1 files changed, 59 insertions, 13 deletions

diff --git a/text_recognizer/networks/transformer/attention.py b/text_recognizer/networks/transformer/attention.py
index 8724691..623d680 100644
--- a/text_recognizer/networks/transformer/attention.py
+++ b/text_recognizer/networks/transformer/attention.py
@@ -1,9 +1,10 @@
 """Implementes the attention module for the transformer."""
 from typing import Optional, Tuple
 
+from einops import rearrange
 from einops.layers.torch import Rearrange
-import numpy as np
 import torch
+from torch import einsum
 from torch import nn
 from torch import Tensor
 import torch.nn.functional as F
@@ -34,7 +35,7 @@ class Attention(nn.Module):
         self.attn_fn = F.softmax
 
         # Feedforward
-        self.proj = nn.Linear(inner_dim, dim)
+        self.fc = nn.Linear(inner_dim, dim)
 
     @staticmethod
     def _apply_rotary_emb(
@@ -47,8 +48,42 @@ class Attention(nn.Module):
         k = torch.cat((kl, kr), dim=-1)
         return q, k
 
-    def _cross_attention(self) -> Tensor:
-        pass
+    @staticmethod
+    def _compute_input_mask(
+        b: int,
+        n: int,
+        k: Tensor,
+        mask: Optional[Tensor],
+        context: Optional[Tensor],
+        context_mask: Optional[Tensor],
+        device: str,
+    ) -> Optional[Tensor]:
+        if any(x is not None for x in (mask, context_mask)):
+            q_mask = (
+                mask if mask is not None else torch.ones((b, n), device=device).bool()
+            )
+            k_mask = q_mask if context is not None else context_mask
+            k_mask = (
+                torch.ones((b, k.shape[-2]), device=device).bool()
+                if k_mask is None
+                else k_mask
+            )
+            q_mask = rearrange(q_mask, "b i -> b () i ()")
+            k_mask = rearrange(k_mask, "b i -> b () () j")
+            return q_mask * k_mask
+        return
+
+    @staticmethod
+    def _apply_causal_mask(
+        energy: Tensor, mask: Tensor, mask_value: Tensor, device: str
+    ) -> Tensor:
+        i, j = energy.shape[-2:]
+        r = torch.arange(i, device=device)
+        mask = rearrange(r, "i -> () () i ()") < rearrange(r, "j -> () () () j")
+        mask = F.pad(mask, (j - i, 0), value=False)
+        energy.masked_fill_(mask, mask_value)
+        del mask
+        return energy
 
     def forward(
         self,
@@ -67,14 +102,25 @@ class Attention(nn.Module):
             k,
         )
 
-        input_mask = None
-        if any(x is not None for x in (mask, context_mask)):
-            q_mask = (
-                mask
-                if mask is not None
-                else lambda: torch.ones((b, n), device=device).bool()
-            )
-            pass
+        input_mask = self._compute_input_mask(
+            b, n, k, mask, context, context_mask, device
+        )
 
         # Compute the attention
-        energy = (q @ k.transpose(-2, -1)) * self.scale
+        energy = einsum("b h i d, b h j d -> b h i j", q, k) * self.scale
+        mask_value = -torch.finfo(energy.dtype).max
+
+        # Apply input mask
+        if input_mask is not None:
+            energy = energy.masked_fill_(~input_mask, mask_value)
+            del input_mask
+
+        if self.causal:
+            energy = self._apply_causal_mask(energy, mask, mask_value, device)
+
+        attn = self.attn_fn(energy, dim=-1)
+        attn = self.dropout(attn)
+        out = einsum("b h i j, b h j d -> b h i d", attn, v)
+        out = rearrange(out, "b h n d -> b n (h d)")
+        out = self.fc(out)
+        return out, attn