Skip to content
This repository was archived by the owner on Jun 21, 2024. It is now read-only.

Trying to solve the dimensionality bug #1

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Trying to solve the dimensionality bug #1

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 8 additions & 10 deletions t5_pytorch/t5_pytorch.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -146,16 +146,16 @@ def forward(self, x, mask = None):

q = q * self.scale

sim = torch.einsum('b h i d, b h j d -> b h i j', q, k)
sim = torch.einsum('b h i d, b h j d -> b h i j', q, k) # (b, h, n, n)

sim = self.relative_position_bias(sim)

# mask
# mask (b, n)

mask_value = -torch.finfo(sim.dtype).max

if mask is not None:
sim = sim.masked_fill_(~mask, mask_value)
sim = sim.masked_fill_(~mask[:, None, :, None], mask_value)

if self.causal:
i, j = sim.shape[-2:]
Expand Down Expand Up @@ -222,19 +222,19 @@ def forward(self, x, context, mask = None, context_mask = None):

q = q * self.scale

sim = torch.einsum('b h i d, b h j d -> b h i j', q, k)
sim = torch.einsum('b h i d, b h j d -> b h i j', q, k) # (b, h, n, n)

#sim = self.relative_position_bias(sim)

# mask
# mask (b, n)

mask_value = -torch.finfo(sim.dtype).max

if mask is not None:
sim = sim.masked_fill_(~mask, mask_value)
sim = sim.masked_fill_(~mask[:, None, :, None], mask_value)

if context_mask is not None:
sim = sim.masked_fill_(~context_mask[:, None, :], mask_value)
sim = sim.masked_fill_(~context_mask[:, None, None, :], mask_value)

# attention

Expand Down Expand Up @@ -360,7 +360,6 @@ def __init__(
):
super().__init__()

self.embedding = nn.Embedding(enc_num_tokens, dim)
#self.pos_emb = nn.Embedding(max_seq_len, dim)

self.encoder = T5Encoder(
Expand Down Expand Up @@ -392,10 +391,9 @@ def __init__(
self.encoder.token_emb.weight = self.decoder.token_emb.weight

def forward(self, src, tgt, mask = None, context_mask = None):
x = self.embedding(src)
#x = x + self.pos_emb(torch.arange(x.shape[1], device = x.device))
x = self.encoder(src, mask = mask)
x = self.decoder(tgt, x, mask = mask, context_mask = context_mask)
x = self.decoder(tgt, x, mask = context_mask, context_mask = mask)
x = self.to_logits(x)
return x

Expand Down