lamba_cpu: cuda-guvenli (GQACPU._rope arange + Mamba3CPU.forward zeros device=)

code/kod/lamba_cpu.py

@@ -104,10 +104,10 @@ class Mamba3CPU(nn.Module):
         gamma = trap * DT
 
         # KALİBRE: angle birikimi (işaret/dt-ölçeği) — ilk tahmin: cum += DT·angles
-        h = torch.zeros(B, H, P, S)            # ssm_state (V,QK) = (P,S)
-        x_prev = torch.zeros(B, H, P)
-        Bk_prev = torch.zeros(B, H, S)
-        cum = torch.zeros(B, H, self.n_ang)
+        h = torch.zeros(B, H, P, S, device=u.device)            # ssm_state (V,QK) = (P,S)
+        x_prev = torch.zeros(B, H, P, device=u.device)
+        Bk_prev = torch.zeros(B, H, S, device=u.device)
+        cum = torch.zeros(B, H, self.n_ang, device=u.device)
         ys = []
         for t in range(L):
             # fork: angle = angle_state + tanh(angle_proj)·DT·π  (mamba3_mimo_rotary_step referans)
@@ -182,7 +182,7 @@ class GQACPU(nn.Module):
         self.register_buffer("inv", 1.0 / (base ** (torch.arange(0, self.hd, 2).float() / self.hd)), persistent=False)
 
     def _rope(self, x, T):
-        f = torch.outer(torch.arange(T).float(), self.inv); e = torch.cat((f, f), -1)
+        f = torch.outer(torch.arange(T, device=x.device).float(), self.inv); e = torch.cat((f, f), -1)
         return (x * e.cos()[None, None] + _rot_half(x) * e.sin()[None, None]).to(x.dtype)
 
     def forward(self, x):