versão com multiplos agentes e comandos

.dockerignore

@@ -0,0 +1 @@
+app/models/

.gitignore

@@ -1,4 +1,4 @@
 __pycache__/
 .vscode/
 .env
-
+app/models/*

Dockerfile

@@ -4,6 +4,13 @@ FROM python:3.8
 # Copy requirements file and install dependencies
 COPY app/requirements.txt .
 RUN pip install --no-cache-dir -r requirements.txt
+# Instale as dependências necessárias para OpenBLAS e pip (se precisar)
+RUN apt-get update && apt-get install -y \
+    libopenblas-dev \
+    && rm -rf /var/lib/apt/lists/*
+
+# Instale o pacote Python com os argumentos CMake
+RUN CMAKE_ARGS="-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS" pip install llama-cpp-python
 
 # Copy your application code
 COPY app /app

app/agent/lex_chatgpt.py

@@ -0,0 +1,36 @@
+import os
+
+#LLM
+from langchain import OpenAI
+from langchain.chat_models import ChatOpenAI
+from langchain.schema import SystemMessage
+from langchain.prompts import ChatPromptTemplate, HumanMessagePromptTemplate, MessagesPlaceholder
+#Memory
+from langchain.memory import ConversationBufferMemory
+from langchain.chains import LLMChain
+
+#CallBack
+from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler
+
+#Prompts
+from .prompts import SYS_PROMPT
+OPENAI_API_KEY = os.getenv('OPENAI_API_KEY')
+
+#Define o LLM
+llm = ChatOpenAI(model_name="gpt-3.5-turbo")
+
+prompt = ChatPromptTemplate.from_messages([
+    SystemMessage(content=SYS_PROMPT), # The persistent system prompt
+    MessagesPlaceholder(variable_name="chat_history"), # Where the memory will be stored.
+    HumanMessagePromptTemplate.from_template("{human_input}"), # Where the human input will injected
+])
+
+memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)
+llm = ChatOpenAI(streaming=True, callbacks=[StreamingStdOutCallbackHandler()])
+
+chat_llm_chain = LLMChain(
+    llm=llm,
+    prompt=prompt,
+    verbose=True,
+    memory=memory,
+)

app/agent/lex_llama.py

@@ -0,0 +1,59 @@
+import os
+
+#LLM
+from langchain.llms import LlamaCpp
+from langchain.schema import SystemMessage
+from langchain.prompts import PromptTemplate, ChatPromptTemplate, HumanMessagePromptTemplate, MessagesPlaceholder
+
+#Prompt and Memory
+from agent.prompts import SYS_PROMPT
+from langchain.memory import ConversationBufferMemory
+
+#Chain
+from langchain.chains import LLMChain
+
+#Callback para Streaming
+from langchain.callbacks.manager import CallbackManager
+from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler
+
+from utils import download_model
+
+OPENAI_API_KEY = os.getenv('OPENAI_APIKEY')
+
+# Encontra o caminho atual do script
+
+
+
+LLM_MODEL = "llama-2-7b-chat.Q4_K_M.gguf"
+# Concatena com o caminho relativo do modelo
+
+MODEL_PATH = download_model("TheBloke/Llama-2-7b-Chat-GGUF","llama-2-7b-chat.Q4_K_M.gguf")
+
+
+#Define o LLM
+
+# Callbacks support token-wise streaming
+
+callback_manager = CallbackManager([StreamingStdOutCallbackHandler()])
+
+llm = LlamaCpp(
+    model_path=MODEL_PATH,
+    temperature=0.75,
+    max_tokens=512,
+    top_p=1,
+    callback_manager=callback_manager, 
+    verbose=True, # Verbose is required to pass to the callback manager
+    streaming=True,
+    stop=["Human:"]
+)
+#Memória
+prompt = PromptTemplate.from_template("<s>[INST] <<SYS>>\n" + SYS_PROMPT + "\n<</SYS>>\n\n{human_input} [/INST]</s>")
+
+memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)
+
+chat_llm_chain = LLMChain(
+    llm=llm,
+    prompt=prompt,
+    memory=memory,
+    verbose=True
+)

app/agent/prompts.py

@@ -33,4 +33,6 @@
 {chat_history}
 
 Novo input: {input}
-{agent_scratchpad}"""
+{agent_scratchpad}"""
+
+SYS_PROMPT = "You're Lex, an AI Legislative Assisant. From now on you'll speak and write ONLY in Portuguese. Responda sempre e apenas em Português. Não responda em Inglês.\n\nVocê é Lex, uma Inteligência Artificial, enérgica, bem humorada e um pouco sarcástica que atua como co-mandatária em uma proposta de candidatura coletiva progressista para a gestão pública."

app/app.py

@@ -1,70 +1,31 @@
 import os
+from flask import Flask, render_template, request, session
+from flask_socketio import SocketIO
+from importlib import import_module
 from dotenv import load_dotenv
-load_dotenv()
-#LLM
-from langchain import OpenAI
-from langchain.chat_models import ChatOpenAI
-from langchain.schema import SystemMessage
-from langchain.prompts import ChatPromptTemplate, HumanMessagePromptTemplate, MessagesPlaceholder
-
-#Memory
-from langchain.memory import ConversationBufferMemory
-
-from langchain.chains import LLMChain
-
-#Agents
-from langchain.agents.conversational.base import ConversationalAgent #Uses GPT-3.5 Format
-from langchain.agents.conversational_chat.base import ConversationalChatAgent #Uses ChatGPT Format
-from langchain.agents import AgentExecutor
-
-from agent.prompts import PREFIX, SUFFIX, FORMAT_INSTRUCTIONS
-from langchain.prompts import PromptTemplate
-from tools.library import Library
+load_dotenv(os.path.join(os.path.dirname(__file__), '..', '.env'))
 
+ACTIVE_AGENTS = ["lex_chatgpt", "lex_llama"]
 
-from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler
+loaded_agents = {}
+for agent in ACTIVE_AGENTS:
+    module = import_module(f"agent.{agent}")
+    loaded_agents[agent] = module.chat_llm_chain
 
-OPENAI_API_KEY = os.getenv('OPENAI_APIKEY')
+current_agent = os.getenv('DEFAULT_AGENT', loaded_agents[ACTIVE_AGENTS[0]])
 
-#Define o LLM
-llm = OpenAI()
-llm = ChatOpenAI(model_name="gpt-3.5-turbo")
 
-#Define os prefixos e configurações
-ai_prefix = "Lex"
-human_prefix = "Usuário"
+# Carregando comandos
+ACTIVE_COMMANDS = ["agent"]
 
-#Ferramentas
-#biblioteca = Library()
-#library_dir = os.getenv("LIBRARY_DIR") or os.path.join(os.path.dirname(os.path.abspath(__file__)), 'library')
-#library_tools = biblioteca.generate_tools_for_library(library_dir)
-#tools = []# + library_tools
-
-#Memória
-prompt = ChatPromptTemplate.from_messages([
-    SystemMessage(content="You are a chatbot having a conversation with a human."), # The persistent system prompt
-    MessagesPlaceholder(variable_name="chat_history"), # Where the memory will be stored.
-    HumanMessagePromptTemplate.from_template("{human_input}"), # Where the human input will injected
-])
+loaded_commands = {}
+for command in ACTIVE_COMMANDS:
+    module = import_module(f"commands.{command}_command")
+    loaded_commands[command] = module.run
 
-memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)
-
-llm = ChatOpenAI(streaming=True, callbacks=[StreamingStdOutCallbackHandler()])
-
-chat_llm_chain = LLMChain(
-    llm=llm,
-    prompt=prompt,
-    verbose=True,
-    memory=memory,
-)
-
-
-
-from flask import Flask, render_template, request, jsonify
-from flask_socketio import SocketIO
-
 app = Flask(__name__)
-socketio = SocketIO(app, cors_allowed_origins="*")
+app.secret_key = os.getenv('SECRET_KEY', 'supersecret')  # Você precisa definir uma chave secreta para usar sessions
+socketio = SocketIO(app, cors_allowed_origins="*", ping_timeout=120)
 
 @app.route('/')
 def index():
@@ -77,20 +38,39 @@ def index():
 def robots():
     return app.send_static_file('robots.txt')
 
+@socketio.on('connect')
+def initialize_session():
+    session['current_agent'] = current_agent  # Definindo o agente padrão
+
+@socketio.on('select_agent')
+def handle_agent_selection(agent_name):
+    if agent_name in loaded_agents.keys():
+        session['current_agent'] = agent_name
+
 @socketio.on('message')
 def handle_message(message):
     user_input = message.get('message')
     room = request.sid  # Obtém o ID da sessão atual
-
+    current_agent_name = session.get('current_agent', ACTIVE_AGENTS[0])
+    current_agent = loaded_agents[current_agent_name]
+
     if not user_input:
-        socketio.emit('message', {'resposta': 'mensagem não fornecida'}, room=room)  # Envia para o room especificado
-        return
+        return None
+
+    socketio.emit('start_message')  # Envia para o room especificado
+
+    if user_input.startswith('!'):
+        command, *args = user_input[1:].split()
+        if command in loaded_commands:
+            response = f"[{loaded_commands[command](args, session, ACTIVE_AGENTS, loaded_agents)}]"
+
+        else:
+            response = "Comando desconhecido. Por favor, tente novamente."
+    else:
+        response = current_agent.run(human_input=user_input)
 
-    response = chat_llm_chain.predict(human_input=user_input)
-    socketio.emit('start_message', room=room)  # Envia para o room especificado
-    for chunk in response:
-        socketio.emit('message', chunk, room=room)  # Envia para o room especificado
+    socketio.emit('message', response, room=room)  # Envia para o room especificado
     socketio.emit('end_message', room=room)  # Envia para o room especificado
 
 if __name__ == "__main__":
-    socketio.run(app, host='0.0.0.0', port=5000, debug=True, allow_unsafe_werkzeug=True)
+    socketio.run(app, host='0.0.0.0', port=5000, debug=True)

app/commands/agent_command.py

@@ -0,0 +1,17 @@
+# commands/agent_command.py
+
+def run(args, session, ACTIVE_AGENTS, loaded_agents):
+    if args:
+        subcommand = args[0]
+        if subcommand == "list":
+            return f"Agente disponíveis: {', '.join(ACTIVE_AGENTS)}"
+        elif subcommand == "set":
+            new_agent = args[1] if len(args) > 1 else None
+            if new_agent in loaded_agents.keys():
+                session['current_agent'] = new_agent
+                return f"Agente configurado para {new_agent}"
+            else:
+                return "Agente invalido."
+    else:
+        current_agent = session.get('current_agent', ACTIVE_AGENTS[0])
+        return f"O agente atual é {current_agent}"

app/requirements.txt

@@ -26,4 +26,6 @@ flask
 flask_socketio
 
 gunicorn
-eventlet
+eventlet
+
+huggingface-hub==0.17.3

app/templates/index.html

@@ -18,16 +18,22 @@ <h1>Lex.AI</h1>
     </div>
 
     <script>
-        var socket = io.connect(window.location.origin);
+        var socket = io.connect(window.location.origin, {
+            reconnection: true,
+            reconnectionDelay: 1000,
+            reconnectionDelayMax : 5000,
+            reconnectionAttempts: Infinity
+        });
+
         let aiMessageDiv;
+        var chatBox = document.getElementById('chat-box');
 
         function addMessage(message, classe) {
-            var chatBox = document.getElementById('chat-box');
             var messageDiv = document.createElement('div');
             messageDiv.classList.add('message', classe);
             messageDiv.textContent = message;
             chatBox.appendChild(messageDiv);
-            //chatBox.scrollTop = chatBox.scrollHeight;
+            chatBox.scrollTop = chatBox.scrollHeight;
 
             if (classe === 'ai-message') {
                 aiMessageDiv = messageDiv;
@@ -54,6 +60,7 @@ <h1>Lex.AI</h1>
         });
 
         socket.on('end_message', function() {
+            chatBox.scrollTop = chatBox.scrollHeight;
             aiMessageDiv = null;
         });
 

app/utils.py

@@ -0,0 +1,22 @@
+import os
+from huggingface_hub import hf_hub_download
+
+def download_model(repo, model_file):
+    current_path = os.path.dirname(os.path.abspath(__file__))
+    model_directory = os.path.join(current_path, "models")
+
+    # Cria o diretório 'model' se ele não existir
+    if not os.path.exists(model_directory):
+        os.mkdir(model_directory)
+
+    model_path = os.path.join(model_directory, model_file)
+
+    if not os.path.exists(model_path):
+        print(f"O modelo {repo}/{model_file} não existe. Baixando...")
+        hf_hub_download(
+            repo_id=f"{repo}",
+            filename=model_file,
+            cache_dir=model_directory
+        )
+
+    return(model_path)

docker-compose.yml

@@ -31,4 +31,6 @@ services:
       OPENAI_API_KEY: $OPENAI_APIKEY
       WEAVIATE_URL: 'http://localhost:8080'
     depends_on:
-      - weaviate
+      - weaviate
+    volumes:
+      - ./app/models:/app/models