onnxscript 0.7.1.dev20260604__tar.gz → 0.7.1.dev20260610__tar.gz

{onnxscript-0.7.1.dev20260604/onnxscript.egg-info → onnxscript-0.7.1.dev20260610}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: onnxscript
-Version: 0.7.1.dev20260604
+Version: 0.7.1.dev20260610
 Summary: Naturally author ONNX functions and models using a subset of Python
 Author-email: Microsoft Corporation <onnx@microsoft.com>
 License: MIT License
@@ -27,7 +27,7 @@ License: MIT License
         
 Project-URL: Homepage, https://microsoft.github.io/onnxscript/
 Project-URL: Repository, https://github.com/microsoft/onnxscript
-Project-URL: Commit, https://github.com/microsoft/onnxscript/tree/e0d3edc6eb902a2dcc07e618b055234a2ef66665
+Project-URL: Commit, https://github.com/microsoft/onnxscript/tree/c72e810b5806f4ace9d747db8ec798387899d04a
 Classifier: Development Status :: 4 - Beta
 Classifier: Environment :: Console
 Classifier: Intended Audience :: Developers

{onnxscript-0.7.1.dev20260604 → onnxscript-0.7.1.dev20260610}/onnxscript/function_libs/torch_lib/ops/core.py

@@ -4348,6 +4348,11 @@ def aten_gru(
         # Extract hidden_size from hx shape: [num_layers * num_directions, batch, hidden_size]
         hidden_size_attr = hx.shape[2]
 
+        # linear_before_reset=1 matches PyTorch's GRU formulation where the linear
+        # transformation is applied before multiplying by the reset gate:
+        #   ht = g(Xt*(Wh^T) + rt (.) (Ht-1*(Rh^T) + Rbh) + Wbh)
+        # The ONNX default (linear_before_reset=0) uses a different equation and
+        # would produce numerically incorrect results.
         if B is not None:
             Y, Y_h = op.GRU(
                 current_input,
@@ -4357,6 +4362,7 @@ def aten_gru(
                 initial_h=layer_h,
                 direction=direction,
                 hidden_size=hidden_size_attr,
+                linear_before_reset=1,
             )
         else:
             Y, Y_h = op.GRU(
@@ -4366,6 +4372,7 @@ def aten_gru(
                 initial_h=layer_h,
                 direction=direction,
                 hidden_size=hidden_size_attr,
+                linear_before_reset=1,
             )
 
         # Y shape: [seq_length, num_directions, batch_size, hidden_size]
@@ -6081,10 +6088,12 @@ def aten_masked_fill(self: TTensor, mask: BOOL, value: TTensor) -> TTensor:
 def aten_masked_scatter(self: TTensor, mask: TTensor, source: TTensor) -> TTensor:
     """masked_scatter(Tensor self, Tensor mask, Tensor source) -> Tensor"""
 
-    if len(mask.shape) < len(self.shape):
-        mask = op.Expand(mask, op.Shape(self))
-    else:
-        self = op.Expand(self, op.Shape(mask))
+    # Broadcast self and mask to their common shape so NonZero enumerates every
+    # masked element. The previous rank-only check missed same-rank broadcasting
+    # (e.g. mask (1, S, 1) vs self (1, S, D)): it left mask un-expanded, so only a
+    # subset of masked positions were scattered (pytorch/pytorch#186146).
+    self = op.Expand(self, op.Shape(mask))
+    mask = op.Expand(mask, op.Shape(self))
     index = op.Transpose(op.NonZero(mask), perm=[1, 0])
 
     # NOTE: source can have more elements than needed.

{onnxscript-0.7.1.dev20260604 → onnxscript-0.7.1.dev20260610}/onnxscript/function_libs/torch_lib/ops/nn.py

@@ -2320,7 +2320,7 @@ def aten_unflatten_dense_tensors(
 
 
 def _get_upsample_align_corners_mode(align_corners: bool) -> str:
-    return "align_corners" if align_corners else "pytorch_half_pixel"
+    return "align_corners" if align_corners else "half_pixel"
 
 
 def _aten_upsample_output_size(

{onnxscript-0.7.1.dev20260604 → onnxscript-0.7.1.dev20260610}/onnxscript/optimizer/_constant_folding.py

@@ -22,6 +22,7 @@ import numpy as np
 import onnx
 import onnx.reference.ops
 import onnx_ir as ir
+import onnx_ir.passes.common as ir_passes_common
 
 import onnxscript.utils.utils as utils
 from onnxscript._internal.tape_builder import BuilderBase, TapeBuilder
@@ -1394,6 +1395,11 @@ class FoldConstantsPass(ir.passes.InPlacePass):
         for function in model.functions.values():
             # TODO(rama): Should we specialize functions?
             self.visit_function(function)
+        if self._modified:
+            # TapeBuilder may create values with names that clash with existing graph
+            # values when nodes are inserted via replace_nodes_and_values.
+            # NameFixPass ensures all value names are unique before returning.
+            ir_passes_common.NameFixPass()(model)
         return FoldConstantsResult(model, self._modified, self._state.symbolic_value_map)
 
 

{onnxscript-0.7.1.dev20260604 → onnxscript-0.7.1.dev20260610}/onnxscript/rewriter/_rewrite_rule.py

@@ -13,6 +13,8 @@ from typing import (
     TypeVar,
 )
 
+import onnx_ir.passes.common as ir_passes_common
+
 import onnxscript.optimizer
 import onnxscript.rewriter._basics as _basics
 import onnxscript.rewriter._context as _context
@@ -835,6 +837,11 @@ class RewriteRuleSet:
             )
         if self.remove_unused_nodes:
             onnxscript.optimizer.remove_unused_nodes(model)
+        if count > 0:
+            # TapeBuilder may create values with names that clash with existing graph
+            # values when nodes are inserted via replace_nodes_and_values.
+            # NameFixPass ensures all value names are unique before returning.
+            ir_passes_common.NameFixPass()(model)
         return count
 
     def __iter__(self):

{onnxscript-0.7.1.dev20260604 → onnxscript-0.7.1.dev20260610}/onnxscript/version_converter/_version_converter.py

@@ -10,6 +10,7 @@ import logging
 from typing import Callable, Sequence, Union
 
 import onnx_ir.convenience as ir_convenience
+import onnx_ir.passes.common as ir_passes_common
 
 import onnxscript.utils.metadata_merger as metadata_merger
 from onnxscript import ir
@@ -239,6 +240,7 @@ def groupnormalization_20_21(node: ir.Node, op):
 class _VersionConverter:
     def __init__(self, target_version: int):
         self._target_version = target_version
+        self._modified: bool = False
         # Default metadata merger: no merging should be needed; keep the first value.
         self._default_metadata_merger: metadata_merger.MetadataMerger = (
             metadata_merger.MetadataMerger(
@@ -269,6 +271,7 @@ class _VersionConverter:
         ir_convenience.replace_nodes_and_values(
             root, node, [node], replacement.new_nodes, node.outputs, replacement.new_outputs
         )
+        self._modified = True
 
     def visit_attribute(self, attr: ir.Attr) -> None:
         if attr.is_ref():
@@ -341,6 +344,11 @@ class _VersionConverter:
             self.visit_graph_or_function(function)
             _set_onnx_opset_version(function, self._target_version)
         _set_onnx_opset_version(model, self._target_version)
+        if self._modified:
+            # TapeBuilder may create values with names that clash with existing graph
+            # values when nodes are inserted via replace_nodes_and_values.
+            # NameFixPass ensures all value names are unique before returning.
+            ir_passes_common.NameFixPass()(model)
 
 
 def convert_version(model: ir.Model, target_version: int) -> None:

{onnxscript-0.7.1.dev20260604 → onnxscript-0.7.1.dev20260610/onnxscript.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: onnxscript
-Version: 0.7.1.dev20260604
+Version: 0.7.1.dev20260610
 Summary: Naturally author ONNX functions and models using a subset of Python
 Author-email: Microsoft Corporation <onnx@microsoft.com>
 License: MIT License
@@ -27,7 +27,7 @@ License: MIT License
         
 Project-URL: Homepage, https://microsoft.github.io/onnxscript/
 Project-URL: Repository, https://github.com/microsoft/onnxscript
-Project-URL: Commit, https://github.com/microsoft/onnxscript/tree/e0d3edc6eb902a2dcc07e618b055234a2ef66665
+Project-URL: Commit, https://github.com/microsoft/onnxscript/tree/c72e810b5806f4ace9d747db8ec798387899d04a
 Classifier: Development Status :: 4 - Beta
 Classifier: Environment :: Console
 Classifier: Intended Audience :: Developers