mlx - fix error with loading models with h5 and update mlx.core.while_loop (#20819)

* fix error with loading models with h5 and core while_loop

* adjust h5py import error

Author: acsweet

keras/src/backend/mlx/core.py

@@ -7,6 +7,11 @@
 from keras.src.backend.common.keras_tensor import KerasTensor
 from keras.src.backend.common.stateless_scope import StatelessScope
 
+try:
+    import h5py
+except ImportError:
+    h5py = None
+
 SUPPORTS_SPARSE_TENSORS = False
 
 MLX_DTYPES = {
@@ -55,6 +60,13 @@ def __array__(self, dtype=None):
         return value
 
 
+def _is_h5py_dataset(obj):
+    return (
+        type(obj).__module__.startswith("h5py.")
+        and type(obj).__name__ == "Dataset"
+    )
+
+
 def convert_to_tensor(x, dtype=None, sparse=None):
     if sparse:
         raise ValueError("`sparse=True` is not supported with mlx backend")
@@ -89,6 +101,14 @@ def to_scalar_list(x):
 
         return mx.array(to_scalar_list(x), dtype=mlx_dtype)
 
+    if _is_h5py_dataset(x):
+        if h5py is None:
+            raise ImportError(
+                "h5py must be installed in order to load HDF5 datasets."
+            )
+        # load h5py._hl.dataset.Dataset object with numpy
+        x = np.array(x)
+
     return mx.array(x, dtype=mlx_dtype)
 
 
@@ -279,18 +299,32 @@ def while_loop(
     loop_vars,
     maximum_iterations=None,
 ):
-    # TODO: How should we avoid evaluating cond when tracing?
     current_iter = 0
     iteration_check = (
         lambda iter: maximum_iterations is None or iter < maximum_iterations
     )
-    loop_vars = tuple([convert_to_tensor(v) for v in loop_vars])
-    while cond(*loop_vars) and iteration_check(current_iter):
-        loop_vars = body(*loop_vars)
-        if not isinstance(loop_vars, (list, tuple)):
-            loop_vars = (loop_vars,)
-        loop_vars = tuple(loop_vars)
+
+    is_sequence = isinstance(loop_vars, (tuple, list))
+
+    if is_sequence:
+        loop_vars = tuple(convert_to_tensor(v) for v in loop_vars)
+    else:
+        loop_vars = tree.map_structure(convert_to_tensor, loop_vars)
+
+    while (
+        cond(*loop_vars) if is_sequence else cond(loop_vars)
+    ) and iteration_check(current_iter):
+        new_vars = body(*loop_vars) if is_sequence else body(loop_vars)
+
+        if is_sequence:
+            if not isinstance(new_vars, (tuple, list)):
+                new_vars = (new_vars,)
+            loop_vars = tuple(convert_to_tensor(v) for v in new_vars)
+        else:
+            loop_vars = tree.map_structure(convert_to_tensor, new_vars)
+
         current_iter += 1
+
     return loop_vars