feat: refatorando código para também conseguir fazer requisções assíncronas com o claude

_utils/gerar_relatorio_modelo_usuario/contextual_retriever.py

@@ -1,7 +1,7 @@
 import os
 from langchain_openai import ChatOpenAI
 from typing import List, Dict, Tuple, Optional
-from anthropic import Anthropic
+from anthropic import Anthropic, AsyncAnthropic
 import logging
 from langchain.schema import Document
 import asyncio
@@ -10,6 +10,7 @@ from typing import List
 from multiprocessing import Process, Barrier, Queue
 from dataclasses import dataclass
 from langchain_core.messages import HumanMessage
+from asgiref.sync import sync_to_async
 
 from _utils.gerar_relatorio_modelo_usuario.llm_calls import claude_answer, gpt_answer
 from _utils.gerar_relatorio_modelo_usuario.prompts import contextual_prompt
@@ -21,12 +22,14 @@ from _utils.models.gerar_relatorio import (
 
 lista_contador = []
 
+
 class ContextualRetriever:
     def __init__(
         self, config: RetrievalConfig, claude_api_key: str, claude_context_model: str
     ):
         self.config = config
-        self.claude_client = Anthropic(api_key=claude_api_key)
+        # self.claude_client = Anthropic(api_key=claude_api_key)
+        self.claude_client = AsyncAnthropic(api_key=claude_api_key)
         self.logger = logging.getLogger(__name__)
         self.bm25 = None
         self.claude_context_model = claude_context_model
@@ -36,7 +39,10 @@ class ContextualRetriever:
         try:
             print("COMEÇOU A REQUISIÇÃO")
             prompt = contextual_prompt(full_text, chunk.content)
-            # response = claude_answer(self.claude_client, self.claude_context_model, prompt)
+            # response = await claude_answer(
+            #     self.claude_client, self.claude_context_model, prompt
+            # )
+
             response = await gpt_answer(prompt)
             return response
         except Exception as e:
@@ -48,12 +54,15 @@ class ContextualRetriever:
     async def create_contextualized_chunk(self, chunk, full_text):
         lista_contador.append(0)
         print("contador: ", len(lista_contador))
-        page_content = ""
-        for i in range(
-            max(0, chunk.page_number - 1),
-            min(len(full_text), chunk.page_number + 2),
-        ):
-            page_content += full_text[i].page_content if full_text[i] else ""
+        # Código comentado abaixo é para ler as páginas ao redor da página atual do chunk
+        # page_content = ""
+        # for i in range(
+        #     max(0, chunk.page_number - 1),
+        #     min(len(full_text), chunk.page_number + 2),
+        # ):
+        #     page_content += full_text[i].page_content if full_text[i] else ""
+        page_number = chunk.page_number - 1
+        page_content = full_text[page_number].page_content
 
         context = await self.llm_generate_context(page_content, chunk)
         return ContextualizedChunk(
@@ -70,6 +79,7 @@ class ContextualRetriever:
     ) -> List[ContextualizedChunk]:
         """Add context to all chunks"""
         contextualized_chunks = []
+        lista_contador = []
 
         async with asyncio.TaskGroup() as tg:
             tasks = [

_utils/gerar_relatorio_modelo_usuario/llm_calls.py

@@ -3,8 +3,11 @@ from langchain_core.messages import HumanMessage
 from langchain_openai import ChatOpenAI
 
 
-def claude_answer(claude_client, claude_context_model, prompt):
-    response = claude_client.messages.create(
+async def claude_answer(claude_client, claude_context_model, prompt):
+    print("\n")
+    print("Começou uma requisição pelo Claude")
+    print("\n")
+    response = await claude_client.messages.create(
         model=claude_context_model,
         max_tokens=100,
         messages=[{"role": "user", "content": prompt}],