables=["question"],
template=rewrite_hyde_1
)
)
retriever_2 = MultiQueryRetriever.from_llm(
llm=agent.llm,
retriever=agent.embedding.get_vector_store().as_retriever(
search_type="similarity",
search_kwargs={"k": retrieve_document_count}
),
prompt=PromptTemplate(
input_variables=["question"],
template=rewrite_hyde_2
)
)
merge_retriever = MergerRetriever(
retrievers=[retriever_1, retriever_2],
)
compressor = CohereRerank(
model="rerank-multilingual-v3.0",
top_n=retrieve_document_count
)
compression_retriever = ContextualCompressionRetriever(
base_compressor=compressor,
base_retriever=merge_retriever,
search_kwargs={"k": retrieve_document_count},
)
return compression_retriever
def hyde_2_retrieval(agent, retrieve_document_count):
compressor = CohereRerank(
model="rerank-multilingual-v3.0",
top_n=retrieve_document_count / 2
)
retriever_1 = MultiQueryRetriever.from_llm(
llm=agent.llm,
retriever=agent.embedding.get_vector_store().as_retriever(
search_type="similarity",
search_kwargs={"k": min(retrieve_document_count * 10, 300)}
),
prompt=PromptTemplate(
input_variables=["question"],
template=rewrite_1
)
)
compression_retriever_1 = ContextualCompressionRetriever(
base_compressor=compressor,
base_retriever=retriever_1
)
retriever_2 = MultiQueryRetriever.from_llm(
llm=agent.llm,
retriever=agent.embedding.get_vector_store().as_retriever(
search_type="similarity",
search_kwargs={"k": min(retrieve_document_count * 10, 300)}
),
prompt=PromptTemplate(
input_variables=["question"],
template=rewrite_2
)
)
compression_retriever_2 = ContextualCompressionRetriever(
base_compressor=compressor,
base_retriever=retriever_2
)
retriever_3 = MultiQueryRetriever.from_llm(
llm=agent.llm,
retriever=agent.embedding.get_vector_store().as_retriever(
search_type="similarity",
search_kwargs={"k": min(retrieve_document_count * 10, 300)}
),
prompt=PromptTemplate(
input_variables=["question"],
template=rewrite_hyde
)
)
compression_retriever_3 = ContextualCompressionRetriever(
base_compressor=compressor,
base_retriever=retriever_3
)
merge_retriever = EnsembleRetriever(
retrievers=[compression_retriever_1, compression_retriever_2, compression_retriever_3],
weights=[1.0, 1.0, 1.0]
)
return merge_retriever
def retrieve_subqueries(agent, retrieve_document_count, rewritten_list) -> list[Document]:
contexts: list[Document] = []
for rewritten in rewritten_list:
contexts.extend(retrieve_with_rerank(agent.embedding, rewritten, retrieve_document_count))
contexts.sort(key=lambda x: -x.metadata["relevance_score"])
deduplicated: list[Document] = []
for doc in contexts:
already_in = False
for de_doc in deduplicated:
if doc.page_content == de_doc.page_content:
already_in = True
if not already_in:
deduplicated.append(doc)
return deduplicated[:retrieve_document_count]
def retrieve_subqueries_hyde(agent, retrieve_document_count, rewritten_list) -> list[Document]:
contexts: list[Document] = []
for rewritten in rewritten_list:
answer = agent.llm.invoke(rewritten).content
contexts.extend(retrieve_with_rerank(agent.embedding, rewritten + "\n" + answer, retrieve_document_count))
contexts.sort(key=lambda x: -x.metadata["relevance_score"])
deduplicated: list[Document] = []
for doc in contexts:
already_in = False
for de_doc in deduplicated:
if doc.page_content == de_doc.page_content:
already_in = True
if not already_in:
deduplicated.append(doc)
return deduplicated[:retrieve_document_count]
def answer_pipeline(agent, context_doc, prompt, q):
prompt_template = PromptTemplate(
input_variables=["context", "question"],
template=prompt
)
llm_chain = LLMChain(
llm=agent.llm,
prompt=prompt_template,
verbose=False
)
result: dict[str, Any] = llm_chain.invoke(
input={
"question": q,
"context": context_doc,
"actual_date": datetime.date.today().isoformat()
}
)
return result
def check_pipeline(answer, check_prompt, context_doc, q):
prompt_template = PromptTemplate(
input_variables=["context", "question", "answer"],
template=check_prompt
)
llm_chain = LLMChain(
llm=deepinfra_chat("meta-llama/Meta-Llama-3-70B-Instruct", "0.4"),
prompt=prompt_template,
verbose=False
)
try:
check_result = llm_chain.invoke(
input={
"question": q[:2000],
"context": context_doc,
"answer": answer
}
)["text"]
except Exception as e:
check_result = traceback.format_exc()
return check_result
def rag_with_rerank(agent: Agent, q: str, retrieve_document_count: int, prompt: str = None, check_prompt: str = None):
context_doc: list[Document] = retrieve_with_rerank(agent.embedding, q, retrieve_document_count)
try:
result: dict[str, Any] = answer_pipeline(agent, context_doc, prompt, q)
answer = result["text"]
check_result = ""
if check_prompt is not None:
check_result = check_pipeline(answer, check_prompt, context_doc, q)
return answer, check_result, context_doc
except Exception as e:
return "", traceback.format_exc(), ""
def save_conversation(answer: str, check_result: str, context_doc: list[Document], gramatika: str, question: str,
prompt_id: str, check_prompt_id: str, grammar_prompt_id: str):
if len(answer) > 0:
conversation_store.save_content(
q=question,
a=answer,
sources=list(map(lambda doc: doc.page_content, context_doc)),
params=
{
"prompt_id": prompt_id,
"check_prompt_id": check_prompt_id,
"grammar_prompt_id": grammar_prompt_id,
"check_result": check_result,
"grammar_result": gramatika,
"temperature": os.environ["temperature"],
}
)
def check_slovak_agent(text: str) -> str:
prompt_template = PromptTemplate(
input_variables=["text"],
template=grammar_check_1
)
llm_chain = LLMChain(
llm=chat_openai_llm(),
prompt=prompt_template,
verbose=False
)
result: dict[str, Any] = llm_chain.invoke(input={"text": text})
return result["text"]
|