What's Changed

• [Feature] Support vanilla C++ STL in FFI. by @DarkSharpness in #228
• fix: Use-after-move when handling std::tuple by @junrushao in #293
• chore(release): Version bump after v0.1.4 release by @junrushao in #296
• Workaround Segfaults and clang-tidy warnings by @junrushao in #297
• [TEST] Further fix potential use after free issue by @tqchen in #298
• [ADDON] Merge torch c dlpack actions by @cyx-6 in #299
• [FIX] Fixes from_dlpack issue fallback by @tqchen in #301
• Change log level from info to debug for JIT initialization by @merrymercy in #303
• [METADATA] Always return metadata string as created in main lib by @tqchen in #305
• [FEAT] kwargs wrapping utility by @tqchen in #309

New Contributors

• @merrymercy made their first contribution in #303

Full Changelog: v0.1.4...v0.1.5-rc2

What's Changed

• [Feature] Support vanilla C++ STL in FFI. by @DarkSharpness in #228
• fix: Use-after-move when handling std::tuple by @junrushao in #293
• chore(release): Version bump after v0.1.4 release by @junrushao in #296
• Workaround Segfaults and clang-tidy warnings by @junrushao in #297
• [TEST] Further fix potential use after free issue by @tqchen in #298
• [ADDON] Merge torch c dlpack actions by @cyx-6 in #299
• [FIX] Fixes from_dlpack issue fallback by @tqchen in #301
• Change log level from info to debug for JIT initialization by @merrymercy in #303

New Contributors

• @merrymercy made their first contribution in #303

Full Changelog: v0.1.4...v0.1.5-rc1

What's Changed

• [Feature] Support vanilla C++ STL in FFI. by @DarkSharpness in #228
• fix: Use-after-move when handling std::tuple by @junrushao in #293
• chore(release): Version bump after v0.1.4 release by @junrushao in #296
• Workaround Segfaults and clang-tidy warnings by @junrushao in #297
• [TEST] Further fix potential use after free issue by @tqchen in #298
• [ADDON] Merge torch c dlpack actions by @cyx-6 in #299
• [FIX] Fixes from_dlpack issue fallback by @tqchen in #301

Full Changelog: v0.1.4...v0.1.5-rc0

What's Changed

• [Docs] Add kernel library guide by @cyx-6 in #169
• feat: Embed metadata for FFI exported functions by @Kathryn-cat in #230
• fix: clang tidy issue by @Kathryn-cat in #279
• [TORCH] Fix precise version of f8e8m0 support by @tqchen in #282
• [Feature] Add cubin launcher utility as an extra header by @yaoyaoding in #283
• Add cuda DeviceGuard by @cyx-6 in #287
• [DLPACK] Upgrade DLPack Exchange API to pass by capsule by @tqchen in #288
• [DOCS] Update kernel library guide with device guard by @cyx-6 in #289
• chore(release): Add GPG Key from Junru Shao by @junrushao in #290

Full Changelog: v0.1.3...v0.1.4

What's Changed

• feat(cython): Expose TypeAttr via _lookup_type_attr by @junrushao in #247
• fix(cython): Make sure TypeInfo is properly registered by all classes by @junrushao in #246
• chore(cython): Specify --module-name when building Cython by @junrushao in #248
• feat: Introduce tvm.registry.get_registered_type_keys() by @junrushao in #249
• [misc] support various arg types by @DarkSharpness in #229
• [PyTorch] Allow tensor conversion on rocm backend by @MasterJH5574 in #253
• [ADDON] Add torch c dlpack ext wheels for windows by @cyx-6 in #252
• [Fix] Enhanced check for CUDA availability by @MasterJH5574 in #258
• [ADDON] Add torch c dlpack for macos by @cyx-6 in #256
• [DLPack] Leverage exchange api when possible by @tqchen in #260
• [DTYPE] Align bool parsing to align with DLPack by @tqchen in #262
• [DTYPE] Include dtype literals by @tqchen in #263
• feat(stubgen): Refactor into staged pipeline, Introduce directive import by @junrushao in #259
• feat(stubgen): Generate __all__ for proper exporting by @junrushao in #268
• Add Siyuan Feng's GPG key to KEYS file by @Hzfengsy in #269
• [TORCH] Remove ninja dep on non-windows when JIT optional torch-c-dlpack by @tqchen in #272
• [Minor] use perfect forwarding for template types. by @DarkSharpness in #266
• [TEST] Add test for Optional by @tqchen in #274
• [LINT][MINOR] Fix clang-tidy by @tqchen in #275
• [BUILD][MINOR] Remove --no-as-needed by @tqchen in #276
• [chore] Hide symbols by default for extension build by @tqchen in #278

New Contributors

• @MasterJH5574 made their first contribution in #253
• @Hzfengsy made their first contribution in #269

Full Changelog: v0.1.2...v0.1.3-rc1

This is a minor patch release, mainly to include the fixes support for corner cases and enable cuda-python CUstream compactibility

What's Changed

• Add torch c dlpack release workflow by @cyx-6 in #220
• [Fix] fix iterator behavior by @DarkSharpness in #218
• [Test] Add test for filelock utility by @yaoyaoding in #216
• [LINT] Bump macro version after release by @tqchen in #223
• [EXTRA] Improve the Error propagation in alloc by @tqchen in #224
• [Minor] Improve tuple functionality by @DarkSharpness in #219
• [Feature] Support build and build_and_load. by @DarkSharpness in #213
• fix: More fix of ml_dtypes<0.5 by @oraluben in #222
• [ADDON] Fix import issue by @tqchen in #227
• [ADDON] Improved github action for torch-c-dlpack-ext by @cyx-6 in #225
• [CYTHON] Fix ctypes.c_void_p for nullptr by @tqchen in #235
• [STREAM] Enable compact with cuda-python driver stream by @tqchen in #236
• Allow handling of load-bearing compiler flags for dlpack by @directhex in #231
• [ERROR] Make Error conform more to std by @tqchen in #240
• [FIX] Fix missing static registration for DLTensor* by @tqchen in #239
• [CYTHON] Improve fallback and dtype convert behavior by @tqchen in #241
• [TORCH] Update the extension build to o3 by @tqchen in #243
• [PYTHON] Further streamline number handling by @tqchen in #242

New Contributors

• @directhex made their first contribution in #231

Full Changelog: v0.1.1...v0.1.2.rc0

This is a minor patch release, mainly to include the latest support for AOT torch-dlpack-ext and reduce JIT deadlock issues

What's Changed

• doc: Polish quickstart doc and add Examples by @junrushao in #170
• feat: Automatically add __init__ method if available by @junrushao in #174
• doc: Migrate .pyi docstrings into Cython by @junrushao in #175
• Introduce dtype protocol by @tqchen in #178
• Introduce Device Protocol by @tqchen in #179
• doc: Handle builtin and inheritance by @junrushao in #177
• Update readme text by @tqchen in #182
• Fix README.md by @junrushao in #183
• [PYTHON] Robustify libinfo finding by @tqchen in #185
• [Minor] Fix missing function ref in type traits by @DarkSharpness in #186
• feat: Enhance load_module to accept PathLike objects by @Tunghohin in #187
• chore(build): Introduce setuptools_scm for better versioning by @junrushao in #154
• doc: Stable C ABI by @junrushao in #191
• fix(doc): SVG images not rendering by @junrushao in #193
• fix(doc): Remove ?sanitize=true in SVG URL by @junrushao in #196
• feat: support ml_dtypes<0.5 by @oraluben in #198
• chore: Fix clang-tidy warnings by @junrushao in #197
• feat: add clang-tidy to CI by @junrushao in #95
• [Feature] Add a function to build c-dlpack for torch by @yaoyaoding in #192
• [Fix] Keep a reference to loaded patch lib by @yaoyaoding in #200
• doc: Remove stale notes by @junrushao in #204
• [Lint] Lint the build torch c dlpack utility by @yaoyaoding in #203
• [DOCS] add author by @tqchen in #207
• doc(python): Add example coverage to docstrings by @junrushao in #206
• [Fix] Remove python lib during building torch c dlpack ext by @yaoyaoding in #210
• [Docs][Examples] Add CUDA C++ loading example and improve development documentation by @yifeifang in #209
• feat: Build PyTorch DLPack extension ahead of time by @yzh119 in #189
• [DOCS] Add README to torch-c-dlpack-ext by @tqchen in #214
• Minor improvement on type hint of register_object by @KEKE046 in #212

New Contributors

• @DarkSharpness made their first contribution in #186
• @Tunghohin made their first contribution in #187
• @KEKE046 made their first contribution in #212

Full Changelog: v0.1.0...v0.1.1.rc0

This is the first release of tvm-ffi

Apache TVM FFI is an open ABI and FFI for ML systems. It is a minimal, framework-agnostic, yet flexible open convention with the following systems in mind:

• Kernel libraries: ship one wheel to support multiple frameworks, Python versions, and different languages.
• Kernel DSLs: reusable open ABI for JIT and AOT kernel exposure to PyTorch, JAX, and other ML runtimes.
• ML frameworks and runtimes: unified mechanism to connect libraries and DSLs that adopt the ABI convention.
• Coding agents: unified mechanism to package and ship generated code to production environments.
• ML infrastructure: cross-language support for Python, C++, and Rust, and DSLs.

It has the following technical features:

• DLPack-compatible Tensor data ABI to seamlessly support many frameworks such as PyTorch, JAX, CuPy and others that support DLPack convention.
• Compact value and function calling convention for common data types in machine learning.
• Stable, minimal, and flexible C ABI to support machine learning system use-cases.
• Out-of-the-box multi-language support for Python, C++, Rust, and future path for other languages.

Full Changelog: https://github.com/apache/tvm-ffi/commits/v0.1.0