tensorstudio 0.1.0__tar.gz → 0.1.1__tar.gz

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/.github/workflows/wheels.yml

@@ -10,6 +10,8 @@ jobs:
   build_wheels:
     name: Build wheels on ${{ matrix.os }}
     runs-on: ${{ matrix.os }}
+    permissions:
+      contents: read
     strategy:
       fail-fast: false
       matrix:
@@ -22,6 +24,13 @@ jobs:
           python-version: "3.12"
       - name: Build wheels
         uses: pypa/cibuildwheel@v2.23.3
+        env:
+          CIBW_BUILD: "cp310-* cp311-* cp312-* cp313-*"
+          CIBW_SKIP: "*-musllinux*"
+          CIBW_ARCHS_WINDOWS: "AMD64"
+          CIBW_ARCHS_LINUX: "x86_64"
+          CIBW_MANYLINUX_X86_64_IMAGE: "manylinux_2_28"
+          CIBW_TEST_COMMAND: "python -c \"import tensorstudio as ts; x = ts.tensor([1.0, 2.0], requires_grad=True); y = (x * x).sum(); y.backward(); print(ts.__version__, x.grad.tolist())\""
       - uses: actions/upload-artifact@v4
         with:
           name: wheels-${{ matrix.os }}

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/CHANGELOG.md

@@ -1,5 +1,12 @@
 # Changelog
 
+## 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.

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/CMakeLists.txt

@@ -1,6 +1,6 @@
 cmake_minimum_required(VERSION 3.18)
 
-project(tensorstudio VERSION 0.1.0 LANGUAGES CXX)
+project(tensorstudio VERSION 0.1.1 LANGUAGES CXX)
 
 find_package(pybind11 CONFIG REQUIRED)
 

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: tensorstudio
-Version: 0.1.0
+Version: 0.1.1
 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
@@ -65,7 +65,7 @@ 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 is experimental v0.1.0 software. It is CPU-only, eager-only, and
+TensorStudio is experimental v0.1.1 software. It is CPU-only, eager-only, and
 intentionally small enough to read and modify.
 
 ## Install

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/README.md

@@ -7,7 +7,7 @@
 TensorStudio is a compact C++ tensor and autograd engine with a Python API for
 learning, experimentation, and lightweight ML workloads.
 
-TensorStudio is experimental v0.1.0 software. It is CPU-only, eager-only, and
+TensorStudio is experimental v0.1.1 software. It is CPU-only, eager-only, and
 intentionally small enough to read and modify.
 
 ## Install

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/SECURITY.md

@@ -1,6 +1,6 @@
 # Security Policy
 
-TensorStudio is experimental v0.1.0 software.
+TensorStudio is experimental v0.1.x software.
 
 ## Reporting a Vulnerability
 
@@ -16,6 +16,6 @@ Include:
 
 ## Serialization Warning
 
-`tensorstudio.load()` uses Python pickle for v0.1.0 internal object roundtrips.
+`tensorstudio.load()` uses Python pickle for v0.1.x internal object roundtrips.
 Loading pickle files from untrusted sources is unsafe because pickle can execute
 arbitrary code during deserialization.

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/docs/Autograde/index.md

@@ -40,7 +40,7 @@ Calling `backward()`:
 
 ## Supported Gradient Operations
 
-v0.1.0 supports gradients for:
+v0.1.x supports gradients for:
 
 - `add`
 - `sub`

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/docs/Benchmarks/index.md

@@ -33,7 +33,7 @@ multiplication on the same shapes.
 ## Interpreting Results
 
 Expect NumPy to be faster for many operations because it uses highly optimized
-native kernels. TensorStudio v0.1.0 favors clarity and correctness over kernel
+native kernels. TensorStudio v0.1.x favors clarity and correctness over kernel
 performance.
 
 Useful benchmark notes:

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/docs/Hardware/cpu-backend.md

@@ -1,13 +1,13 @@
 # CPU Backend
 
-TensorStudio v0.1.0 supports CPU tensors only.
+TensorStudio v0.1.x supports CPU tensors only.
 
 ## Device Abstraction
 
 The C++ core includes a device abstraction so future backends can fit into the
 same public model. Today, every tensor lives on CPU.
 
-## What CPU Means In v0.1.0
+## What CPU Means In v0.1.x
 
 - Storage is host memory.
 - Operations run synchronously.
@@ -36,4 +36,4 @@ CUDA and other accelerator backends are future work. Adding them would require:
 - kernel implementations
 - expanded testing infrastructure
 
-The current CPU implementation keeps those concerns out of v0.1.0.
+The current CPU implementation keeps those concerns out of v0.1.x.

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/docs/Neural Networks/modules.md

@@ -97,6 +97,6 @@ model.train()
 model.eval()
 ```
 
-These currently toggle the `training` flag recursively. v0.1.0 modules do not
+These currently toggle the `training` flag recursively. v0.1.x modules do not
 include dropout or batch normalization, so the flag is mostly scaffolding for
 future modules.

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/docs/Neural Networks/training.md

@@ -58,7 +58,7 @@ w = ts.randn((4, 4), seed=123)
 
 ## Practical Tips
 
-- Keep tensors small in v0.1.0; kernels are simple CPU loops.
+- Keep tensors small in v0.1.x; kernels are simple CPU loops.
 - Prefer `float32` unless you need `float64`.
 - Compare with NumPy in tests for expected numerical values.
 - Watch shape errors carefully around broadcasting and matrix multiplication.

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/docs/Roadmap/index.md

@@ -1,6 +1,6 @@
 # Roadmap
 
-TensorStudio v0.1.0 is a foundation. This roadmap describes likely directions
+TensorStudio v0.1.x is a foundation. This roadmap describes likely directions
 without promising dates.
 
 ## Backend And Performance

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/docs/Usage/serialization.md

@@ -1,6 +1,6 @@
 # Serialization
 
-TensorStudio v0.1.0 provides simple Python-level serialization:
+TensorStudio v0.1.x provides simple Python-level serialization:
 
 ```python
 import tensorstudio as ts
@@ -34,7 +34,7 @@ loaded = ts.load("model.tsmodel")
 
 ## Version Compatibility
 
-The v0.1.0 format is not guaranteed to be stable across future releases. For
+The v0.1.x format is not guaranteed to be stable across future releases. For
 long-term model exchange, ONNX import/export is planned as future work.
 
 ## Recommended File Extension

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/docs/Usage/tensors.md

@@ -94,7 +94,7 @@ Rules:
 
 ## Views
 
-TensorStudio v0.1.0 supports:
+TensorStudio v0.1.x supports:
 
 - `reshape`
 - `flatten`

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/docs/api.md

@@ -32,20 +32,12 @@ Exports:
 Creates a tensor from Python numeric scalars, nested sequences, another
 TensorStudio tensor, or a NumPy array.
 
-```python
-x = ts.tensor([[1, 2], [3, 4]], dtype="float32")
-```
-
 Nested lists must be rectangular. Ragged lists raise a shape error.
 
 ### `from_numpy(array, requires_grad=False)`
 
 Creates a TensorStudio tensor copy from a NumPy array.
 
-```python
-x = ts.from_numpy(np.array([1, 2, 3], dtype=np.int64))
-```
-
 ### Fill And Range Helpers
 
 ```python
@@ -68,30 +60,25 @@ ts.arange(0, 10, 2)
 
 ## Tensor Methods
 
-### Conversion
+Conversion:
 
-- `numpy()`: returns a NumPy copy
-- `tolist()`: returns Python scalar or nested lists
-- `item()`: returns a scalar for single-element tensors
+- `numpy()`
+- `tolist()`
+- `item()`
 
-### Views
+Views:
 
 - `reshape(*shape)` or `reshape(shape)`
 - `flatten()`
 - `transpose()`
 - `T`
 
-Views reuse storage where supported. v0.1.0 supports contiguous reshape and 2D
-transpose.
-
-### Reductions
+Reductions:
 
 - `sum()`
 - `mean()`
 
-Reductions currently reduce all elements to a scalar.
-
-### Activations And Math
+Activations and math:
 
 - `relu()`
 - `sigmoid()`
@@ -99,13 +86,11 @@ Reductions currently reduce all elements to a scalar.
 - `exp()`
 - `log()`
 
-### Autograd
+Autograd:
 
 - `backward(gradient=None)`
 - `zero_grad()`
 
-`backward()` without a gradient is only valid for scalar outputs.
-
 ## Operators
 
 TensorStudio overloads common Python operators:
@@ -144,7 +129,7 @@ Optimizers implement `zero_grad()` and `step()`.
 
 ## Error Types
 
-The native extension maps C++ exceptions to Python exceptions:
+Import native exception aliases from `tensorstudio.errors`:
 
 - `TensorStudioError`
 - `ShapeError`
@@ -152,8 +137,6 @@ The native extension maps C++ exceptions to Python exceptions:
 - `DeviceError`
 - `AutogradError`
 
-Import them from `tensorstudio.errors`.
-
 ## Serialization
 
 ```python
@@ -161,4 +144,4 @@ ts.save(obj, "object.tsmodel")
 loaded = ts.load("object.tsmodel")
 ```
 
-Serialization uses pickle in v0.1.0. Only load files from trusted sources.
+Serialization uses pickle in v0.1.x. Only load files from trusted sources.

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/docs/development.md

@@ -22,7 +22,7 @@ TensorStudio builds a C++20 extension module named `tensorstudio._C`.
 Windows:
 
 ```powershell
-# Use a Visual Studio Developer PowerShell, or make sure cl/nmake are on PATH.
+# Use a Visual Studio Developer PowerShell, or install wheels from PyPI.
 python -m pip install -e ".[dev]"
 ```
 
@@ -40,6 +40,9 @@ sudo apt-get install build-essential cmake
 python -m pip install -e ".[dev]"
 ```
 
+End users should normally install prebuilt wheels from PyPI and should not need
+CMake or a compiler.
+
 ## Development Install
 
 ```bash
@@ -93,14 +96,8 @@ as superior to mature ML frameworks.
 
 ## Testing Strategy
 
-Add tests for:
-
-- Tensor creation and dtype behavior
-- Broadcasting and shape errors
-- NumPy numerical comparisons
-- Autograd gradients for new differentiable ops
-- Python module or optimizer behavior
-- Serialization roundtrips
+Add tests for tensor creation, dtype handling, broadcasting, shape errors, NumPy
+comparisons, autograd gradients, modules, optimizers, and serialization.
 
 For numerical tests, prefer `np.testing.assert_allclose`.
 
@@ -116,5 +113,5 @@ python -m build
 twine check dist/*
 ```
 
-The wheel build requires a working native compiler. If the local machine cannot
-compile the extension, use CI wheel builds.
+Wheels are built in CI with cibuildwheel so users can install without a local
+C++ toolchain.

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/docs/index.md

@@ -11,7 +11,7 @@ examples, benchmarks, and release tooling.
 
 ## Project Status
 
-TensorStudio v0.1.0 is experimental. It is useful for learning how tensor
+TensorStudio v0.1.1 is experimental. It is useful for learning how tensor
 frameworks are assembled, for small experiments, and for lightweight workloads.
 It is not a replacement for mature production ML frameworks.
 
@@ -23,7 +23,7 @@ It is not a replacement for mature production ML frameworks.
 - NumPy-style broadcasting for binary elementwise operations.
 - Matrix multiplication for 2D tensors.
 - Scalar reductions, common activations, reshape, flatten, and 2D transpose.
-- Reverse-mode autograd for the v0.1.0 operation set.
+- Reverse-mode autograd for the v0.1.x operation set.
 - Python `nn.Module`, `Parameter`, `Linear`, `Sequential`, activations, and
   `MSELoss`.
 - Python `optim.SGD` and `optim.Adam`.
@@ -42,7 +42,7 @@ TensorStudio prioritizes:
 
 ## Current Boundaries
 
-TensorStudio v0.1.0 is CPU-only. It has no CUDA backend, no graph compiler, no
+TensorStudio v0.1.1 is CPU-only. It has no CUDA backend, no graph compiler, no
 distributed runtime, no advanced indexing, no sparse tensors, and limited dtype
 casting. Serialization uses pickle, so loading untrusted files is unsafe.
 

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/docs/publishing.md

@@ -13,8 +13,8 @@ long-lived PyPI tokens in developer environments.
 4. Create and push a version tag:
 
    ```bash
-   git tag v0.1.0
-   git push origin v0.1.0
+   git tag v0.1.1
+   git push origin v0.1.1
    ```
 
 5. Publish a GitHub release, or manually run the publish workflow.

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/include/tensorstudio/version.hpp

@@ -2,6 +2,6 @@
 
 namespace tensorstudio {
 
-inline constexpr const char* version = "0.1.0";
+inline constexpr const char* version = "0.1.1";
 
 }  // namespace tensorstudio

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/pyproject.toml

@@ -7,7 +7,7 @@ build-backend = "scikit_build_core.build"
 
 [project]
 name = "tensorstudio"
-version = "0.1.0"
+version = "0.1.1"
 description = "TensorStudio is a compact C++ tensor and autograd engine with a Python API for learning, experimentation, and lightweight ML workloads."
 readme = "README.md"
 requires-python = ">=3.10"

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/python/tensorstudio/_version.py

@@ -1,5 +1,5 @@
 """Package version."""
 
-__version__ = "0.1.0"
+__version__ = "0.1.1"
 
 __all__ = ["__version__"]

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/python/tensorstudio/serialization.py

@@ -1,6 +1,6 @@
 """Serialization helpers.
 
-TensorStudio v0.1.0 uses pickle for internal object roundtrips. Loading pickle
+TensorStudio v0.1.x uses pickle for internal object roundtrips. Loading pickle
 data from untrusted sources is unsafe because pickle can execute arbitrary code
 during deserialization.
 """

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/src/bindings/bind_tensor.cpp

@@ -111,11 +111,11 @@ py::list nested_list_from_tensor(const Tensor& tensor, std::size_t dim, int64_t&
   return result;
 }
 
-std::vector<ssize_t> ssize_shape(const Shape& shape) {
-  std::vector<ssize_t> result;
+std::vector<py::ssize_t> ssize_shape(const Shape& shape) {
+  std::vector<py::ssize_t> result;
   result.reserve(shape.size());
   for (const auto dim : shape) {
-    result.push_back(static_cast<ssize_t>(dim));
+    result.push_back(static_cast<py::ssize_t>(dim));
   }
   return result;
 }
@@ -201,7 +201,7 @@ Tensor tensor_from_numpy(py::array array, bool requires_grad) {
   py::array flat = py::reinterpret_borrow<py::array>(contiguous.attr("ravel")());
   std::vector<double> values;
   values.reserve(static_cast<std::size_t>(flat.size()));
-  for (ssize_t i = 0; i < flat.size(); ++i) {
+  for (py::ssize_t i = 0; i < flat.size(); ++i) {
     values.push_back(py::cast<double>(flat.attr("__getitem__")(i)));
   }
   return Tensor::from_flat_values(values, shape, dtype, requires_grad);

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/src/core/serialization.cpp

@@ -3,7 +3,7 @@
 namespace tensorstudio {
 
 std::string serialization_note() {
-  return "TensorStudio Python serialization uses pickle for v0.1.0 internal objects.";
+  return "TensorStudio Python serialization uses pickle for v0.1.x internal objects.";
 }
 
 }  // namespace tensorstudio

{tensorstudio-0.1.0 → tensorstudio-0.1.1}/tests/test_import.py

@@ -4,7 +4,7 @@ from __future__ import annotations
 def test_import_and_version() -> None:
     import tensorstudio as ts
 
-    assert ts.__version__ == "0.1.0"
+    assert ts.__version__ == "0.1.1"
     assert hasattr(ts, "Tensor")
     assert hasattr(ts, "nn")
     assert hasattr(ts, "optim")