Aryan contri3

docs/overview/related_work/what_does_ivy_add.rst

@@ -105,6 +105,6 @@ Firstly, we are adhering to the `Array API Standard`_ defined by Quansight.
 In essence, they have written the standard and we have implemented it, which is pretty much as complementary as it gets.
 Similarly, OctoML makes it easy for anyone to *deploy* their model anywhere, while Ivy makes it easy for anyone to mix and match any code from any frameworks and versions to *train* their model anywhere.
 Again very complementary objectives.
-Finally, Modular will perhaps make it possible for developers to make changes at various levels of the stack when creating ML models using their "", and this would also be a great addition to the field.
+Finally, Modular will perhaps make it possible for developers to make changes at various levels of the stack when creating ML models using their own, and this would also be a great addition to the field.
 Compared to Modular which focuses on the lower levels of the stack, Ivy instead unifies the ML frameworks at the functional API level, enabling code conversions to and from the user-facing APIs themselves, without diving into any of the lower level details.
 All of these features are entirely complementary, and together would form a powerful suite of unifying tools for ML practitioners.

ivy/data_classes/array/experimental/manipulation.py

@@ -173,6 +173,11 @@ def vstack(
             ivy.array([[1, 2],
                        [5, 6],
                        [7, 8]])
+
+                    x = ivy.array([1, 2])
+                    y = [ivy.array([[5, 6]]), ivy.array([[7, 8]])]
+                    ivy.vstack((x, y))
+                    ivy.vstack((x, y, x, y))
         """
         if not isinstance(arrays, (list, tuple)):
             arrays = [arrays]

ivy/functional/backends/jax/general.py

@@ -120,6 +120,26 @@ def to_list(x: JaxArray, /) -> list:
     return _to_array(x).tolist()
 
 
+# ivy/utils/assertions.py
+
+def get_positive_axis_for_gather(axis, ndims):
+    if not isinstance(axis, int):
+        raise TypeError(f"{axis} must be an int; got {type(axis).__name__}")
+    if ndims is not None:
+        if 0 <= axis < ndims:
+            return axis
+        elif -ndims <= axis < 0:
+            return axis + ndims
+        else:
+            raise ValueError(f"{axis}={axis} out of bounds: "
+                             f"expected {-ndims}<={axis}<{ndims}")
+    elif axis < 0:
+        raise ValueError(f"{axis} may only be negative "
+                         f"if {ndims} is statically known.")
+    return axis
+
+# ivy/functional/backends/jax/general.py
+
 def gather(
     params: JaxArray,
     indices: JaxArray,
@@ -129,7 +149,7 @@ def gather(
     batch_dims: int = 0,
     out: Optional[JaxArray] = None,
 ) -> JaxArray:
-    axis = axis % len(params.shape)
+    axis = get_positive_axis_for_gather(axis, params.ndim)
     batch_dims = batch_dims % len(params.shape)
     ivy.utils.assertions.check_gather_input_valid(params, indices, axis, batch_dims)
     result = []