From: François Fleuret <francois@fleuret.org>
Date: Sun, 7 Jan 2024 15:21:02 +0000 (+0100)
Subject: Update.
X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=commitdiff_plain;h=3c5ce93138700c33a055f83ac1a46efb2975e28a;p=mygptrnn.git

Update.
---

diff --git a/mygpt.py b/mygpt.py
index 6e13ff8..5ea927e 100755
--- a/mygpt.py
+++ b/mygpt.py
@@ -656,23 +656,14 @@ class Caterpillar(nn.Module):
         self.rec_K[:, :, t0:t1] = next_K.flatten(2, 3)
 
         if self.training and self.proba_flashback:
+            # insert_flash_back(self.rec_V,V,self.rec_K,K,t0,t1,CL,proba=self.proba_flashback / CL,)
+
             # This piece of code makes the assumption that there is
             # nothing informative before t0, otherwise we'd have to
             # implement a cache for V and K too. This should not be
             # too much of a problem since this is used only during
             # train, where full sequence are available
 
-            # insert_flash_back(
-            # self.rec_V,
-            # V,
-            # self.rec_K,
-            # K,
-            # t0,
-            # t1,
-            # CL,
-            # proba=self.proba_flashback / CL,
-            # )
-
             n = torch.arange(N, device=X.device)[:, None, None, None]
             t = torch.arange(t0, t1, device=X.device)[None, None, :, None]
             dv = torch.arange(DV, device=X.device)[None, None, None, :]