Update.

[mygpt.git] / mygpt.py

diff --git a/mygpt.py b/mygpt.py
index 212e1a5..9da2e68 100755 (executable)
--- a/mygpt.py
+++ b/mygpt.py
@@ -14,7 +14,7 @@ from torch.nn import functional as F
 
 ##############################
 
-class Residual(nn.Module):
+class WithResidual(nn.Module):
     def __init__(self, *f):
         super().__init__()
         self.f = f[0] if len(f) == 1 else nn.Sequential(*f)
@@ -103,8 +103,8 @@ class MyGPT(nn.Module):
 
         for _ in range(nb_blocks):
             trunk_blocks += [
-                Residual(
-                    nn.LayerNorm(dim_model),
+                WithResidual(
+                    nn.LayerNorm((dim_model,)),
                     QKVAttention(
                         dim_in = dim_model,
                         dim_qk = dim_keys,
@@ -113,8 +113,8 @@ class MyGPT(nn.Module):
                         causal = True, attention_dropout = dropout
                     ),
                 ),
-                Residual(
-                    nn.LayerNorm(dim_model),
+                WithResidual(
+                    nn.LayerNorm((dim_model,)),
                     nn.Linear(in_features = dim_model, out_features = dim_hidden),
                     nn.ReLU(),
                     nn.Linear(in_features = dim_hidden, out_features = dim_model),
@@ -131,7 +131,8 @@ class MyGPT(nn.Module):
         x = self.embedding(x)
         x = self.trunk(x)
         x = self.readout(x)
-        return x[:, :-1]
+        x = F.pad(x, (0, 0, 0, -1))
+        return x
 
 ######################################################################