haliax 1.4.dev291__tar.gz → 1.4.dev293__tar.gz

{haliax-1.4.dev291 → haliax-1.4.dev293}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: haliax
-Version: 1.4.dev291
+Version: 1.4.dev293
 Summary: Named Tensors for Legible Deep Learning in JAX
 Project-URL: Homepage, https://github.com/stanford-crfm/haliax
 Project-URL: Bug Tracker, https://github.com/stanford-crfm/haliax/issues/

{haliax-1.4.dev291 → haliax-1.4.dev293}/docs/indexing.md

@@ -153,9 +153,11 @@ It's not a huge improvement, but it's a bit more convenient.
 ## Advanced Indexing
 
 NumPy's [Advanced Indexing](https://numpy.org/doc/stable/user/basics.indexing.html#advanced-indexing) is supported, though we use named arrays for the indices instead of normal arrays.
-In NumPy, the indexed arrays much be broadcastable to the same shape.Advanced indexing in Haliax is similar, except that
-they follow Haliax's broadcasting rules, meaning that the axis names determine broadcasting.Axes with the same name must
-have the same size.
+(Though, as noted above, you can use 1-D JAX arrays of integers as well.)
+In NumPy, the indexed arrays must be broadcastable to the same shape. Advanced indexing in Haliax is similar,
+except that it follows Haliax's broadcasting rules, meaning that shared names are broadcasted together,
+while non-shared names are treated as separate axes and are cross-producted.
+In particular, axes with the same name must have the same size.
 
 ```python
 import haliax as hax

haliax-1.4.dev293/src/haliax/__about__.py

@@ -0,0 +1 @@
+__version__ = "1.4.dev293"

{haliax-1.4.dev291 → haliax-1.4.dev293}/src/haliax/partitioning.py

@@ -38,6 +38,7 @@ class ResourceAxis(StringHolderEnum):
 
     MODEL = "model"
     DATA = "data"
+    REPLICA = "replica"
 
 
 class _ResourceMappingHolder:

haliax-1.4.dev291/src/haliax/__about__.py

@@ -1 +0,0 @@
-__version__ = "1.4.dev291"