File size: 4,439 Bytes
0f0b357 |
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 |
"""
Vector Memory Module
Mem贸ria de longo prazo usando ChromaDB + Sentence Transformers
"""
import chromadb
from sentence_transformers import SentenceTransformer
from datetime import datetime
import hashlib
# Usar modelo leve para embeddings
EMBEDDING_MODEL = "all-MiniLM-L6-v2" # ~80MB, r谩pido
# Singleton para evitar recarregar
_memory_instance = None
def get_memory():
"""Get or create memory instance."""
global _memory_instance
if _memory_instance is None:
_memory_instance = VectorMemory()
return _memory_instance
class VectorMemory:
def __init__(self):
print("Inicializando mem贸ria vetorial...")
# Modelo de embeddings
self.model = SentenceTransformer(EMBEDDING_MODEL)
print(f"Modelo carregado: {EMBEDDING_MODEL}")
# ChromaDB em mem贸ria (persiste enquanto o servidor estiver rodando)
self.client = chromadb.Client()
self.collection = self.client.get_or_create_collection(
name="chat_memory",
metadata={"hnsw:space": "cosine"}
)
print(f"Mem贸ria pronta. {self.collection.count()} mem贸rias existentes.")
def _generate_id(self, text: str) -> str:
"""Generate unique ID for a memory."""
timestamp = datetime.now().isoformat()
content = f"{timestamp}:{text}"
return hashlib.md5(content.encode()).hexdigest()
def add_memory(self, user_message: str, bot_response: str):
"""
Add a conversation exchange to memory.
Stores the combined context for better retrieval.
"""
# Combinar mensagem e resposta para contexto completo
combined = f"Usu谩rio: {user_message}\nAssistente: {bot_response}"
# Gerar embedding
embedding = self.model.encode(combined).tolist()
# Gerar ID 煤nico
doc_id = self._generate_id(combined)
# Metadados
metadata = {
"user_message": user_message[:500], # Truncar se muito longo
"bot_response": bot_response[:500],
"timestamp": datetime.now().isoformat()
}
# Adicionar ao banco
self.collection.add(
ids=[doc_id],
embeddings=[embedding],
documents=[combined],
metadatas=[metadata]
)
print(f"Mem贸ria adicionada. Total: {self.collection.count()}")
def search_memories(self, query: str, k: int = 3) -> list[dict]:
"""
Search for relevant memories based on the query.
Returns list of {text, user_message, bot_response, score}
"""
if self.collection.count() == 0:
return []
# Gerar embedding da query
query_embedding = self.model.encode(query).tolist()
# Buscar similares
results = self.collection.query(
query_embeddings=[query_embedding],
n_results=min(k, self.collection.count())
)
memories = []
if results and results['documents'] and results['documents'][0]:
for i, doc in enumerate(results['documents'][0]):
metadata = results['metadatas'][0][i] if results['metadatas'] else {}
distance = results['distances'][0][i] if results['distances'] else 0
memories.append({
"text": doc,
"user_message": metadata.get("user_message", ""),
"bot_response": metadata.get("bot_response", ""),
"score": 1 - distance, # Converter dist芒ncia em similaridade
"timestamp": metadata.get("timestamp", "")
})
return memories
def clear_memories(self):
"""Clear all memories."""
# Recriar collection
self.client.delete_collection("chat_memory")
self.collection = self.client.get_or_create_collection(
name="chat_memory",
metadata={"hnsw:space": "cosine"}
)
print("Mem贸rias limpas.")
def get_stats(self) -> dict:
"""Get memory statistics."""
return {
"total_memories": self.collection.count(),
"model": EMBEDDING_MODEL
}
|