From be092b9d616934edddef63556ad133e9ad9aaf2b Mon Sep 17 00:00:00 2001 From: =?utf8?q?Fran=C3=A7ois=20Fleuret?= Date: Sat, 6 Jan 2024 12:21:45 +0100 Subject: [PATCH] Update. --- mygpt.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/mygpt.py b/mygpt.py index daaec01..7cecc22 100755 --- a/mygpt.py +++ b/mygpt.py @@ -533,7 +533,7 @@ class Caterpillar(nn.Module): # This is the Gating sequence that modulates if they key and # values should be stored in one of the CH pairs of the # current stack. The CH gating values are independent, which - # means that the same thing could be stored multiple times or + # means that the same thing could be stored up to CH times or # not at all G = ( @@ -586,7 +586,7 @@ class Caterpillar(nn.Module): self.rec_K[:, :, t0 - CL + 1 : t1], dim=2, win_dim=3, win_size=CL ) - # We have an attention score for each of the CHxCL value + # We have an attention score for each of the CHxCL values ar = torch.einsum( "nhtd,nftld->nhtfl", -- 2.39.5