PyPI - broccoli-ml - Versions diffs - 1.2.0__tar.gz → 1.2.1__tar.gz - Mend

broccoli-ml 1.2.0tar.gz → 1.2.1tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (17) hide show

{broccoli_ml-1.2.0 → broccoli_ml-1.2.1}/PKG-INFO RENAMED Viewed

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: broccoli-ml
-Version: 1.2.0
+Version: 1.2.1
 Summary: Some useful Pytorch models, circa 2025
 License: MIT
 Author: Nicholas Bailey

{broccoli_ml-1.2.0 → broccoli_ml-1.2.1}/broccoli/transformer.py RENAMED Viewed

@@ -219,7 +219,8 @@ class MHAttention(nn.Module):
         if self.scaling == "sqrtd":
             qk_scores /= math.sqrt(self.head_dim)
         elif self.scaling == "d":
-            qk_scores /= self.head_dim
+            # for backwards compatibility, per https://github.com/microsoft/mup
+            qk_scores *= 8 / self.head_dim
         else:
             raise ValueError('`scaling` argument to MHAttention must be "d" or "sqrtd"')

{broccoli_ml-1.2.0 → broccoli_ml-1.2.1}/pyproject.toml RENAMED Viewed

@@ -1,6 +1,6 @@
 [project]
 name = "broccoli-ml"
-version = "1.2.0"
+version = "1.2.1"
 description = "Some useful Pytorch models, circa 2025"
 authors = [
     {name = "Nicholas Bailey"}