tensorstudio 1.0.0rc2__tar.gz → 1.1.0__tar.gz

{tensorstudio-1.0.0rc2 → tensorstudio-1.1.0}/.github/workflows/ci.yml

@@ -30,6 +30,10 @@ jobs:
         run: mypy --python-version ${{ matrix.python-version }} python/tensorstudio
       - name: Tests
         run: pytest -q
+      - name: Runner scripts
+        run: |
+          python test_all.py --help
+          python benchmark_all.py --help
       - name: Example
         run: python examples/basic_tensor_ops.py
 
@@ -59,6 +63,10 @@ jobs:
         run: mypy --python-version ${{ matrix.python-version }} python/tensorstudio
       - name: Tests
         run: pytest -q
+      - name: Runner scripts
+        run: |
+          python test_all.py --help
+          python benchmark_all.py --help
       - name: Example
         run: python examples/basic_tensor_ops.py
 
@@ -88,5 +96,9 @@ jobs:
         run: mypy --python-version ${{ matrix.python-version }} python/tensorstudio
       - name: Tests
         run: pytest -q
+      - name: Runner scripts
+        run: |
+          python test_all.py --help
+          python benchmark_all.py --help
       - name: Example
         run: python examples/basic_tensor_ops.py

{tensorstudio-1.0.0rc2 → tensorstudio-1.1.0}/.github/workflows/publish-testpypi.yml

@@ -106,3 +106,4 @@ jobs:
         uses: pypa/gh-action-pypi-publish@release/v1
         with:
           repository-url: https://test.pypi.org/legacy/
+          skip-existing: true

{tensorstudio-1.0.0rc2 → tensorstudio-1.1.0}/.github/workflows/publish.yml

@@ -4,9 +4,6 @@ on:
   workflow_dispatch:
   release:
     types: [published]
-  push:
-    tags:
-      - "v1.0.0"
 
 permissions:
   contents: read
@@ -97,11 +94,26 @@ jobs:
     needs: [windows_wheels, linux_wheels, macos_wheels, build_sdist]
     runs-on: ubuntu-latest
     environment: pypi
+    env:
+      HAS_PYPI_TOKEN: ${{ secrets.PYPI_TOKEN != '' }}
     steps:
       - uses: actions/download-artifact@v4
         with:
           pattern: dist-*
           path: dist
           merge-multiple: true
-      - name: Publish to PyPI
+      - name: Verify artifacts
+        run: |
+          python -m pip install -U twine
+          python -m twine check dist/*
+      - name: Publish to PyPI with configured API token
+        if: env.HAS_PYPI_TOKEN == 'true'
         uses: pypa/gh-action-pypi-publish@release/v1
+        with:
+          password: ${{ secrets.PYPI_TOKEN }}
+          skip-existing: true
+      - name: Publish to PyPI with Trusted Publishing
+        if: env.HAS_PYPI_TOKEN != 'true'
+        uses: pypa/gh-action-pypi-publish@release/v1
+        with:
+          skip-existing: true

tensorstudio-1.1.0/CHANGELOG.md

@@ -0,0 +1,89 @@
+# Changelog
+
+## Unreleased
+
+## 1.1.0 - 2026-07-07
+
+- Added native CPU NCHW `conv2d` with stride, padding, dilation, optional bias,
+  and reverse-mode gradients for input, weight, and bias.
+- Added `tensorstudio.conv2d`, `tensorstudio.ops.conv2d`,
+  `tensorstudio.nn.functional.conv2d`, and `tensorstudio.nn.Conv2d`.
+- Added convolution tests, finite-difference autograd coverage, and a
+  standalone convolution benchmark script.
+- Added convolution rows to the full benchmark report and updated docs to mark
+  convolution as partially implemented rather than missing.
+- Added native CPU NCHW `max_pool2d` and `avg_pool2d` operations with autograd,
+  Python top-level and functional APIs, `nn.MaxPool2d`, `nn.AvgPool2d`, tests,
+  docs, and benchmark coverage.
+- Added native single-axis `sum`, `mean`, `max`, and `min` reductions with
+  `keepdims` and reverse-mode gradients.
+- Added a typed contiguous 2D fast path for common axis `sum`/`mean`
+  reductions.
+- Added `nn.softmax`, `nn.log_softmax`, `nn.cross_entropy`, and
+  `nn.CrossEntropyLoss` for small multiclass classification workloads.
+- Added native `astype`, `concat`, and `stack` operations with Python wrappers,
+  Tensor methods for casting, and autograd support for floating tensors.
+
+## 1.0.1
+
+- Added `test_all.py` for one-command local release checks covering lint,
+  typing, tests, examples, docs, build, and package metadata validation.
+- Added `benchmark_all.py` for one-command full benchmark report generation.
+- Added explicit benchmark win columns for NumPy, TensorFlow, PyTorch, and JAX,
+  plus fastest-library reporting for each benchmark case.
+- Hardened release publishing triggers so future `v*` tags are eligible for
+  the publish workflow instead of only `v1.0.0`.
+- Updated docs to describe benchmark interpretation, local release checks, and
+  site navigation for benchmark, hardware, roadmap, and usage pages.
+
+## 1.0.0
+
+- Promoted the CPU-only stable API foundation from release candidate status.
+- Added direct C++ buffer copies for NumPy import/export paths.
+- Added faster typed tensor construction, flat-vector export, contiguous tensor
+  copy, and floating-point optimizer update paths.
+- Added a `float32` contiguous matmul fast path that avoids double accumulation
+  and per-element dtype conversion for the common training dtype.
+- Added typed contiguous reduction kernels for `sum` and `mean`.
+- Expanded benchmark reports with framework version metadata and PyTorch
+  win/loss summaries.
+- Updated documentation to describe honest `1.0.0` performance boundaries.
+
+## 1.0.0rc2
+
+- Added `*_like` tensor creation helpers for zero, one, full, empty, uniform
+  random, and normal random tensors.
+- Added `nn.Identity`, `nn.LeakyReLU`, `nn.Softplus`,
+  `nn.BCEWithLogitsLoss`, `nn.HuberLoss`, and matching functional helpers.
+- Expanded `nn.Module` with child/module introspection, trainable parameter
+  filtering, freeze/unfreeze helpers, parameter counting, recursive `apply`,
+  and richer `repr` details.
+- Added optimizer utilities for gradient clipping and lightweight
+  `StepLR`/`ExponentialLR` learning-rate schedulers.
+- Added `len(DataLoader)` batch counting with `drop_last` support.
+- Added typed C++ fast paths for contiguous elementwise/unary kernels and a
+  contiguous matmul kernel that avoids per-element Tensor method dispatch.
+- Bumped the next release candidate after public `1.0.0rc1` artifacts.
+
+## 1.0.0rc1
+
+- Release-candidate hardening toward a CPU-only v1.0.0 API foundation.
+- Added additional tensor creation helpers, math operations, comparisons,
+  no-grad mode, state APIs, data utilities, and Windows-first CI/release docs.
+- This is not final v1.0.0; remaining platform checklist items must pass before
+  a stable release.
+
+## 0.1.1
+
+- Publishable wheel release for CPython 3.10-3.13.
+- Fixed MSVC build portability by removing non-portable `ssize_t` usage.
+- Updated wheel CI to build 64-bit Windows wheels and modern Linux wheels.
+- Expanded documentation with MkDocs navigation and detailed usage guides.
+
+## 0.1.0
+
+- Initial experimental release.
+- C++20 CPU tensor core with Python bindings.
+- Basic broadcasting, matrix multiplication, reductions, activations, and views.
+- Reverse-mode autograd for the initial operation set.
+- Python `nn`, `optim`, serialization, examples, tests, docs, and CI workflows.

{tensorstudio-1.0.0rc2 → tensorstudio-1.1.0}/CMakeLists.txt

@@ -1,7 +1,9 @@
 cmake_minimum_required(VERSION 3.18)
 
-project(tensorstudio VERSION 1.0.0 LANGUAGES CXX)
+project(tensorstudio VERSION 1.0.1 LANGUAGES CXX)
 
+find_package(Python COMPONENTS Interpreter Development.Module REQUIRED)
+set(PYBIND11_FINDPYTHON ON)
 find_package(pybind11 CONFIG REQUIRED)
 
 set(TENSORSTUDIO_CORE_SOURCES

{tensorstudio-1.0.0rc2 → tensorstudio-1.1.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: tensorstudio
-Version: 1.0.0rc2
+Version: 1.1.0
 Summary: TensorStudio is a compact C++ tensor and autograd engine with a Python API for learning, experimentation, and lightweight ML workloads.
 Keywords: tensor,autograd,machine-learning,cpp,pybind11
 Author: TensorStudio contributors
@@ -65,14 +65,12 @@ Description-Content-Type: text/markdown
 TensorStudio is a compact C++ tensor and autograd engine with a Python API for
 learning, experimentation, and lightweight ML workloads.
 
-TensorStudio `1.0.0rc2` is a release candidate for a CPU-only stable API
-foundation. It is eager-only, intentionally small, and not a replacement for
-mature ML frameworks. The final `1.0.0` version should only be tagged after the
-Windows, Linux, and macOS release checklist passes.
+TensorStudio `1.1.0` is a CPU-only stable API foundation. It is eager-only,
+intentionally small, and not a replacement for mature ML frameworks.
 
 ## Install
 
-From PyPI, once release-candidate wheels are published:
+From PyPI:
 
 ```bash
 python -m pip install tensorstudio
@@ -154,6 +152,8 @@ d = ts.linspace(0.0, 1.0, 5)
 
 print(a.shape, a.strides, a.device, a.is_contiguous)
 print((b.clamp(0.2, 0.8) + 1).mean().item())
+print(b.sum(axis=1).tolist())
+print(ts.concat([b, b], axis=0).shape, b.astype("float64").dtype)
 print(c.tolist(), d.tolist())
 print(ts.zeros_like(b).shape, ts.randn_like(b, seed=11).dtype)
 ```
@@ -219,8 +219,8 @@ for features, targets in loader:
     print(features.shape, targets.shape)
 ```
 
-The v1 release candidate DataLoader is intentionally single-process so it works
-cleanly on Windows without multiprocessing setup.
+The v1 DataLoader is intentionally single-process so it works cleanly on
+Windows without multiprocessing setup.
 
 ## Performance
 
@@ -229,34 +229,45 @@ performance is still experimental. Benchmarks live in `benchmarks/` and can be
 run locally:
 
 ```bash
+python benchmark_all.py
 python benchmarks/benchmark_report.py
 ```
 
-On one Windows CPython 3.10 run for `1.0.0rc2`, TensorStudio beat NumPy on 11
-small activation/reduction benchmark cases and lost on 78 NumPy-comparable
-cases. The strongest local wins were small `sigmoid`, `sum`, and `mean` cases;
-medium elementwise operations and matrix multiplication remain slower than
-NumPy. The C++ contiguous matmul fast path is much faster than the original
-naive storage-access loop, but it is not a BLAS replacement. See
-`benchmarks/results.md` for the full table, platform details, and exact timings.
+`benchmark_all.py` writes `benchmarks/results.md` and includes explicit win
+columns for NumPy, TensorFlow, PyTorch, and JAX when those libraries are
+available locally.
+
+On one Windows CPython 3.10 development run reporting `1.1.0`, TensorStudio
+beat NumPy on 23 small operation benchmark cases and lost on 80
+NumPy-comparable cases. Against PyTorch CPU `2.12.1+cpu`, TensorStudio won 74
+local cases and lost 34. The strongest local wins were small eager operations,
+small contiguous axis reductions, and the simple NumPy convolution/pooling
+references where framework dispatch or Python loops dominate; larger matrix
+multiplication, PyTorch convolution and pooling, larger axis reductions, larger
+transcendental activations, and larger autograd workloads remain faster in
+PyTorch and NumPy.
+See `benchmarks/results.md` for the full table, platform details, and exact
+timings.
 
 Snapshot from that local run:
 
-| operation | shape | TensorStudio median | NumPy median | speedup | result |
-|---|---:|---:|---:|---:|---|
-| `sigmoid` | `(1,)` | 0.0016 ms | 0.0036 ms | 2.2170x | win |
-| `mean` | `(1,)` | 0.0014 ms | 0.0081 ms | 5.8562x | win |
-| `mean` | `(8,)` | 0.0016 ms | 0.0079 ms | 5.0005x | win |
-| `add` | `(16384,)` | 0.0100 ms | 0.0037 ms | 0.3682x | loss |
-| `matmul` | `(256, 256)` | 4.1393 ms | 0.4222 ms | 0.1020x | loss |
+| operation | shape | TensorStudio | NumPy | PyTorch CPU | TS vs NumPy | TS vs PyTorch |
+|---|---:|---:|---:|---:|---:|---:|
+| `sigmoid` | `(32,)` | 0.0017 ms | 0.0036 ms | 0.0580 ms | 2.1201x | 33.8536x |
+| `mean` | `(32,)` | 0.0018 ms | 0.0078 ms | 0.0127 ms | 4.2927x | 7.0047x |
+| `sum_axis1` | `(16, 16)` | 0.0021 ms | 0.0029 ms | 0.0068 ms | 1.3727x | 3.2193x |
+| `chain_relu` | `(128,)` | 0.0086 ms | 0.0039 ms | 0.0559 ms | 0.4478x | 6.5042x |
+| `matmul` | `(256, 256)` | 4.1154 ms | 0.3679 ms | 0.0931 ms | 0.0894x | 0.0226x |
+| `conv2d_3x3_padding1` | `(1, 1, 8, 8)` | 0.1788 ms | 1.2241 ms | 0.0131 ms | 6.8478x | 0.0731x |
+| `max_pool2d_2x2` | `(1, 1, 16, 16)` | 0.0146 ms | 0.2649 ms | 0.0062 ms | 18.1659x | 0.4242x |
+| `avg_pool2d_2x2` | `(1, 1, 16, 16)` | 0.0139 ms | 0.5486 ms | 0.0052 ms | 39.5974x | 0.3748x |
+| `elementwise_backward` | `(1024,)` | 2.3885 ms | n/a | 0.1947 ms | n/a | 0.0815x |
 
-Speedup is `NumPy median / TensorStudio median`, so values above `1.0x` favor
-TensorStudio. The matmul result is from the optimized C++ contiguous path; a
-pre-optimization run was about 259 ms for the same `(256, 256)` case on this
-machine.
+Speedup is `competitor median / TensorStudio median`, so values above `1.0x`
+favor TensorStudio.
 
-Do not treat these results as universal. TensorStudio does not currently claim
-to be faster than NumPy, TensorFlow, PyTorch, or JAX overall.
+Do not treat these results as universal. TensorStudio does not claim to be
+faster than NumPy, TensorFlow, PyTorch, or JAX overall.
 
 ## Save And Load
 
@@ -276,6 +287,7 @@ unsafe because pickle can execute arbitrary code.
 
 ```bash
 python -m pip install -e ".[dev,docs]"
+python test_all.py --skip-build
 ruff check .
 mypy python/tensorstudio
 pytest -q
@@ -288,6 +300,7 @@ scikit-build-core, and C++20.
 
 ## Release Checklist
 
+- `python test_all.py` passes locally.
 - `ruff check .` passes.
 - `mypy python/tensorstudio` passes.
 - `pytest -q` passes on Windows, Linux, and macOS.
@@ -312,10 +325,14 @@ tokens or print secrets.
 - CPU backend only.
 - Eager execution only.
 - No CUDA or Metal backend yet.
+- No BLAS-backed matrix multiplication yet.
 - No graph compiler or distributed runtime.
-- No convolution layers yet.
+- Convolution and pooling support are currently limited to CPU NCHW
+  `conv2d`, `max_pool2d`, and `avg_pool2d` style workloads.
+- Reductions support all-element or single-axis reductions, not tuple-axis
+  reductions yet.
 - No sparse tensors or advanced indexing.
-- Limited dtype casting.
+- Dtype casting is basic and does not include a full promotion/casting policy.
 - Experimental performance; benchmarks are local references only.
 - Pickle serialization is for trusted TensorStudio objects only.
 
@@ -323,8 +340,8 @@ tokens or print secrets.
 
 - CUDA backend
 - Graph/JIT mode
-- Convolution ops
-- Dataset utilities
+- Broader convolution ops, adaptive/global pooling, and image-model examples
+- Richer dataset utilities
 - Model zoo examples
 - ONNX import/export
 - Improved memory allocator

{tensorstudio-1.0.0rc2 → tensorstudio-1.1.0}/README.md

@@ -7,14 +7,12 @@
 TensorStudio is a compact C++ tensor and autograd engine with a Python API for
 learning, experimentation, and lightweight ML workloads.
 
-TensorStudio `1.0.0rc2` is a release candidate for a CPU-only stable API
-foundation. It is eager-only, intentionally small, and not a replacement for
-mature ML frameworks. The final `1.0.0` version should only be tagged after the
-Windows, Linux, and macOS release checklist passes.
+TensorStudio `1.1.0` is a CPU-only stable API foundation. It is eager-only,
+intentionally small, and not a replacement for mature ML frameworks.
 
 ## Install
 
-From PyPI, once release-candidate wheels are published:
+From PyPI:
 
 ```bash
 python -m pip install tensorstudio
@@ -96,6 +94,8 @@ d = ts.linspace(0.0, 1.0, 5)
 
 print(a.shape, a.strides, a.device, a.is_contiguous)
 print((b.clamp(0.2, 0.8) + 1).mean().item())
+print(b.sum(axis=1).tolist())
+print(ts.concat([b, b], axis=0).shape, b.astype("float64").dtype)
 print(c.tolist(), d.tolist())
 print(ts.zeros_like(b).shape, ts.randn_like(b, seed=11).dtype)
 ```
@@ -161,8 +161,8 @@ for features, targets in loader:
     print(features.shape, targets.shape)
 ```
 
-The v1 release candidate DataLoader is intentionally single-process so it works
-cleanly on Windows without multiprocessing setup.
+The v1 DataLoader is intentionally single-process so it works cleanly on
+Windows without multiprocessing setup.
 
 ## Performance
 
@@ -171,34 +171,45 @@ performance is still experimental. Benchmarks live in `benchmarks/` and can be
 run locally:
 
 ```bash
+python benchmark_all.py
 python benchmarks/benchmark_report.py
 ```
 
-On one Windows CPython 3.10 run for `1.0.0rc2`, TensorStudio beat NumPy on 11
-small activation/reduction benchmark cases and lost on 78 NumPy-comparable
-cases. The strongest local wins were small `sigmoid`, `sum`, and `mean` cases;
-medium elementwise operations and matrix multiplication remain slower than
-NumPy. The C++ contiguous matmul fast path is much faster than the original
-naive storage-access loop, but it is not a BLAS replacement. See
-`benchmarks/results.md` for the full table, platform details, and exact timings.
+`benchmark_all.py` writes `benchmarks/results.md` and includes explicit win
+columns for NumPy, TensorFlow, PyTorch, and JAX when those libraries are
+available locally.
+
+On one Windows CPython 3.10 development run reporting `1.1.0`, TensorStudio
+beat NumPy on 23 small operation benchmark cases and lost on 80
+NumPy-comparable cases. Against PyTorch CPU `2.12.1+cpu`, TensorStudio won 74
+local cases and lost 34. The strongest local wins were small eager operations,
+small contiguous axis reductions, and the simple NumPy convolution/pooling
+references where framework dispatch or Python loops dominate; larger matrix
+multiplication, PyTorch convolution and pooling, larger axis reductions, larger
+transcendental activations, and larger autograd workloads remain faster in
+PyTorch and NumPy.
+See `benchmarks/results.md` for the full table, platform details, and exact
+timings.
 
 Snapshot from that local run:
 
-| operation | shape | TensorStudio median | NumPy median | speedup | result |
-|---|---:|---:|---:|---:|---|
-| `sigmoid` | `(1,)` | 0.0016 ms | 0.0036 ms | 2.2170x | win |
-| `mean` | `(1,)` | 0.0014 ms | 0.0081 ms | 5.8562x | win |
-| `mean` | `(8,)` | 0.0016 ms | 0.0079 ms | 5.0005x | win |
-| `add` | `(16384,)` | 0.0100 ms | 0.0037 ms | 0.3682x | loss |
-| `matmul` | `(256, 256)` | 4.1393 ms | 0.4222 ms | 0.1020x | loss |
+| operation | shape | TensorStudio | NumPy | PyTorch CPU | TS vs NumPy | TS vs PyTorch |
+|---|---:|---:|---:|---:|---:|---:|
+| `sigmoid` | `(32,)` | 0.0017 ms | 0.0036 ms | 0.0580 ms | 2.1201x | 33.8536x |
+| `mean` | `(32,)` | 0.0018 ms | 0.0078 ms | 0.0127 ms | 4.2927x | 7.0047x |
+| `sum_axis1` | `(16, 16)` | 0.0021 ms | 0.0029 ms | 0.0068 ms | 1.3727x | 3.2193x |
+| `chain_relu` | `(128,)` | 0.0086 ms | 0.0039 ms | 0.0559 ms | 0.4478x | 6.5042x |
+| `matmul` | `(256, 256)` | 4.1154 ms | 0.3679 ms | 0.0931 ms | 0.0894x | 0.0226x |
+| `conv2d_3x3_padding1` | `(1, 1, 8, 8)` | 0.1788 ms | 1.2241 ms | 0.0131 ms | 6.8478x | 0.0731x |
+| `max_pool2d_2x2` | `(1, 1, 16, 16)` | 0.0146 ms | 0.2649 ms | 0.0062 ms | 18.1659x | 0.4242x |
+| `avg_pool2d_2x2` | `(1, 1, 16, 16)` | 0.0139 ms | 0.5486 ms | 0.0052 ms | 39.5974x | 0.3748x |
+| `elementwise_backward` | `(1024,)` | 2.3885 ms | n/a | 0.1947 ms | n/a | 0.0815x |
 
-Speedup is `NumPy median / TensorStudio median`, so values above `1.0x` favor
-TensorStudio. The matmul result is from the optimized C++ contiguous path; a
-pre-optimization run was about 259 ms for the same `(256, 256)` case on this
-machine.
+Speedup is `competitor median / TensorStudio median`, so values above `1.0x`
+favor TensorStudio.
 
-Do not treat these results as universal. TensorStudio does not currently claim
-to be faster than NumPy, TensorFlow, PyTorch, or JAX overall.
+Do not treat these results as universal. TensorStudio does not claim to be
+faster than NumPy, TensorFlow, PyTorch, or JAX overall.
 
 ## Save And Load
 
@@ -218,6 +229,7 @@ unsafe because pickle can execute arbitrary code.
 
 ```bash
 python -m pip install -e ".[dev,docs]"
+python test_all.py --skip-build
 ruff check .
 mypy python/tensorstudio
 pytest -q
@@ -230,6 +242,7 @@ scikit-build-core, and C++20.
 
 ## Release Checklist
 
+- `python test_all.py` passes locally.
 - `ruff check .` passes.
 - `mypy python/tensorstudio` passes.
 - `pytest -q` passes on Windows, Linux, and macOS.
@@ -254,10 +267,14 @@ tokens or print secrets.
 - CPU backend only.
 - Eager execution only.
 - No CUDA or Metal backend yet.
+- No BLAS-backed matrix multiplication yet.
 - No graph compiler or distributed runtime.
-- No convolution layers yet.
+- Convolution and pooling support are currently limited to CPU NCHW
+  `conv2d`, `max_pool2d`, and `avg_pool2d` style workloads.
+- Reductions support all-element or single-axis reductions, not tuple-axis
+  reductions yet.
 - No sparse tensors or advanced indexing.
-- Limited dtype casting.
+- Dtype casting is basic and does not include a full promotion/casting policy.
 - Experimental performance; benchmarks are local references only.
 - Pickle serialization is for trusted TensorStudio objects only.
 
@@ -265,8 +282,8 @@ tokens or print secrets.
 
 - CUDA backend
 - Graph/JIT mode
-- Convolution ops
-- Dataset utilities
+- Broader convolution ops, adaptive/global pooling, and image-model examples
+- Richer dataset utilities
 - Model zoo examples
 - ONNX import/export
 - Improved memory allocator

tensorstudio-1.1.0/benchmark_all.py

@@ -0,0 +1,50 @@
+from __future__ import annotations
+
+import argparse
+from pathlib import Path
+
+from benchmarks.benchmark_report import run_benchmarks
+
+ALL_SECTIONS = {
+    "elementwise",
+    "matmul",
+    "conv2d",
+    "pooling",
+    "reductions",
+    "activations",
+    "autograd",
+    "training_loop",
+}
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="Run every TensorStudio benchmark section.")
+    parser.add_argument(
+        "--section",
+        action="append",
+        choices=sorted(ALL_SECTIONS),
+        help="Run only the selected section. Repeat for multiple sections.",
+    )
+    parser.add_argument(
+        "--output",
+        type=Path,
+        default=Path("benchmarks") / "results.md",
+        help="Markdown report path. Defaults to benchmarks/results.md.",
+    )
+    parser.add_argument(
+        "--print",
+        action="store_true",
+        help="Print the full Markdown report after writing it.",
+    )
+    args = parser.parse_args()
+
+    selected_sections = set(args.section or ALL_SECTIONS)
+    report = run_benchmarks(selected_sections, args.output)
+    if args.print:
+        print(report)
+    else:
+        print(f"Wrote benchmark report to {args.output}")
+
+
+if __name__ == "__main__":
+    main()

tensorstudio-1.1.0/benchmarks/__init__.py

@@ -0,0 +1,2 @@
+"""Local TensorStudio benchmark helpers."""
+

tensorstudio-1.1.0/benchmarks/bench_conv2d.py

@@ -0,0 +1,8 @@
+from __future__ import annotations
+
+from pathlib import Path
+
+from benchmark_report import run_benchmarks
+
+if __name__ == "__main__":
+    print(run_benchmarks({"conv2d"}, Path(__file__).with_name("results_conv2d.md")))

tensorstudio-1.1.0/benchmarks/bench_pooling.py

@@ -0,0 +1,8 @@
+from __future__ import annotations
+
+from pathlib import Path
+
+from benchmark_report import run_benchmarks
+
+if __name__ == "__main__":
+    print(run_benchmarks({"pooling"}, Path(__file__).with_name("results_pooling.md")))