Upload positional.py with huggingface_hub

positional.py

@@ -0,0 +1,27 @@
+# positional.py
+import torch
+import torch.nn as nn
+from math import log
+
+class PositionalEncoding(nn.Module):
+    def __init__(self, d_model: int, dropout: float = 0.1, max_len: int = 500):
+        super().__init__()
+        self.dropout = nn.Dropout(p=dropout)
+
+        position = torch.arange(max_len, dtype=torch.float).unsqueeze(1)  # (max_len, 1)
+        div_term = torch.exp(
+            torch.arange(0, d_model, 2, dtype=torch.float) * (-log(10000.0) / d_model)
+        )  # (d_model/2,)
+
+        pe = torch.zeros(max_len, d_model, dtype=torch.float)  # (max_len, d_model)
+        pe[:, 0::2] = torch.sin(position * div_term)          # (max_len, d_model/2)
+        pe[:, 1::2] = torch.cos(position * div_term)
+
+        pe = pe.unsqueeze(0) 
+        self.register_buffer("pe", pe, persistent=False)      # buffer, pas paramètre
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        # x: (B, S, D)
+        s = x.size(1)
+        x = x + self.pe[:, :s, :]  # (1,S,D) broadcast -> (B,S,D)
+        return self.dropout(x)