complete hyper connected alphafold3

Author: lucidrains

alphafold3_pytorch/alphafold3.py

@@ -119,7 +119,7 @@
 
 from colt5_attention import ConditionalRoutedAttention
 
-from hyper_connections import HyperConnections
+from hyper_connections.hyper_connections_with_multi_input_streams import HyperConnections
 
 # other external libs
 
@@ -995,8 +995,8 @@ def __init__(
     @typecheck
     def forward(
         self,
-        *,
         pairwise_repr: Float['b n n d'],
+        *,
         mask: Bool['b n'] | None = None,
         value_residuals: tuple[Tensor, Tensor] | None = None,
         return_values = False,
@@ -1470,8 +1470,8 @@ def __init__(
             single_transition = Transition(dim = dim_single)
 
             layers.append(ModuleList([
-                pairwise_block,
-                init_hyper_conn(dim = dim_single, branch = single_pre_ln(pair_bias_attn)),
+                init_hyper_conn(dim = dim_pairwise, branch = pairwise_block),
+                init_hyper_conn(dim = dim_single, additional_input_paths = [('pairwise_repr', dim_pairwise)], branch = single_pre_ln(pair_bias_attn)),
                 init_hyper_conn(dim = dim_single, branch = single_pre_ln(single_transition)),
             ]))
 
@@ -1508,6 +1508,7 @@ def to_layers(
     ) -> Tuple[Float['b n ds'], Float['b n n dp']]:
 
         single_repr = self.expand_streams(single_repr)
+        pairwise_repr = self.expand_streams(pairwise_repr)
 
         for _ in range(self.recurrent_depth):
 
@@ -1520,7 +1521,7 @@ def to_layers(
                 single_transition
             ) in self.layers:
 
-                pairwise_repr, pairwise_attn_values = pairwise_block(pairwise_repr = pairwise_repr, mask = mask, value_residuals = pairwise_value_residuals, return_values = True)
+                pairwise_repr, pairwise_attn_values = pairwise_block(pairwise_repr, mask = mask, value_residuals = pairwise_value_residuals, return_values = True)
 
                 single_repr, attn_values = pair_bias_attn(single_repr, pairwise_repr = pairwise_repr, mask = mask, return_values = True, value_residual = value_residual)
 
@@ -1531,6 +1532,7 @@ def to_layers(
                 single_repr = single_transition(single_repr)
 
         single_repr = self.reduce_streams(single_repr)
+        pairwise_repr = self.reduce_streams(pairwise_repr)
 
         return single_repr, pairwise_repr
 
@@ -1548,7 +1550,7 @@ def pairwise_block_wrapper(layer):
             @wraps(layer)
             def inner(inputs, *args, **kwargs):
                 single_repr, pairwise_repr, mask, maybe_value_residual, maybe_pairwise_value_residuals = inputs
-                pairwise_repr, pairwise_attn_values = layer(pairwise_repr = pairwise_repr, mask = mask, value_residuals = maybe_pairwise_value_residuals, return_values = True)
+                pairwise_repr, pairwise_attn_values = layer(pairwise_repr, mask = mask, value_residuals = maybe_pairwise_value_residuals, return_values = True)
 
                 if self.add_value_residual:
                     maybe_pairwise_value_residuals = default(maybe_pairwise_value_residuals, pairwise_attn_values)
@@ -1589,6 +1591,7 @@ def inner(inputs, *args, **kwargs):
             wrapped_layers.append(single_transition_wrapper(single_transition))
 
         single_repr = self.expand_streams(single_repr)
+        pairwise_repr = self.expand_streams(pairwise_repr)
 
         for _ in range(self.recurrent_depth):
             inputs = (single_repr, pairwise_repr, mask, None, None)
@@ -1599,6 +1602,7 @@ def inner(inputs, *args, **kwargs):
             single_repr, pairwise_repr, *_ = inputs
 
         single_repr = self.reduce_streams(single_repr)
+        pairwise_repr = self.reduce_streams(pairwise_repr)
 
         return single_repr, pairwise_repr
 

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "alphafold3-pytorch"
-version = "0.7.7"
+version = "0.7.8"
 description = "Alphafold 3 - Pytorch"
 authors = [
     { name = "Phil Wang", email = "[email protected]" },
@@ -41,7 +41,7 @@ dependencies = [
     "fair-esm",
     "fastapi",
     "frame-averaging-pytorch>=0.0.18",
-    "hyper-connections>=0.0.21",
+    "hyper-connections>=0.0.23",
     "gradio",
     "gradio_molecule3d",
     "huggingface_hub>=0.21.4",