pnnx: support npy input tensors

[nihui]

No description provided.

[tencent-adm]

Thank you for your submission, we really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[Copilot]

Pull request overview

This PR adds first-class support for providing .npy files as example input tensors to pnnx, enabling conversions/tests to run with real tensor values (including non-native endianness and Fortran-ordered arrays) instead of synthetic/random inputs.

Changes:

• Add a .npy loader (load_numpy_file) and plumb input= / input2= CLI args through model loading and code generation.
• Update TorchScript/TNN loading paths to accept inferred input shapes/types from .npy and (for TorchScript) construct input tensors from the loaded payload.
• Add new unit tests (TorchScript/ONNX/TNN) and register them in CMake.

Reviewed changes

Copilot reviewed 18 out of 18 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
tools/pnnx/tests/test_pnnx_input_npy.py	New TorchScript test covering .npy inputs (incl. endian + Fortran order) for multi-output model.
tools/pnnx/tests/test_pnnx_input_npy_input2.py	New TorchScript test covering input + input2 with two example input sets.
tools/pnnx/tests/test_pnnx_input_npy_int64.py	New TorchScript test covering int64 .npy inputs (e.g., gather indices).
tools/pnnx/tests/test_pnnx_input_npy_tnn.py	New TNN test covering .npy input for a minimal ReLU network.
tools/pnnx/tests/onnx/test_pnnx_input_npy.py	New ONNX test covering .npy inputs (incl. endian + Fortran order).
tools/pnnx/tests/onnx/CMakeLists.txt	Registers the new ONNX .npy input test.
tools/pnnx/tests/CMakeLists.txt	Registers new .npy input tests (including conditional TNN test).
tools/pnnx/src/utils.h	Introduces NumpyArray + load_numpy_file() declaration.
tools/pnnx/src/utils.cpp	Implements .npy parsing, dtype/shape handling, endian swap, and Fortran→C reordering.
tools/pnnx/src/save_ncnn.h	Extends save_ncnn() signature to accept input .npy paths.
tools/pnnx/src/save_ncnn.cpp	Emits Python that can load provided .npy inputs for generated ncnn inference wrappers.
tools/pnnx/src/main.cpp	Adds input/input2 CLI params, parses .npy inputs, and wires paths/contents into downstream stages.
tools/pnnx/src/load_torchscript.h	Extends load_torchscript() signature to accept raw input payloads.
tools/pnnx/src/load_torchscript.cpp	Creates Torch tensors from .npy payload bytes (with validation) and adds bool handling.
tools/pnnx/src/load_tnn.h	Extends load_tnn() signature to accept user-provided input shapes/types.
tools/pnnx/src/load_tnn.cpp	Validates/overrides TNN input shape/type based on provided .npy/inputshape.
tools/pnnx/src/ir.h	Extends Graph::python() signature to accept input .npy paths.
tools/pnnx/src/ir.cpp	Emits Python that can load provided .npy inputs for generated pnnx inference wrappers.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[nihui]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Swish!

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Copilot]

Pull request overview

Copilot reviewed 12 out of 12 changed files in this pull request and generated 4 comments.

[Copilot]

Pull request overview

Copilot reviewed 12 out of 12 changed files in this pull request and generated no new comments.