complete attention annotations

lucidrains · May 11, 2024 · 3de453f · 3de453f
1 parent 682b1bc
commit 3de453f
Showing 1 changed file with 9 additions and 7 deletions.
diff --git a/alphafold3_pytorch/attention.py b/alphafold3_pytorch/attention.py
@@ -84,10 +84,11 @@ def __init__(
 
     def forward(
         self,
-        seq,
-        mask = None,
-        context = None
-    ):
+        seq: Float['b i d'],
+        mask: Bool['b n']| None = None,
+        context: Float['b j d'] | None = None
+    ) -> Float['b i d']:
+
         q = self.to_q(seq)
 
         context_seq = default(context, seq)
@@ -112,7 +113,7 @@ def __init__(
         self,
         dropout = 0.,
         flash = False,
-        scale = None,
+        scale: float | None = None,
         attn_config: Config = Config(True, True, True)
     ):
         super().__init__()
@@ -138,7 +139,8 @@ def flash_attn(
         k: Float['b h j d'],
         v: Float['b h j d'],
         mask: Bool['b j'] | None = None
-    ):
+    ) -> Float['b h i d']:
+
         _, heads, seq_len, _ = q.shape
 
         attn_mask = None
@@ -162,7 +164,7 @@ def forward(
         k: Float['b h j d'],
         v: Float['b h j d'],
         mask: Bool['b j'] | None = None
-    ):
+    ) -> Float['b h i d']:
 
         if self.flash:
             return self.flash_attn(q, k, v, mask = mask)