Update memory.py
Browse files
memory.py
CHANGED
@@ -5,43 +5,44 @@ from collections import deque
|
|
5 |
from typing import Deque, Dict, Any
|
6 |
|
7 |
class CognitiveMemory(nn.Module):
|
8 |
-
"""Memory system dengan dimensi
|
9 |
def __init__(self, context_size: int, capacity: int = 100):
|
10 |
super().__init__()
|
11 |
self.context_size = context_size
|
12 |
self.capacity = capacity
|
13 |
self.memory_queue: Deque[Dict[str, Any]] = deque(maxlen=capacity)
|
14 |
|
15 |
-
# Proyeksi
|
16 |
self.key_proj = nn.Linear(context_size, context_size)
|
17 |
self.value_proj = nn.Linear(context_size, context_size)
|
18 |
self.importance_decay = nn.Parameter(torch.tensor(0.95))
|
19 |
|
20 |
def add_memory(self, context: torch.Tensor, activation: float):
|
21 |
-
"""Menyimpan memori dengan dimensi
|
22 |
importance = torch.sigmoid(torch.tensor(activation * 0.5 + 0.2))
|
23 |
self.memory_queue.append({
|
24 |
-
'context': context.detach().clone(),
|
25 |
'importance': importance,
|
26 |
'age': torch.tensor(0.0)
|
27 |
})
|
28 |
|
29 |
def consolidate_memories(self):
|
30 |
-
"""Konsolidasi memori dengan
|
31 |
self.memory_queue = deque(
|
32 |
[m for m in self.memory_queue if m['importance'] > 0.2],
|
33 |
maxlen=self.capacity
|
34 |
)
|
35 |
|
36 |
def retrieve(self, query: torch.Tensor) -> torch.Tensor:
|
37 |
-
"""Retrieval dengan penanganan
|
38 |
if not self.memory_queue:
|
39 |
return torch.zeros(self.context_size, device=query.device)
|
40 |
|
|
|
41 |
contexts = torch.stack([m['context'] for m in self.memory_queue])
|
42 |
keys = self.key_proj(contexts)
|
43 |
values = self.value_proj(contexts)
|
44 |
-
query_proj = self.key_proj(query)
|
45 |
|
46 |
scores = F.softmax(keys @ query_proj, dim=0)
|
47 |
return (scores.unsqueeze(1) * values).sum(dim=0)
|
|
|
5 |
from typing import Deque, Dict, Any
|
6 |
|
7 |
class CognitiveMemory(nn.Module):
|
8 |
+
"""Memory system dengan manajemen dimensi yang ketat"""
|
9 |
def __init__(self, context_size: int, capacity: int = 100):
|
10 |
super().__init__()
|
11 |
self.context_size = context_size
|
12 |
self.capacity = capacity
|
13 |
self.memory_queue: Deque[Dict[str, Any]] = deque(maxlen=capacity)
|
14 |
|
15 |
+
# Proyeksi linear dengan dimensi input/output sama
|
16 |
self.key_proj = nn.Linear(context_size, context_size)
|
17 |
self.value_proj = nn.Linear(context_size, context_size)
|
18 |
self.importance_decay = nn.Parameter(torch.tensor(0.95))
|
19 |
|
20 |
def add_memory(self, context: torch.Tensor, activation: float):
|
21 |
+
"""Menyimpan memori dengan dimensi terkontrol"""
|
22 |
importance = torch.sigmoid(torch.tensor(activation * 0.5 + 0.2))
|
23 |
self.memory_queue.append({
|
24 |
+
'context': context.detach().clone().squeeze(),
|
25 |
'importance': importance,
|
26 |
'age': torch.tensor(0.0)
|
27 |
})
|
28 |
|
29 |
def consolidate_memories(self):
|
30 |
+
"""Konsolidasi memori dengan validasi dimensi"""
|
31 |
self.memory_queue = deque(
|
32 |
[m for m in self.memory_queue if m['importance'] > 0.2],
|
33 |
maxlen=self.capacity
|
34 |
)
|
35 |
|
36 |
def retrieve(self, query: torch.Tensor) -> torch.Tensor:
|
37 |
+
"""Retrieval dengan penanganan tensor 1D"""
|
38 |
if not self.memory_queue:
|
39 |
return torch.zeros(self.context_size, device=query.device)
|
40 |
|
41 |
+
# Penanganan dimensi yang konsisten
|
42 |
contexts = torch.stack([m['context'] for m in self.memory_queue])
|
43 |
keys = self.key_proj(contexts)
|
44 |
values = self.value_proj(contexts)
|
45 |
+
query_proj = self.key_proj(query.squeeze())
|
46 |
|
47 |
scores = F.softmax(keys @ query_proj, dim=0)
|
48 |
return (scores.unsqueeze(1) * values).sum(dim=0)
|