Update.

diff --git a/README.txt b/README.txt
new file mode 100644 (file)
index 0000000..abf3490
--- /dev/null
+++ b/README.txt
@@ -0,0 +1,4 @@
+
+./main.py --task=memory --model=4M-C --nb_epochs=5
+
+./main.py --task=grid --model=37M-C

diff --git a/fridge b/fridge
index 5dd85dd..ac7f86c 100644 (file)
--- a/fridge
+++ b/fridge
@@ -74,3 +74,10 @@ def insert_flash_back(rec_V, V, rec_K, K, t0, t1, CL, proba):
 
 
 ######################################################################
+
+######################################################################
+
+2024 Jan 07 21:38:11 (from mygpt.py)
+
+            # insert_flash_back(self.rec_V,V,self.rec_K,K,t0,t1,CL,proba=self.proba_flashback / CL,)
+

diff --git a/mygpt.py b/mygpt.py
index eda8685..de69a75 100755 (executable)
--- a/mygpt.py
+++ b/mygpt.py
@@ -585,8 +585,6 @@ class Caterpillar(nn.Module):
         self.rec_K[:, :, t0:t1] = next_K.flatten(2, 3)
 
         if self.training and self.proba_flashback > 0.0:
-            # 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