rxnn 0.1.55__tar.gz → 0.1.56__tar.gz

{rxnn-0.1.55 → rxnn-0.1.56}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: rxnn
-Version: 0.1.55
+Version: 0.1.56
 Summary: RxNN: Reactive Neural Networks Platform
 License: Apache-2.0
 Keywords: deep-learning,ai,machine-learning

{rxnn-0.1.55 → rxnn-0.1.56}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "poetry.core.masonry.api"
 
 [tool.poetry]
 name = "rxnn"
-version = "0.1.55"
+version = "0.1.56"
 description = "RxNN: Reactive Neural Networks Platform"
 
 license = "Apache-2.0"

{rxnn-0.1.55 → rxnn-0.1.56}/src/rxnn/experimental/attention.py

@@ -283,12 +283,12 @@ class SparseQueryAttention(MultiHeadAttention):
         """Override query, key, and value projections for GQA case - split data into heads and groups"""
         head_dim = d // self.num_heads
         if not self.rel_embed:
-            q = self.q_proj(query).view(b, t, self.num_query_heads, head_dim).transpose(1, 2)
+            q = self.q_proj(query).view(b, t, self.num_query_groups, head_dim).transpose(1, 2)
             k = self.k_proj(key).view(b, -1, self.num_groups, head_dim).transpose(1, 2)
             v = self.v_proj(value).view(b, -1, self.num_groups, head_dim).transpose(1, 2)
         else:
             group_heads = self.num_heads // self.num_groups
-            query_heads = self.num_query_heads // self.num_query_groups
+            query_heads = self.num_heads // self.num_query_groups
             # Process Q
             q = self.q_proj(query).view(b, -1, self.num_query_groups, head_dim).transpose(1, 2)  # (B, Q_G, T, head_dim)