entropack 0.1.0__tar.gz

entropack-0.1.0/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [2026] [ModelScope]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

entropack-0.1.0/PKG-INFO

@@ -0,0 +1,158 @@
+Metadata-Version: 2.4
+Name: entropack
+Version: 0.1.0
+Summary: EntroPack: general-purpose tensor compression for PyTorch with lossless and rate-controlled lossy schemes and GPU encoding and decoding.
+Author: DiffSynth-Studio
+License-Expression: Apache-2.0
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: torch>=2.10
+Requires-Dist: numpy>=1.23
+Requires-Dist: dahuffman>=0.4
+Provides-Extra: cuda13
+Requires-Dist: cupy-cuda13x>=14; extra == "cuda13"
+Provides-Extra: cuda12
+Requires-Dist: cupy-cuda12x>=14; extra == "cuda12"
+Dynamic: license-file
+
+# EntroPack
+
+### General-purpose tensor compression for PyTorch
+
+EntroPack is a general-purpose tensor compression library for PyTorch. It supports lossless
+compression for exact recovery and lossy compression with a target bitrate to balance storage
+and reconstruction accuracy. GPU encoding and decoding compress tensors and restore them
+in their original shape and dtype.
+
+[![License](https://img.shields.io/badge/license-Apache_2.0-blue.svg)](LICENSE)
+![Python](https://img.shields.io/badge/python-%3E%3D3.10-blue.svg)
+
+[Documentation](docs/en/index.rst) · [中文](README_zh.md)
+
+- **Flexible bitrates.** Compress each weight matrix at any non-integer target bitrate,
+  or preserve every input bit with a lossless scheme.
+- **Dtype preservation.** Restore tensors in their input dtype, including BF16, FP16, FP8,
+  and INT8.
+- **PyTorch integration.** Compress and restore tensors through a common API, and save
+  them with `state_dict`. Compressed linear layers provide an integration for model weights.
+
+## Installation
+
+Use Python 3.10 or later and install a CUDA-enabled build of PyTorch 2.10 or later
+for your environment.
+
+### Install from source (recommended)
+
+```bash
+git clone https://github.com/modelscope/entropack.git
+cd entropack
+pip install -e ".[cuda13]"
+```
+
+### Install from PyPI
+
+PyPI releases may lag behind source updates. Install from source for the latest features.
+
+```bash
+pip install "entropack[cuda13]"
+```
+
+Both installation methods above use CUDA 13 and include the matching CuPy package.
+For CUDA 12, replace `cuda13` with `cuda12` in either command. If a compatible CuPy is
+already installed, use `pip install -e .` for source installation or `pip install entropack` for PyPI.
+
+See [Quick start](docs/en/Usage/Quick-start.md) for environment requirements and usage examples.
+
+## Get started
+
+### Direct tensor compression
+
+This example compresses a 2D BF16 tensor at a target of 3.5 bits per element,
+then decompresses it to a tensor with the original shape and dtype:
+
+```python
+import torch
+import entropack as ep
+
+tensor = (torch.randn(256, 256, device="cuda") * 0.02).to(torch.bfloat16)
+config = ep.LatticeRANSConfig(target_bpp=3.5)
+
+compressed = ep.compress(tensor, config)
+restored = ep.decompress(compressed, config)
+
+print(f"Target: {config.target_bpp:.2f} bits per element")
+print(f"Stored: {compressed.actual_bpp:.2f} bits per element")
+print(restored.shape, restored.dtype)
+```
+
+`target_bpp` is the requested number of bits per element (bpp). `actual_bpp` reports the stored
+rate, including metadata. Targets from 1 to 11 are supported, including non-integer values.
+
+### Compressed Linear
+
+`CompressedLinear.from_linear` compresses an existing `torch.nn.Linear`'s weights using
+the supplied Config and returns a new Compressed Linear. Call it as `layer(x)` to compute
+the output, just as with an ordinary linear layer:
+
+```python
+import torch
+import entropack as ep
+
+linear = torch.nn.Linear(256, 256, dtype=torch.bfloat16, device="cuda")
+config = ep.LatticeRANSConfig(target_bpp=4.0)
+layer = ep.CompressedLinear.from_linear(linear, config=config)
+x = torch.randn(8, 256, dtype=torch.bfloat16, device="cuda")
+
+with torch.inference_mode():
+    output = layer(x)
+print(output.shape, f"{layer.compressed_bits:.2f} bits per weight")
+```
+
+See [Compressed Linear usage](docs/en/Usage/Linear-layers.md) for model replacement and checkpoint examples.
+
+## Configuration
+
+Config defines the compression scheme and its settings. It is passed to tensor compression
+and decompression functions or to a compressed Linear layer's constructor.
+
+| Config | Compression and use cases |
+| --- | --- |
+| `DFloat11Config()` | Specialized lossless compression for BF16 tensors. Decompression restores every input bit, for applications requiring exact recovery. |
+| `TileANSConfig()` | Lossless compression for BF16, FP16, FP32, FP8, INT8, and other supported dtypes. The compression ratio depends on the input data distribution. |
+| `LatticeRANSConfig(target_bpp=...)` | Lossy compression of 2D floating-point and integer tensors. `target_bpp` specifies the target bits per element, from 1 to 11 including non-integer values, to balance storage size and reconstruction accuracy. |
+
+See [Compression configuration](docs/en/Usage/Configuration.md) for scheme selection
+and the complete parameter reference.
+
+## Performance
+
+On one NVIDIA H20, EntroPack compresses the weights of all 276 linear layers in
+Z-Image-Turbo's diffusion transformer at a 4 bpp target in **3.5 seconds**. The compressed
+weights occupy **4.02 bpp**, with **7.18%** relative L2 reconstruction error. Inference with
+these weights takes **544.7 ms** per denoising step, only **7.7%** above the original BF16
+model's 505.7 ms.
+
+## Documentation
+
+| Guide | Contents |
+| --- | --- |
+| [Quick start](docs/en/Usage/Quick-start.md) | Install and run tensor compression and Compressed Linear examples |
+| [Compression configuration](docs/en/Usage/Configuration.md) | Choose a scheme and look up supported dtypes and parameters |
+| [Tensor compression](docs/en/Usage/Tensor-compression.md) | Encode, decode, inspect storage, move data, and save or load tensors |
+| [Compressed Linear usage](docs/en/Usage/Linear-layers.md) | Replace model layers, use low-precision computation, and manage checkpoints |
+| [API reference](docs/en/API_Reference/index.md) | Look up functions, classes, and properties |
+
+Compression principles: [DFloat11](docs/en/Principles/DFloat11.md), [tile-ANS](docs/en/Principles/Tile-ANS.md),
+and [EntroPack lattice quantization](docs/en/Principles/Lattice-rANS.md).
+
+## Acknowledgements
+
+EntroPack's design is inspired by [DFloat11](https://github.com/LeanModels/DFloat11),
+[dahuffman](https://github.com/soxofaan/dahuffman),
+[DietGPU](https://github.com/facebookresearch/dietgpu), and
+[tile-ANS](https://arxiv.org/abs/2606.15789).
+
+## License
+
+[Apache License 2.0](LICENSE).

entropack-0.1.0/README.md

@@ -0,0 +1,140 @@
+# EntroPack
+
+### General-purpose tensor compression for PyTorch
+
+EntroPack is a general-purpose tensor compression library for PyTorch. It supports lossless
+compression for exact recovery and lossy compression with a target bitrate to balance storage
+and reconstruction accuracy. GPU encoding and decoding compress tensors and restore them
+in their original shape and dtype.
+
+[![License](https://img.shields.io/badge/license-Apache_2.0-blue.svg)](LICENSE)
+![Python](https://img.shields.io/badge/python-%3E%3D3.10-blue.svg)
+
+[Documentation](docs/en/index.rst) · [中文](README_zh.md)
+
+- **Flexible bitrates.** Compress each weight matrix at any non-integer target bitrate,
+  or preserve every input bit with a lossless scheme.
+- **Dtype preservation.** Restore tensors in their input dtype, including BF16, FP16, FP8,
+  and INT8.
+- **PyTorch integration.** Compress and restore tensors through a common API, and save
+  them with `state_dict`. Compressed linear layers provide an integration for model weights.
+
+## Installation
+
+Use Python 3.10 or later and install a CUDA-enabled build of PyTorch 2.10 or later
+for your environment.
+
+### Install from source (recommended)
+
+```bash
+git clone https://github.com/modelscope/entropack.git
+cd entropack
+pip install -e ".[cuda13]"
+```
+
+### Install from PyPI
+
+PyPI releases may lag behind source updates. Install from source for the latest features.
+
+```bash
+pip install "entropack[cuda13]"
+```
+
+Both installation methods above use CUDA 13 and include the matching CuPy package.
+For CUDA 12, replace `cuda13` with `cuda12` in either command. If a compatible CuPy is
+already installed, use `pip install -e .` for source installation or `pip install entropack` for PyPI.
+
+See [Quick start](docs/en/Usage/Quick-start.md) for environment requirements and usage examples.
+
+## Get started
+
+### Direct tensor compression
+
+This example compresses a 2D BF16 tensor at a target of 3.5 bits per element,
+then decompresses it to a tensor with the original shape and dtype:
+
+```python
+import torch
+import entropack as ep
+
+tensor = (torch.randn(256, 256, device="cuda") * 0.02).to(torch.bfloat16)
+config = ep.LatticeRANSConfig(target_bpp=3.5)
+
+compressed = ep.compress(tensor, config)
+restored = ep.decompress(compressed, config)
+
+print(f"Target: {config.target_bpp:.2f} bits per element")
+print(f"Stored: {compressed.actual_bpp:.2f} bits per element")
+print(restored.shape, restored.dtype)
+```
+
+`target_bpp` is the requested number of bits per element (bpp). `actual_bpp` reports the stored
+rate, including metadata. Targets from 1 to 11 are supported, including non-integer values.
+
+### Compressed Linear
+
+`CompressedLinear.from_linear` compresses an existing `torch.nn.Linear`'s weights using
+the supplied Config and returns a new Compressed Linear. Call it as `layer(x)` to compute
+the output, just as with an ordinary linear layer:
+
+```python
+import torch
+import entropack as ep
+
+linear = torch.nn.Linear(256, 256, dtype=torch.bfloat16, device="cuda")
+config = ep.LatticeRANSConfig(target_bpp=4.0)
+layer = ep.CompressedLinear.from_linear(linear, config=config)
+x = torch.randn(8, 256, dtype=torch.bfloat16, device="cuda")
+
+with torch.inference_mode():
+    output = layer(x)
+print(output.shape, f"{layer.compressed_bits:.2f} bits per weight")
+```
+
+See [Compressed Linear usage](docs/en/Usage/Linear-layers.md) for model replacement and checkpoint examples.
+
+## Configuration
+
+Config defines the compression scheme and its settings. It is passed to tensor compression
+and decompression functions or to a compressed Linear layer's constructor.
+
+| Config | Compression and use cases |
+| --- | --- |
+| `DFloat11Config()` | Specialized lossless compression for BF16 tensors. Decompression restores every input bit, for applications requiring exact recovery. |
+| `TileANSConfig()` | Lossless compression for BF16, FP16, FP32, FP8, INT8, and other supported dtypes. The compression ratio depends on the input data distribution. |
+| `LatticeRANSConfig(target_bpp=...)` | Lossy compression of 2D floating-point and integer tensors. `target_bpp` specifies the target bits per element, from 1 to 11 including non-integer values, to balance storage size and reconstruction accuracy. |
+
+See [Compression configuration](docs/en/Usage/Configuration.md) for scheme selection
+and the complete parameter reference.
+
+## Performance
+
+On one NVIDIA H20, EntroPack compresses the weights of all 276 linear layers in
+Z-Image-Turbo's diffusion transformer at a 4 bpp target in **3.5 seconds**. The compressed
+weights occupy **4.02 bpp**, with **7.18%** relative L2 reconstruction error. Inference with
+these weights takes **544.7 ms** per denoising step, only **7.7%** above the original BF16
+model's 505.7 ms.
+
+## Documentation
+
+| Guide | Contents |
+| --- | --- |
+| [Quick start](docs/en/Usage/Quick-start.md) | Install and run tensor compression and Compressed Linear examples |
+| [Compression configuration](docs/en/Usage/Configuration.md) | Choose a scheme and look up supported dtypes and parameters |
+| [Tensor compression](docs/en/Usage/Tensor-compression.md) | Encode, decode, inspect storage, move data, and save or load tensors |
+| [Compressed Linear usage](docs/en/Usage/Linear-layers.md) | Replace model layers, use low-precision computation, and manage checkpoints |
+| [API reference](docs/en/API_Reference/index.md) | Look up functions, classes, and properties |
+
+Compression principles: [DFloat11](docs/en/Principles/DFloat11.md), [tile-ANS](docs/en/Principles/Tile-ANS.md),
+and [EntroPack lattice quantization](docs/en/Principles/Lattice-rANS.md).
+
+## Acknowledgements
+
+EntroPack's design is inspired by [DFloat11](https://github.com/LeanModels/DFloat11),
+[dahuffman](https://github.com/soxofaan/dahuffman),
+[DietGPU](https://github.com/facebookresearch/dietgpu), and
+[tile-ANS](https://arxiv.org/abs/2606.15789).
+
+## License
+
+[Apache License 2.0](LICENSE).

entropack-0.1.0/entropack/__init__.py

@@ -0,0 +1,15 @@
+from importlib import metadata
+
+try:
+    __version__ = metadata.version("entropack")
+except metadata.PackageNotFoundError:
+    __version__ = "0+unknown"
+
+from .compression import CompressedTensor, compress, decompress
+from .linear import CompressedFP8Linear, CompressedINT8Linear, CompressedLinear
+from .schemes import CompressionConfig, DFloat11Config, LatticeRANSConfig, RawConfig, TileANSConfig
+
+__all__ = [
+    "CompressedFP8Linear", "CompressedINT8Linear", "CompressedLinear", "CompressedTensor", "CompressionConfig",
+    "DFloat11Config", "LatticeRANSConfig", "RawConfig", "TileANSConfig", "__version__", "compress", "decompress",
+]

entropack-0.1.0/entropack/backends/__init__.py

@@ -0,0 +1,19 @@
+from collections.abc import Callable
+from dataclasses import dataclass
+
+import torch
+
+from . import cuda
+
+
+@dataclass(frozen=True)
+class Backend:
+    name: str
+    priority: int = 0
+    probe: Callable[[], str | None] | None = None
+    device: Callable[[], torch.device] | None = None
+
+
+BACKENDS: tuple[Backend, ...] = (
+    Backend("cuda", priority=100, probe=cuda.probe, device=cuda.runs_on), Backend("eager", priority=0),
+)

entropack-0.1.0/entropack/backends/cuda/__init__.py

@@ -0,0 +1,16 @@
+import torch
+
+
+def probe() -> str | None:
+    if not torch.cuda.is_available():
+        return "torch.cuda.is_available() is False (no NVIDIA device / driver)"
+    try:
+        import cupy
+    except ImportError as e:
+        return (f"cupy is not installed ({e}); install a matching build, e.g. "
+                "`pip install cupy-cuda13x` for CUDA 13 or `pip install cupy-cuda12x` for CUDA 12")
+    return None
+
+
+def runs_on() -> torch.device:
+    return torch.device("cuda", torch.cuda.current_device())

entropack-0.1.0/entropack/backends/cuda/device.py

@@ -0,0 +1,96 @@
+from dataclasses import dataclass
+
+import cupy
+import torch
+
+from .kernels import device_index
+
+_FALLBACK_WARP_SIZE = 32
+
+_DEFAULT_GRID_WAVES = 8
+
+__all__ = ["DeviceCaps", "caps", "resolve_threads", "validate_threads_per_block"]
+
+
+def _shared_optin(index: int, properties) -> int:
+    optin = getattr(properties, "shared_memory_per_block_optin", 0)
+    if optin:
+        return int(optin)
+    with cupy.cuda.Device(index):
+        attributes = cupy.cuda.Device(index).attributes
+    return int(attributes["MaxSharedMemoryPerBlockOptin"])
+
+
+@dataclass(frozen=True)
+class DeviceCaps:
+    index: int
+    name: str
+    compute_capability: tuple[int, int]
+    sm_count: int
+    warp_size: int
+    max_threads_per_block: int
+    threads_per_sm: int
+    regs_per_sm: int
+    shared_per_block: int
+    shared_optin: int
+    shared_per_sm: int
+    l2_bytes: int
+
+    def blocks_per_sm(self, threads_per_block: int, shared_per_block: int = 0, regs_per_thread: int = 0) -> int:
+        blocks = self.threads_per_sm // threads_per_block
+        if shared_per_block > 0:
+            blocks = min(blocks, self.shared_per_sm // shared_per_block)
+        if regs_per_thread > 0:
+            blocks = min(blocks, self.regs_per_sm // (regs_per_thread * threads_per_block))
+        return max(1, blocks)
+
+    def resident_blocks(self, threads_per_block: int, shared_per_block: int = 0, regs_per_thread: int = 0) -> int:
+        return self.sm_count * self.blocks_per_sm(threads_per_block, shared_per_block, regs_per_thread)
+
+    def grid(self, wanted: int, threads_per_block: int, shared_per_block: int = 0, waves: int = _DEFAULT_GRID_WAVES) -> int:
+        limit = self.resident_blocks(threads_per_block, shared_per_block) * waves
+        return max(1, min(wanted, limit))
+
+    def shared_limit(self, static_slack: int = 0) -> int:
+        return max(0, min(self.shared_optin, self.shared_per_sm - static_slack))
+
+    def threads_per_block(self, wanted: int) -> int:
+        usable = min(wanted, self.max_threads_per_block)
+        usable -= usable % self.warp_size
+        return max(self.warp_size, usable)
+
+
+_caps_cache: dict[int, DeviceCaps] = {}
+
+
+def caps(device=None) -> DeviceCaps:
+    index = device_index(device)
+    cached = _caps_cache.get(index)
+    if cached is not None:
+        return cached
+    properties = torch.cuda.get_device_properties(index)
+    queried = DeviceCaps(
+        index=index, name=properties.name, compute_capability=(properties.major, properties.minor),
+        sm_count=properties.multi_processor_count, warp_size=getattr(properties, "warp_size", 0) or _FALLBACK_WARP_SIZE,
+        max_threads_per_block=properties.max_threads_per_block, threads_per_sm=properties.max_threads_per_multi_processor,
+        regs_per_sm=properties.regs_per_multiprocessor, shared_per_block=properties.shared_memory_per_block,
+        shared_optin=_shared_optin(index, properties), shared_per_sm=properties.shared_memory_per_multiprocessor,
+        l2_bytes=getattr(properties, "L2_cache_size", 0),
+    )
+    _caps_cache[index] = queried
+    return queried
+
+
+def validate_threads_per_block(caps: DeviceCaps, threads_per_block: int) -> None:
+    if threads_per_block % caps.warp_size or threads_per_block > caps.max_threads_per_block:
+        raise ValueError(
+            f"threads_per_block={threads_per_block} is not launchable on {caps.name}: it must be "
+            f"a multiple of the {caps.warp_size}-thread warp size and at most {caps.max_threads_per_block}"
+        )
+
+
+def resolve_threads(caps: DeviceCaps, requested: int | None, default: int) -> int:
+    if requested is None:
+        return caps.threads_per_block(default)
+    validate_threads_per_block(caps, requested)
+    return requested