python(api): add SessionOptions.add_external_initializers_from_files_… (#26012)

…in_memory (closes #25873)

### Description
Adds a Python binding to load external initializer files from in‑memory
buffers.
Mirrors existing C/C++ API and Node binding to enable full in‑memory
model loading.
Adds explicit type validation to avoid pybind dumping large raw bytes on
argument errors.



### Motivation and Context
Problem: Models that use external_data for initializers can’t be fully
loaded from bytes in Python because the weights are expected on disk.
Goal: Allow providing the external file contents directly from Python
memory (e.g., bytes, memoryview, numpy), eliminating filesystem
dependency and supporting serverless/remote asset scenarios.
Issue: Fixes #25873.

### Changes

#### New Python API on SessionOptions:
Name: add_external_initializers_from_files_in_memory
Signature: names: list[str], buffers: list[bytes-like], lengths:
list[int]
File: onnxruntime/python/onnxruntime_pybind_state.cc

#### Input validation to prevent noisy errors:
Validates top-level types are lists and that list lengths match.
Validates each name is str, each buffer supports the buffer protocol,
and each length is an int.
Raises clear RuntimeError messages instead of pybind11’s verbose dumps
for mismatched types.

#### Test added:
onnxruntime/test/python/onnxruntime_test_python.py
test_session_options_add_external_initializers_from_files_in_memory:
supplies “Pads_not_on_disk.bin” content from a numpy array’s bytes and
loads model_with_external_initializer_come_from_user.onnx without
touching the filesystem.


### Usage

Provide the external file name(s) as referenced by the model’s
external_data “location”, plus their bytes and lengths:
so.add_external_initializers_from_files_in_memory(["weights.bin"],
[weights_bytes], [len(weights_bytes)])
sess = onnxruntime.InferenceSession(model_bytes, sess_options=so)

---------

Signed-off-by: Jonah Bernard <[email protected]>
Co-authored-by: Jonah Bernard <[email protected]>

Author: Jonahcb

onnxruntime/python/onnxruntime_pybind_state.cc

@@ -2300,7 +2300,50 @@ Applies to session load, initialization, etc. Default is 0.)pbdoc")
         ORT_UNUSED_PARAMETER(ort_values);
         ORT_THROW("External initializers are not supported in this build.");
 #endif
-      });
+      })
+      .def("add_external_initializers_from_files_in_memory", [](PySessionOptions* options, std::vector<std::string> names, std::vector<py::buffer> buffers, std::vector<size_t> lengths) -> void {
+#if !defined(ORT_MINIMAL_BUILD) && !defined(DISABLE_EXTERNAL_INITIALIZERS)
+        const auto num = names.size();
+        ORT_ENFORCE(num == buffers.size() && num == lengths.size(),
+                    "add_external_initializers_from_files_in_memory: expecting 'names', 'buffers' and 'lengths' to have equal length");
+
+        InlinedVector<PathString> file_names;
+        InlinedVector<std::pair<char*, const size_t>> files_buffers;
+        file_names.reserve(num);
+        files_buffers.reserve(num);
+
+        for (size_t i = 0; i < num; ++i) {
+          // Convert name and buffer using pybind-provided conversions
+          file_names.emplace_back(ToPathString(names[i]));
+
+          // buffers[i] is a py::buffer; request() retrieves pointer without copying
+          py::buffer_info info = buffers[i].request();
+          char* data_ptr = static_cast<char*>(info.ptr);
+
+          files_buffers.emplace_back(std::make_pair(data_ptr, lengths[i]));
+        }
+
+        ORT_THROW_IF_ERROR(options->value.AddExternalInitializersFromFilesInMemory(file_names, files_buffers));
+#else
+        ORT_UNUSED_PARAMETER(options);
+        ORT_UNUSED_PARAMETER(names);
+        ORT_UNUSED_PARAMETER(buffers);
+        ORT_UNUSED_PARAMETER(lengths);
+        ORT_THROW("External initializers are not supported in this build.");
+#endif
+      },
+           R"pbdoc(
+Provide external initializer file contents from memory.
+
+Args:
+  names: sequence[str] of external file names (as referenced by the model's external_data locations).
+  buffers: sequence[bytes-like] objects exposing the buffer protocol (e.g., bytes, bytearray, memoryview, numpy uint8 array) containing the corresponding file contents.
+  lengths: sequence[int] sizes in bytes for each buffer.
+
+Notes:
+  - Keep the provided buffers alive until after session creation completes. ONNX Runtime copies needed data during session creation.
+  - The bytestream must match the external file layout expected by the model (raw tensor bytes at the specified offsets).
+)pbdoc");
 
   py::class_<RunOptions>(m, "RunOptions", R"pbdoc(Configuration information for a single Run.)pbdoc")
       .def(py::init())

onnxruntime/test/python/onnxruntime_test_python.py

@@ -1305,6 +1305,25 @@ def test_session_options_add_external_initializers(self):
             providers=["CPUExecutionProvider"],
         )
 
+    def test_session_options_add_external_initializers_from_files_in_memory(self):
+        # Provide external initializer file content directly from memory
+        # The model references an external file named "Pads_not_on_disk.bin" for the initializer
+        pads_bytes = np.array([0, 0, 1, 1], dtype=np.int64).tobytes()
+
+        so = onnxrt.SessionOptions()
+        so.add_external_initializers_from_files_in_memory(
+            ["Pads_not_on_disk.bin"],
+            [pads_bytes],
+            [len(pads_bytes)],
+        )
+
+        # This should not throw
+        onnxrt.InferenceSession(
+            get_name("model_with_external_initializer_come_from_user.onnx"),
+            sess_options=so,
+            providers=["CPUExecutionProvider"],
+        )
+
     def test_register_custom_ops_library(self):
         if sys.platform.startswith("win"):
             shared_library = os.path.abspath("custom_op_library.dll")