tensorstudio 1.3.4__tar.gz → 1.3.6__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (163) hide show
  1. tensorstudio-1.3.6/.gitattributes +17 -0
  2. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/CHANGELOG.md +25 -0
  3. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/CMakeLists.txt +1 -1
  4. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/PKG-INFO +6 -2
  5. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/README.md +5 -1
  6. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/benchmarks/benchmark_report.py +5 -2
  7. tensorstudio-1.3.6/benchmarks/results.md +343 -0
  8. tensorstudio-1.3.4/docs/Autograde/index.md → tensorstudio-1.3.6/docs/autograd/coverage.md +1 -1
  9. tensorstudio-1.3.6/docs/autograd/gradient-notes.md +46 -0
  10. tensorstudio-1.3.4/docs/Usage/api.md → tensorstudio-1.3.6/docs/core/api-reference.md +47 -0
  11. tensorstudio-1.3.6/docs/core/broadcasting.md +52 -0
  12. tensorstudio-1.3.6/docs/core/native-cpp-core.md +75 -0
  13. {tensorstudio-1.3.4/docs/Usage → tensorstudio-1.3.6/docs/core}/tensors.md +24 -0
  14. tensorstudio-1.3.4/docs/development.md → tensorstudio-1.3.6/docs/development/contributing.md +1 -1
  15. tensorstudio-1.3.6/docs/development/cpp-first.md +42 -0
  16. tensorstudio-1.3.6/docs/getting-started/installation.md +66 -0
  17. {tensorstudio-1.3.4/docs → tensorstudio-1.3.6/docs/getting-started}/quickstart.md +1 -0
  18. tensorstudio-1.3.6/docs/getting-started/source-build.md +50 -0
  19. {tensorstudio-1.3.4/docs/Hardware → tensorstudio-1.3.6/docs/hardware}/cpu-backend.md +1 -1
  20. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/docs/index.md +16 -14
  21. tensorstudio-1.3.6/docs/interchange/model-formats.md +37 -0
  22. tensorstudio-1.3.4/docs/Usage/interchange.md → tensorstudio-1.3.6/docs/interchange/onnx.md +1 -1
  23. {tensorstudio-1.3.4/docs/Usage → tensorstudio-1.3.6/docs/interchange}/serialization.md +1 -1
  24. tensorstudio-1.3.4/docs/Neural Networks/training.md → tensorstudio-1.3.6/docs/nn/training.md +1 -1
  25. tensorstudio-1.3.4/docs/Benchmarks/index.md → tensorstudio-1.3.6/docs/performance/benchmarks.md +1 -1
  26. {tensorstudio-1.3.4/docs → tensorstudio-1.3.6/docs/release}/publishing.md +4 -4
  27. tensorstudio-1.3.4/docs/Roadmap/index.md → tensorstudio-1.3.6/docs/roadmap/roadmap.md +4 -3
  28. tensorstudio-1.3.6/docs/vision/dataset-creation.md +58 -0
  29. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/include/tensorstudio/ops.hpp +7 -0
  30. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/include/tensorstudio/version.hpp +1 -1
  31. tensorstudio-1.3.6/mkdocs.yml +62 -0
  32. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/pyproject.toml +1 -1
  33. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/_C.pyi +16 -0
  34. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/__init__.py +26 -0
  35. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/_version.py +1 -1
  36. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/ops.py +25 -0
  37. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/src/bindings/bind_ops.cpp +27 -0
  38. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/src/bindings/bind_tensor.cpp +41 -0
  39. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/src/bindings/bindings.hpp +7 -0
  40. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/src/core/ops.cpp +215 -0
  41. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/tests/test_autograd.py +23 -0
  42. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/tests/test_import.py +1 -1
  43. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/tests/test_ops.py +69 -0
  44. tensorstudio-1.3.4/benchmarks/results.md +0 -353
  45. tensorstudio-1.3.4/mkdocs.yml +0 -47
  46. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/.github/workflows/ci.yml +0 -0
  47. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/.github/workflows/publish-testpypi.yml +0 -0
  48. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/.github/workflows/publish.yml +0 -0
  49. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/.github/workflows/wheels.yml +0 -0
  50. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/.gitignore +0 -0
  51. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/.pre-commit-config.yaml +0 -0
  52. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/CONTRIBUTING.md +0 -0
  53. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/LICENSE +0 -0
  54. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/SECURITY.md +0 -0
  55. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/benchmark_all.py +0 -0
  56. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/benchmarks/__init__.py +0 -0
  57. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/benchmarks/bench_activations.py +0 -0
  58. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/benchmarks/bench_autograd.py +0 -0
  59. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/benchmarks/bench_conv2d.py +0 -0
  60. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/benchmarks/bench_elementwise.py +0 -0
  61. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/benchmarks/bench_matmul.py +0 -0
  62. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/benchmarks/bench_pooling.py +0 -0
  63. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/benchmarks/bench_reductions.py +0 -0
  64. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/benchmarks/bench_tensor_ops.py +0 -0
  65. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/benchmarks/bench_training_loop.py +0 -0
  66. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/benchmarks/results_conv2d.md +0 -0
  67. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/benchmarks/results_matmul.md +0 -0
  68. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/benchmarks/results_pooling.md +0 -0
  69. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/benchmarks/results_reductions.md +0 -0
  70. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/benchmarks/results_tensor_ops.md +0 -0
  71. /tensorstudio-1.3.4/docs/Autograde/autograd.md → /tensorstudio-1.3.6/docs/autograd/overview.md +0 -0
  72. /tensorstudio-1.3.4/docs/Usage/data.md → /tensorstudio-1.3.6/docs/data/datasets-and-dataloaders.md +0 -0
  73. {tensorstudio-1.3.4/docs/Hardware → tensorstudio-1.3.6/docs/hardware}/linux.md +0 -0
  74. {tensorstudio-1.3.4/docs/Hardware → tensorstudio-1.3.6/docs/hardware}/macos.md +0 -0
  75. {tensorstudio-1.3.4/docs/Hardware → tensorstudio-1.3.6/docs/hardware}/windows.md +0 -0
  76. {tensorstudio-1.3.4/docs/Usage → tensorstudio-1.3.6/docs/interchange}/numpy-interop.md +0 -0
  77. /tensorstudio-1.3.4/docs/serialization.md → /tensorstudio-1.3.6/docs/interchange/trusted-serialization.md +0 -0
  78. {tensorstudio-1.3.4/docs/Neural Networks → tensorstudio-1.3.6/docs/nn}/modules.md +0 -0
  79. /tensorstudio-1.3.4/docs/Usage/optim.md → /tensorstudio-1.3.6/docs/nn/optimizers.md +0 -0
  80. /tensorstudio-1.3.4/docs/Neural Networks/nn.md → /tensorstudio-1.3.6/docs/nn/overview.md +0 -0
  81. /tensorstudio-1.3.4/docs/Projects/index.md → /tensorstudio-1.3.6/docs/project/workflows.md +0 -0
  82. /tensorstudio-1.3.4/docs/Vision/index.md → /tensorstudio-1.3.6/docs/vision/overview.md +0 -0
  83. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/examples/basic_tensor_ops.py +0 -0
  84. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/examples/export_onnx.py +0 -0
  85. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/examples/image_folder_classification.py +0 -0
  86. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/examples/linear_regression.py +0 -0
  87. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/examples/project_training.py +0 -0
  88. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/examples/save_load_model.py +0 -0
  89. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/examples/tiny_mlp.py +0 -0
  90. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/examples/vision_classifier.py +0 -0
  91. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/include/tensorstudio/autograd.hpp +0 -0
  92. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/include/tensorstudio/device.hpp +0 -0
  93. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/include/tensorstudio/dtype.hpp +0 -0
  94. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/include/tensorstudio/errors.hpp +0 -0
  95. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/include/tensorstudio/module.hpp +0 -0
  96. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/include/tensorstudio/optim.hpp +0 -0
  97. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/include/tensorstudio/random.hpp +0 -0
  98. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/include/tensorstudio/serialization.hpp +0 -0
  99. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/include/tensorstudio/shape.hpp +0 -0
  100. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/include/tensorstudio/storage.hpp +0 -0
  101. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/include/tensorstudio/tensor.hpp +0 -0
  102. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/autograd.py +0 -0
  103. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/data/__init__.py +0 -0
  104. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/data/dataloader.py +0 -0
  105. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/data/dataset.py +0 -0
  106. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/dtypes.py +0 -0
  107. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/errors.py +0 -0
  108. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/grad_mode.py +0 -0
  109. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/interchange/__init__.py +0 -0
  110. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/interchange/onnx.py +0 -0
  111. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/math.py +0 -0
  112. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/nn/__init__.py +0 -0
  113. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/nn/functional.py +0 -0
  114. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/nn/losses.py +0 -0
  115. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/nn/modules.py +0 -0
  116. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/optim/__init__.py +0 -0
  117. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/optim/adam.py +0 -0
  118. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/optim/adamw.py +0 -0
  119. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/optim/lr_scheduler.py +0 -0
  120. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/optim/sgd.py +0 -0
  121. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/optim/utils.py +0 -0
  122. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/project/__init__.py +0 -0
  123. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/project/checkpoint.py +0 -0
  124. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/project/config.py +0 -0
  125. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/project/trainer.py +0 -0
  126. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/py.typed +0 -0
  127. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/serialization.py +0 -0
  128. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/tensor.py +0 -0
  129. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/typing.py +0 -0
  130. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/vision/__init__.py +0 -0
  131. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/vision/datasets.py +0 -0
  132. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/vision/io.py +0 -0
  133. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/vision/metrics.py +0 -0
  134. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/vision/models.py +0 -0
  135. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/vision/transforms.py +0 -0
  136. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/python/tensorstudio/vision/visualization.py +0 -0
  137. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/src/bindings/bind_autograd.cpp +0 -0
  138. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/src/bindings/bind_nn.cpp +0 -0
  139. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/src/bindings/bind_optim.cpp +0 -0
  140. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/src/bindings/bindings.cpp +0 -0
  141. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/src/core/autograd.cpp +0 -0
  142. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/src/core/device.cpp +0 -0
  143. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/src/core/dtype.cpp +0 -0
  144. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/src/core/errors.cpp +0 -0
  145. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/src/core/module.cpp +0 -0
  146. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/src/core/optim.cpp +0 -0
  147. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/src/core/random.cpp +0 -0
  148. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/src/core/serialization.cpp +0 -0
  149. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/src/core/shape.cpp +0 -0
  150. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/src/core/storage.cpp +0 -0
  151. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/src/core/tensor.cpp +0 -0
  152. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/test_all.py +0 -0
  153. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/tests/test_benchmark_report.py +0 -0
  154. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/tests/test_broadcasting.py +0 -0
  155. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/tests/test_data.py +0 -0
  156. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/tests/test_interchange.py +0 -0
  157. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/tests/test_nn.py +0 -0
  158. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/tests/test_numpy_interop.py +0 -0
  159. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/tests/test_optim.py +0 -0
  160. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/tests/test_project.py +0 -0
  161. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/tests/test_serialization.py +0 -0
  162. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/tests/test_tensor.py +0 -0
  163. {tensorstudio-1.3.4 → tensorstudio-1.3.6}/tests/test_vision.py +0 -0
tensorstudio-1.3.6/.gitattributes ADDED
@@ -0,0 +1,17 @@
1
+ # Keep GitHub language statistics focused on maintained implementation code.
2
+ # The Python package remains counted; docs, examples, tests, and generated
3
+ # outputs are supporting material for release quality rather than core runtime.
4
+ *.md linguist-documentation
5
+ docs/** linguist-documentation
6
+ examples/** linguist-documentation
7
+ tests/** linguist-documentation
8
+ benchmarks/** linguist-documentation
9
+ benchmark_all.py linguist-documentation
10
+ test_all.py linguist-documentation
11
+ site/** linguist-generated
12
+ build/** linguist-generated
13
+ dist/** linguist-generated
14
+ *.egg-info/** linguist-generated
15
+
16
+ include/** linguist-language=C++
17
+ src/** linguist-language=C++
{tensorstudio-1.3.4 → tensorstudio-1.3.6}/CHANGELOG.md RENAMED
@@ -2,6 +2,31 @@
2
2
 
3
3
  ## Unreleased
4
4
 
5
+ ## 1.3.6 - 2026-07-07
6
+
7
+ - Completed the next ordered correctness-roadmap item with native C++
8
+ `where`, `maximum`, and `minimum` operations.
9
+ - Exposed top-level comparison helpers, `clamp`/`clip`, selection helpers, and
10
+ Tensor method helpers through the Python API while keeping kernels in C++.
11
+ - Added NumPy parity and autograd tests for `where`, `maximum`, and `minimum`,
12
+ including broadcasting and tie-gradient behavior.
13
+ - Reorganized the MkDocs tree into categorized sections and added deeper docs
14
+ for installation, source builds, native C++ architecture, broadcasting,
15
+ gradient behavior, model formats, dataset creation, and the C++-first policy.
16
+ - Added GitHub Linguist attributes so language statistics emphasize the native
17
+ C++ library and do not count docs, examples, tests, generated site output, or
18
+ benchmark scripts as implementation code.
19
+
20
+ ## 1.3.5 - 2026-07-07
21
+
22
+ - Completed the next ordered correctness-roadmap item with native C++
23
+ `argmax` and `argmin` operations.
24
+ - Added Tensor methods, functional ops, top-level Python exports, and type
25
+ stubs for arg reductions.
26
+ - Added NumPy parity tests for all-element and axis arg reductions, including
27
+ `keepdims`, negative axes, first-tie behavior, dtype, and error cases.
28
+ - Documented arg-reduction semantics and limitations.
29
+
5
30
  ## 1.3.4 - 2026-07-07
6
31
 
7
32
  - Completed the next ordered correctness-roadmap item with tuple/list-axis
{tensorstudio-1.3.4 → tensorstudio-1.3.6}/CMakeLists.txt RENAMED
@@ -1,6 +1,6 @@
1
1
  cmake_minimum_required(VERSION 3.18)
2
2
 
3
- project(tensorstudio VERSION 1.0.1 LANGUAGES CXX)
3
+ project(tensorstudio VERSION 1.3.6 LANGUAGES CXX)
4
4
 
5
5
  find_package(Python COMPONENTS Interpreter Development.Module REQUIRED)
6
6
  set(PYBIND11_FINDPYTHON ON)
{tensorstudio-1.3.4 → tensorstudio-1.3.6}/PKG-INFO RENAMED
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.2
2
2
  Name: tensorstudio
3
- Version: 1.3.4
3
+ Version: 1.3.6
4
4
  Summary: TensorStudio is a compact C++ tensor and autograd engine with a Python API for learning, experimentation, and lightweight ML workloads.
5
5
  Keywords: tensor,autograd,machine-learning,cpp,pybind11
6
6
  Author: TensorStudio contributors
@@ -71,7 +71,7 @@ Description-Content-Type: text/markdown
71
71
  TensorStudio is a compact C++ tensor and autograd engine with a Python API for
72
72
  learning, experimentation, and lightweight ML workloads.
73
73
 
74
- TensorStudio `1.3.4` is a CPU-only stable API foundation. It is eager-only,
74
+ TensorStudio `1.3.6` is a CPU-only stable API foundation. It is eager-only,
75
75
  intentionally small, and not a replacement for mature ML frameworks.
76
76
 
77
77
  ## Install
@@ -474,6 +474,10 @@ tokens or print secrets.
474
474
  - ONNX support is export-only for a limited set of TensorStudio modules.
475
475
  - Reductions support all-element, single-axis, and tuple/list-axis reductions
476
476
  for `sum`, `mean`, `max`, and `min`.
477
+ - Arg reductions support all-element flat indices or one axis at a time for
478
+ `argmax` and `argmin`.
479
+ - Selection helpers `where`, `maximum`, and `minimum` are native C++ tensor ops
480
+ with broadcasting and autograd support for floating-point branches.
477
481
  - No sparse tensors or advanced indexing.
478
482
  - Dtype casting is basic and does not include a full promotion/casting policy.
479
483
  - Experimental performance; benchmarks are local references only.
{tensorstudio-1.3.4 → tensorstudio-1.3.6}/README.md RENAMED
@@ -7,7 +7,7 @@
7
7
  TensorStudio is a compact C++ tensor and autograd engine with a Python API for
8
8
  learning, experimentation, and lightweight ML workloads.
9
9
 
10
- TensorStudio `1.3.4` is a CPU-only stable API foundation. It is eager-only,
10
+ TensorStudio `1.3.6` is a CPU-only stable API foundation. It is eager-only,
11
11
  intentionally small, and not a replacement for mature ML frameworks.
12
12
 
13
13
  ## Install
@@ -410,6 +410,10 @@ tokens or print secrets.
410
410
  - ONNX support is export-only for a limited set of TensorStudio modules.
411
411
  - Reductions support all-element, single-axis, and tuple/list-axis reductions
412
412
  for `sum`, `mean`, `max`, and `min`.
413
+ - Arg reductions support all-element flat indices or one axis at a time for
414
+ `argmax` and `argmin`.
415
+ - Selection helpers `where`, `maximum`, and `minimum` are native C++ tensor ops
416
+ with broadcasting and autograd support for floating-point branches.
413
417
  - No sparse tensors or advanced indexing.
414
418
  - Dtype casting is basic and does not include a full promotion/casting policy.
415
419
  - Experimental performance; benchmarks are local references only.
{tensorstudio-1.3.4 → tensorstudio-1.3.6}/benchmarks/benchmark_report.py RENAMED
@@ -2,13 +2,14 @@ from __future__ import annotations
2
2
 
3
3
  import argparse
4
4
  import importlib
5
+ import io
5
6
  import os
6
7
  import platform
7
8
  import statistics
8
9
  import sys
9
10
  import time
10
11
  from collections.abc import Callable, Iterable
11
- from contextlib import suppress
12
+ from contextlib import redirect_stderr, redirect_stdout, suppress
12
13
  from dataclasses import dataclass
13
14
  from pathlib import Path
14
15
  from typing import Any
@@ -17,6 +18,7 @@ import numpy as np
17
18
  import tensorstudio as ts
18
19
 
19
20
  os.environ.setdefault("CUDA_VISIBLE_DEVICES", "-1")
21
+ os.environ.setdefault("TF_CPP_MIN_LOG_LEVEL", "3")
20
22
 
21
23
 
22
24
  ArrayFactory = Callable[[np.ndarray], Any]
@@ -59,7 +61,8 @@ class Stats:
59
61
 
60
62
  def _optional_import(module_name: str) -> Any | None:
61
63
  try:
62
- return importlib.import_module(module_name)
64
+ with redirect_stdout(io.StringIO()), redirect_stderr(io.StringIO()):
65
+ return importlib.import_module(module_name)
63
66
  except Exception:
64
67
  return None
65
68
 
tensorstudio-1.3.6/benchmarks/results.md ADDED
@@ -0,0 +1,343 @@
1
+ # TensorStudio Benchmark Results
2
+
3
+ These benchmarks are local references, not marketing claims.
4
+ TensorStudio is compared on CPU only and uses the same shapes and dtype.
5
+
6
+ ## Environment
7
+
8
+ - Platform: `Windows-10-10.0.26200-SP0`
9
+ - Processor: `Intel64 Family 6 Model 158 Stepping 10, GenuineIntel`
10
+ - Python: `3.10.11`
11
+ - TensorStudio: `1.3.6`
12
+ - NumPy: `1.26.4`
13
+ - TensorFlow CPU eager: unavailable (not installed)
14
+ - PyTorch CPU: unavailable (not installed)
15
+ - JAX CPU dispatch: available (0.6.2)
16
+
17
+ ## Summary
18
+
19
+ - TensorStudio wins versus NumPy: `27`
20
+ - TensorStudio losses versus NumPy: `76`
21
+ - TensorStudio wins versus JAX CPU dispatch: `85`
22
+ - TensorStudio losses versus JAX CPU dispatch: `13`
23
+
24
+ TensorStudio beat NumPy on at least one local benchmark case. See the detailed table before making any narrow performance claim.
25
+
26
+ ## Detailed Results
27
+
28
+ | category | operation | shape | library | median ms | mean ms | min ms | max ms | std ms | TS vs NumPy | TS vs TensorFlow | TS vs PyTorch | TS vs JAX | win vs NumPy | win vs TensorFlow | win vs PyTorch | win vs JAX | fastest library | result |
29
+ |---|---|---:|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|---|---|---|---|---|---|
30
+ | elementwise | add | `(1,)` | TensorStudio | 0.0026 | 0.0026 | 0.0025 | 0.0027 | 0.0001 | 0.2604 | n/a | n/a | 4.3415 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
31
+ | elementwise | add | `(1,)` | NumPy | 0.0007 | 0.0007 | 0.0007 | 0.0007 | 0.0000 | 0.2604 | n/a | n/a | 4.3415 | no | n/a | n/a | yes | NumPy | NumPy baseline |
32
+ | elementwise | add | `(1,)` | JAX CPU dispatch | 0.0113 | 0.0112 | 0.0109 | 0.0115 | 0.0002 | 0.2604 | n/a | n/a | 4.3415 | no | n/a | n/a | yes | NumPy | reference |
33
+ | elementwise | sub | `(1,)` | TensorStudio | 0.0027 | 0.0027 | 0.0026 | 0.0030 | 0.0001 | 0.2655 | n/a | n/a | 4.1326 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
34
+ | elementwise | sub | `(1,)` | NumPy | 0.0007 | 0.0007 | 0.0007 | 0.0008 | 0.0000 | 0.2655 | n/a | n/a | 4.1326 | no | n/a | n/a | yes | NumPy | NumPy baseline |
35
+ | elementwise | sub | `(1,)` | JAX CPU dispatch | 0.0110 | 0.0110 | 0.0107 | 0.0116 | 0.0003 | 0.2655 | n/a | n/a | 4.1326 | no | n/a | n/a | yes | NumPy | reference |
36
+ | elementwise | mul | `(1,)` | TensorStudio | 0.0025 | 0.0026 | 0.0025 | 0.0027 | 0.0001 | 0.3038 | n/a | n/a | 4.5050 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
37
+ | elementwise | mul | `(1,)` | NumPy | 0.0008 | 0.0008 | 0.0008 | 0.0008 | 0.0000 | 0.3038 | n/a | n/a | 4.5050 | no | n/a | n/a | yes | NumPy | NumPy baseline |
38
+ | elementwise | mul | `(1,)` | JAX CPU dispatch | 0.0113 | 0.0123 | 0.0110 | 0.0144 | 0.0014 | 0.3038 | n/a | n/a | 4.5050 | no | n/a | n/a | yes | NumPy | reference |
39
+ | elementwise | div | `(1,)` | TensorStudio | 0.0026 | 0.0026 | 0.0026 | 0.0026 | 0.0000 | 0.2896 | n/a | n/a | 3.2589 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
40
+ | elementwise | div | `(1,)` | NumPy | 0.0008 | 0.0008 | 0.0007 | 0.0008 | 0.0000 | 0.2896 | n/a | n/a | 3.2589 | no | n/a | n/a | yes | NumPy | NumPy baseline |
41
+ | elementwise | div | `(1,)` | JAX CPU dispatch | 0.0085 | 0.0087 | 0.0085 | 0.0089 | 0.0002 | 0.2896 | n/a | n/a | 3.2589 | no | n/a | n/a | yes | NumPy | reference |
42
+ | elementwise | chain_relu((x*2+1)/3) | `(1,)` | TensorStudio | 0.0093 | 0.0094 | 0.0093 | 0.0096 | 0.0001 | 0.6152 | n/a | n/a | 8.9306 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
43
+ | elementwise | chain_relu((x*2+1)/3) | `(1,)` | NumPy | 0.0058 | 0.0059 | 0.0056 | 0.0064 | 0.0003 | 0.6152 | n/a | n/a | 8.9306 | no | n/a | n/a | yes | NumPy | NumPy baseline |
44
+ | elementwise | chain_relu((x*2+1)/3) | `(1,)` | JAX CPU dispatch | 0.0835 | 0.0843 | 0.0826 | 0.0883 | 0.0021 | 0.6152 | n/a | n/a | 8.9306 | no | n/a | n/a | yes | NumPy | reference |
45
+ | elementwise | add | `(8,)` | TensorStudio | 0.0025 | 0.0026 | 0.0025 | 0.0026 | 0.0000 | 0.2712 | n/a | n/a | 4.3466 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
46
+ | elementwise | add | `(8,)` | NumPy | 0.0007 | 0.0007 | 0.0007 | 0.0007 | 0.0000 | 0.2712 | n/a | n/a | 4.3466 | no | n/a | n/a | yes | NumPy | NumPy baseline |
47
+ | elementwise | add | `(8,)` | JAX CPU dispatch | 0.0111 | 0.0111 | 0.0108 | 0.0113 | 0.0001 | 0.2712 | n/a | n/a | 4.3466 | no | n/a | n/a | yes | NumPy | reference |
48
+ | elementwise | sub | `(8,)` | TensorStudio | 0.0026 | 0.0026 | 0.0026 | 0.0027 | 0.0001 | 0.2749 | n/a | n/a | 4.5515 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
49
+ | elementwise | sub | `(8,)` | NumPy | 0.0007 | 0.0007 | 0.0007 | 0.0008 | 0.0000 | 0.2749 | n/a | n/a | 4.5515 | no | n/a | n/a | yes | NumPy | NumPy baseline |
50
+ | elementwise | sub | `(8,)` | JAX CPU dispatch | 0.0119 | 0.0119 | 0.0117 | 0.0122 | 0.0002 | 0.2749 | n/a | n/a | 4.5515 | no | n/a | n/a | yes | NumPy | reference |
51
+ | elementwise | mul | `(8,)` | TensorStudio | 0.0026 | 0.0027 | 0.0025 | 0.0028 | 0.0001 | 0.2912 | n/a | n/a | 4.4156 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
52
+ | elementwise | mul | `(8,)` | NumPy | 0.0008 | 0.0008 | 0.0008 | 0.0009 | 0.0001 | 0.2912 | n/a | n/a | 4.4156 | no | n/a | n/a | yes | NumPy | NumPy baseline |
53
+ | elementwise | mul | `(8,)` | JAX CPU dispatch | 0.0116 | 0.0117 | 0.0113 | 0.0120 | 0.0003 | 0.2912 | n/a | n/a | 4.4156 | no | n/a | n/a | yes | NumPy | reference |
54
+ | elementwise | div | `(8,)` | TensorStudio | 0.0027 | 0.0027 | 0.0026 | 0.0028 | 0.0001 | 0.2860 | n/a | n/a | 3.2805 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
55
+ | elementwise | div | `(8,)` | NumPy | 0.0008 | 0.0008 | 0.0008 | 0.0009 | 0.0001 | 0.2860 | n/a | n/a | 3.2805 | no | n/a | n/a | yes | NumPy | NumPy baseline |
56
+ | elementwise | div | `(8,)` | JAX CPU dispatch | 0.0088 | 0.0088 | 0.0084 | 0.0091 | 0.0002 | 0.2860 | n/a | n/a | 3.2805 | no | n/a | n/a | yes | NumPy | reference |
57
+ | elementwise | chain_relu((x*2+1)/3) | `(8,)` | TensorStudio | 0.0095 | 0.0096 | 0.0094 | 0.0100 | 0.0002 | 0.6266 | n/a | n/a | 9.3277 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
58
+ | elementwise | chain_relu((x*2+1)/3) | `(8,)` | NumPy | 0.0060 | 0.0060 | 0.0057 | 0.0062 | 0.0002 | 0.6266 | n/a | n/a | 9.3277 | no | n/a | n/a | yes | NumPy | NumPy baseline |
59
+ | elementwise | chain_relu((x*2+1)/3) | `(8,)` | JAX CPU dispatch | 0.0887 | 0.0892 | 0.0873 | 0.0934 | 0.0022 | 0.6266 | n/a | n/a | 9.3277 | no | n/a | n/a | yes | NumPy | reference |
60
+ | elementwise | add | `(32,)` | TensorStudio | 0.0027 | 0.0029 | 0.0025 | 0.0035 | 0.0004 | 0.2683 | n/a | n/a | 4.3975 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
61
+ | elementwise | add | `(32,)` | NumPy | 0.0007 | 0.0008 | 0.0007 | 0.0009 | 0.0001 | 0.2683 | n/a | n/a | 4.3975 | no | n/a | n/a | yes | NumPy | NumPy baseline |
62
+ | elementwise | add | `(32,)` | JAX CPU dispatch | 0.0117 | 0.0120 | 0.0111 | 0.0134 | 0.0008 | 0.2683 | n/a | n/a | 4.3975 | no | n/a | n/a | yes | NumPy | reference |
63
+ | elementwise | sub | `(32,)` | TensorStudio | 0.0027 | 0.0030 | 0.0025 | 0.0042 | 0.0006 | 0.3366 | n/a | n/a | 4.2809 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
64
+ | elementwise | sub | `(32,)` | NumPy | 0.0009 | 0.0011 | 0.0007 | 0.0016 | 0.0004 | 0.3366 | n/a | n/a | 4.2809 | no | n/a | n/a | yes | NumPy | NumPy baseline |
65
+ | elementwise | sub | `(32,)` | JAX CPU dispatch | 0.0116 | 0.0117 | 0.0115 | 0.0120 | 0.0002 | 0.3366 | n/a | n/a | 4.2809 | no | n/a | n/a | yes | NumPy | reference |
66
+ | elementwise | mul | `(32,)` | TensorStudio | 0.0028 | 0.0030 | 0.0024 | 0.0043 | 0.0007 | 0.2792 | n/a | n/a | 4.4103 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
67
+ | elementwise | mul | `(32,)` | NumPy | 0.0008 | 0.0008 | 0.0007 | 0.0010 | 0.0001 | 0.2792 | n/a | n/a | 4.4103 | no | n/a | n/a | yes | NumPy | NumPy baseline |
68
+ | elementwise | mul | `(32,)` | JAX CPU dispatch | 0.0122 | 0.0123 | 0.0116 | 0.0131 | 0.0005 | 0.2792 | n/a | n/a | 4.4103 | no | n/a | n/a | yes | NumPy | reference |
69
+ | elementwise | div | `(32,)` | TensorStudio | 0.0027 | 0.0027 | 0.0025 | 0.0029 | 0.0001 | 0.2858 | n/a | n/a | 3.4552 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
70
+ | elementwise | div | `(32,)` | NumPy | 0.0008 | 0.0009 | 0.0008 | 0.0011 | 0.0001 | 0.2858 | n/a | n/a | 3.4552 | no | n/a | n/a | yes | NumPy | NumPy baseline |
71
+ | elementwise | div | `(32,)` | JAX CPU dispatch | 0.0092 | 0.0093 | 0.0089 | 0.0101 | 0.0004 | 0.2858 | n/a | n/a | 3.4552 | no | n/a | n/a | yes | NumPy | reference |
72
+ | elementwise | chain_relu((x*2+1)/3) | `(32,)` | TensorStudio | 0.0097 | 0.0097 | 0.0094 | 0.0101 | 0.0002 | 0.6673 | n/a | n/a | 9.6285 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
73
+ | elementwise | chain_relu((x*2+1)/3) | `(32,)` | NumPy | 0.0064 | 0.0063 | 0.0057 | 0.0066 | 0.0003 | 0.6673 | n/a | n/a | 9.6285 | no | n/a | n/a | yes | NumPy | NumPy baseline |
74
+ | elementwise | chain_relu((x*2+1)/3) | `(32,)` | JAX CPU dispatch | 0.0931 | 0.0938 | 0.0897 | 0.0999 | 0.0034 | 0.6673 | n/a | n/a | 9.6285 | no | n/a | n/a | yes | NumPy | reference |
75
+ | elementwise | add | `(128,)` | TensorStudio | 0.0027 | 0.0027 | 0.0026 | 0.0029 | 0.0001 | 0.2631 | n/a | n/a | 4.4624 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
76
+ | elementwise | add | `(128,)` | NumPy | 0.0007 | 0.0007 | 0.0007 | 0.0009 | 0.0001 | 0.2631 | n/a | n/a | 4.4624 | no | n/a | n/a | yes | NumPy | NumPy baseline |
77
+ | elementwise | add | `(128,)` | JAX CPU dispatch | 0.0119 | 0.0120 | 0.0112 | 0.0132 | 0.0007 | 0.2631 | n/a | n/a | 4.4624 | no | n/a | n/a | yes | NumPy | reference |
78
+ | elementwise | sub | `(128,)` | TensorStudio | 0.0027 | 0.0029 | 0.0026 | 0.0034 | 0.0004 | 0.2600 | n/a | n/a | 4.1726 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
79
+ | elementwise | sub | `(128,)` | NumPy | 0.0007 | 0.0008 | 0.0007 | 0.0010 | 0.0001 | 0.2600 | n/a | n/a | 4.1726 | no | n/a | n/a | yes | NumPy | NumPy baseline |
80
+ | elementwise | sub | `(128,)` | JAX CPU dispatch | 0.0115 | 0.0116 | 0.0110 | 0.0121 | 0.0004 | 0.2600 | n/a | n/a | 4.1726 | no | n/a | n/a | yes | NumPy | reference |
81
+ | elementwise | mul | `(128,)` | TensorStudio | 0.0028 | 0.0030 | 0.0026 | 0.0034 | 0.0004 | 0.2797 | n/a | n/a | 4.2982 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
82
+ | elementwise | mul | `(128,)` | NumPy | 0.0008 | 0.0009 | 0.0008 | 0.0013 | 0.0002 | 0.2797 | n/a | n/a | 4.2982 | no | n/a | n/a | yes | NumPy | NumPy baseline |
83
+ | elementwise | mul | `(128,)` | JAX CPU dispatch | 0.0122 | 0.0122 | 0.0116 | 0.0129 | 0.0005 | 0.2797 | n/a | n/a | 4.2982 | no | n/a | n/a | yes | NumPy | reference |
84
+ | elementwise | div | `(128,)` | TensorStudio | 0.0029 | 0.0036 | 0.0026 | 0.0048 | 0.0010 | 0.2942 | n/a | n/a | 3.2582 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
85
+ | elementwise | div | `(128,)` | NumPy | 0.0009 | 0.0009 | 0.0007 | 0.0011 | 0.0001 | 0.2942 | n/a | n/a | 3.2582 | no | n/a | n/a | yes | NumPy | NumPy baseline |
86
+ | elementwise | div | `(128,)` | JAX CPU dispatch | 0.0096 | 0.0096 | 0.0093 | 0.0102 | 0.0003 | 0.2942 | n/a | n/a | 3.2582 | no | n/a | n/a | yes | NumPy | reference |
87
+ | elementwise | chain_relu((x*2+1)/3) | `(128,)` | TensorStudio | 0.0100 | 0.0127 | 0.0099 | 0.0227 | 0.0050 | 0.5998 | n/a | n/a | 8.9405 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
88
+ | elementwise | chain_relu((x*2+1)/3) | `(128,)` | NumPy | 0.0060 | 0.0063 | 0.0059 | 0.0071 | 0.0005 | 0.5998 | n/a | n/a | 8.9405 | no | n/a | n/a | yes | NumPy | NumPy baseline |
89
+ | elementwise | chain_relu((x*2+1)/3) | `(128,)` | JAX CPU dispatch | 0.0894 | 0.0927 | 0.0886 | 0.1009 | 0.0048 | 0.5998 | n/a | n/a | 8.9405 | no | n/a | n/a | yes | NumPy | reference |
90
+ | elementwise | add | `(1024,)` | TensorStudio | 0.0036 | 0.0036 | 0.0031 | 0.0041 | 0.0004 | 0.2806 | n/a | n/a | 4.9749 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
91
+ | elementwise | add | `(1024,)` | NumPy | 0.0010 | 0.0010 | 0.0010 | 0.0010 | 0.0000 | 0.2806 | n/a | n/a | 4.9749 | no | n/a | n/a | yes | NumPy | NumPy baseline |
92
+ | elementwise | add | `(1024,)` | JAX CPU dispatch | 0.0181 | 0.0182 | 0.0175 | 0.0191 | 0.0005 | 0.2806 | n/a | n/a | 4.9749 | no | n/a | n/a | yes | NumPy | reference |
93
+ | elementwise | sub | `(1024,)` | TensorStudio | 0.0030 | 0.0031 | 0.0030 | 0.0036 | 0.0002 | 0.3588 | n/a | n/a | 5.6346 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
94
+ | elementwise | sub | `(1024,)` | NumPy | 0.0011 | 0.0011 | 0.0010 | 0.0014 | 0.0001 | 0.3588 | n/a | n/a | 5.6346 | no | n/a | n/a | yes | NumPy | NumPy baseline |
95
+ | elementwise | sub | `(1024,)` | JAX CPU dispatch | 0.0171 | 0.0166 | 0.0142 | 0.0180 | 0.0013 | 0.3588 | n/a | n/a | 5.6346 | no | n/a | n/a | yes | NumPy | reference |
96
+ | elementwise | mul | `(1024,)` | TensorStudio | 0.0030 | 0.0032 | 0.0029 | 0.0035 | 0.0002 | 0.3595 | n/a | n/a | 5.9298 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
97
+ | elementwise | mul | `(1024,)` | NumPy | 0.0011 | 0.0011 | 0.0011 | 0.0011 | 0.0000 | 0.3595 | n/a | n/a | 5.9298 | no | n/a | n/a | yes | NumPy | NumPy baseline |
98
+ | elementwise | mul | `(1024,)` | JAX CPU dispatch | 0.0179 | 0.0164 | 0.0126 | 0.0189 | 0.0025 | 0.3595 | n/a | n/a | 5.9298 | no | n/a | n/a | yes | NumPy | reference |
99
+ | elementwise | div | `(1024,)` | TensorStudio | 0.0036 | 0.0035 | 0.0032 | 0.0041 | 0.0003 | 0.3111 | n/a | n/a | 3.9700 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
100
+ | elementwise | div | `(1024,)` | NumPy | 0.0011 | 0.0011 | 0.0011 | 0.0011 | 0.0000 | 0.3111 | n/a | n/a | 3.9700 | no | n/a | n/a | yes | NumPy | NumPy baseline |
101
+ | elementwise | div | `(1024,)` | JAX CPU dispatch | 0.0141 | 0.0139 | 0.0127 | 0.0146 | 0.0007 | 0.3111 | n/a | n/a | 3.9700 | no | n/a | n/a | yes | NumPy | reference |
102
+ | elementwise | chain_relu((x*2+1)/3) | `(1024,)` | TensorStudio | 0.0121 | 0.0129 | 0.0118 | 0.0150 | 0.0012 | 0.6188 | n/a | n/a | 13.2424 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
103
+ | elementwise | chain_relu((x*2+1)/3) | `(1024,)` | NumPy | 0.0075 | 0.0081 | 0.0074 | 0.0098 | 0.0009 | 0.6188 | n/a | n/a | 13.2424 | no | n/a | n/a | yes | NumPy | NumPy baseline |
104
+ | elementwise | chain_relu((x*2+1)/3) | `(1024,)` | JAX CPU dispatch | 0.1606 | 0.1591 | 0.1525 | 0.1614 | 0.0034 | 0.6188 | n/a | n/a | 13.2424 | no | n/a | n/a | yes | NumPy | reference |
105
+ | elementwise | add | `(4096,)` | TensorStudio | 0.0048 | 0.0050 | 0.0047 | 0.0057 | 0.0004 | 0.3565 | n/a | n/a | 4.0521 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
106
+ | elementwise | add | `(4096,)` | NumPy | 0.0017 | 0.0017 | 0.0016 | 0.0017 | 0.0000 | 0.3565 | n/a | n/a | 4.0521 | no | n/a | n/a | yes | NumPy | NumPy baseline |
107
+ | elementwise | add | `(4096,)` | JAX CPU dispatch | 0.0192 | 0.0192 | 0.0184 | 0.0198 | 0.0005 | 0.3565 | n/a | n/a | 4.0521 | no | n/a | n/a | yes | NumPy | reference |
108
+ | elementwise | sub | `(4096,)` | TensorStudio | 0.0046 | 0.0047 | 0.0045 | 0.0049 | 0.0001 | 0.3165 | n/a | n/a | 3.6751 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
109
+ | elementwise | sub | `(4096,)` | NumPy | 0.0015 | 0.0015 | 0.0015 | 0.0015 | 0.0000 | 0.3165 | n/a | n/a | 3.6751 | no | n/a | n/a | yes | NumPy | NumPy baseline |
110
+ | elementwise | sub | `(4096,)` | JAX CPU dispatch | 0.0170 | 0.0158 | 0.0122 | 0.0191 | 0.0030 | 0.3165 | n/a | n/a | 3.6751 | no | n/a | n/a | yes | NumPy | reference |
111
+ | elementwise | mul | `(4096,)` | TensorStudio | 0.0048 | 0.0051 | 0.0044 | 0.0059 | 0.0007 | 0.3480 | n/a | n/a | 4.0870 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
112
+ | elementwise | mul | `(4096,)` | NumPy | 0.0017 | 0.0018 | 0.0016 | 0.0022 | 0.0002 | 0.3480 | n/a | n/a | 4.0870 | no | n/a | n/a | yes | NumPy | NumPy baseline |
113
+ | elementwise | mul | `(4096,)` | JAX CPU dispatch | 0.0195 | 0.0190 | 0.0167 | 0.0203 | 0.0012 | 0.3480 | n/a | n/a | 4.0870 | no | n/a | n/a | yes | NumPy | reference |
114
+ | elementwise | div | `(4096,)` | TensorStudio | 0.0055 | 0.0055 | 0.0054 | 0.0058 | 0.0001 | 0.3327 | n/a | n/a | 2.6388 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
115
+ | elementwise | div | `(4096,)` | NumPy | 0.0018 | 0.0021 | 0.0017 | 0.0029 | 0.0004 | 0.3327 | n/a | n/a | 2.6388 | no | n/a | n/a | yes | NumPy | NumPy baseline |
116
+ | elementwise | div | `(4096,)` | JAX CPU dispatch | 0.0146 | 0.0134 | 0.0100 | 0.0155 | 0.0023 | 0.3327 | n/a | n/a | 2.6388 | no | n/a | n/a | yes | NumPy | reference |
117
+ | elementwise | chain_relu((x*2+1)/3) | `(4096,)` | TensorStudio | 0.0202 | 0.0198 | 0.0189 | 0.0204 | 0.0006 | 0.5374 | n/a | n/a | 8.1573 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
118
+ | elementwise | chain_relu((x*2+1)/3) | `(4096,)` | NumPy | 0.0108 | 0.0108 | 0.0102 | 0.0116 | 0.0005 | 0.5374 | n/a | n/a | 8.1573 | no | n/a | n/a | yes | NumPy | NumPy baseline |
119
+ | elementwise | chain_relu((x*2+1)/3) | `(4096,)` | JAX CPU dispatch | 0.1646 | 0.1623 | 0.1517 | 0.1660 | 0.0053 | 0.5374 | n/a | n/a | 8.1573 | no | n/a | n/a | yes | NumPy | reference |
120
+ | elementwise | add | `(16384,)` | TensorStudio | 0.0113 | 0.0117 | 0.0105 | 0.0130 | 0.0009 | 0.3327 | n/a | n/a | 1.9066 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
121
+ | elementwise | add | `(16384,)` | NumPy | 0.0038 | 0.0038 | 0.0037 | 0.0038 | 0.0000 | 0.3327 | n/a | n/a | 1.9066 | no | n/a | n/a | yes | NumPy | NumPy baseline |
122
+ | elementwise | add | `(16384,)` | JAX CPU dispatch | 0.0216 | 0.0224 | 0.0212 | 0.0262 | 0.0019 | 0.3327 | n/a | n/a | 1.9066 | no | n/a | n/a | yes | NumPy | reference |
123
+ | elementwise | sub | `(16384,)` | TensorStudio | 0.0113 | 0.0113 | 0.0099 | 0.0131 | 0.0011 | 0.3442 | n/a | n/a | 1.2470 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
124
+ | elementwise | sub | `(16384,)` | NumPy | 0.0039 | 0.0040 | 0.0038 | 0.0043 | 0.0002 | 0.3442 | n/a | n/a | 1.2470 | no | n/a | n/a | yes | NumPy | NumPy baseline |
125
+ | elementwise | sub | `(16384,)` | JAX CPU dispatch | 0.0141 | 0.0143 | 0.0140 | 0.0150 | 0.0004 | 0.3442 | n/a | n/a | 1.2470 | no | n/a | n/a | yes | NumPy | reference |
126
+ | elementwise | mul | `(16384,)` | TensorStudio | 0.0158 | 0.0149 | 0.0102 | 0.0210 | 0.0038 | 0.2569 | n/a | n/a | 0.8206 | no | n/a | n/a | no | NumPy | loss vs NumPy |
127
+ | elementwise | mul | `(16384,)` | NumPy | 0.0041 | 0.0041 | 0.0040 | 0.0045 | 0.0002 | 0.2569 | n/a | n/a | 0.8206 | no | n/a | n/a | no | NumPy | NumPy baseline |
128
+ | elementwise | mul | `(16384,)` | JAX CPU dispatch | 0.0130 | 0.0133 | 0.0129 | 0.0140 | 0.0005 | 0.2569 | n/a | n/a | 0.8206 | no | n/a | n/a | no | NumPy | reference |
129
+ | elementwise | div | `(16384,)` | TensorStudio | 0.0132 | 0.0139 | 0.0132 | 0.0162 | 0.0012 | 0.3312 | n/a | n/a | 1.3575 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
130
+ | elementwise | div | `(16384,)` | NumPy | 0.0044 | 0.0045 | 0.0044 | 0.0047 | 0.0001 | 0.3312 | n/a | n/a | 1.3575 | no | n/a | n/a | yes | NumPy | NumPy baseline |
131
+ | elementwise | div | `(16384,)` | JAX CPU dispatch | 0.0180 | 0.0206 | 0.0170 | 0.0281 | 0.0042 | 0.3312 | n/a | n/a | 1.3575 | no | n/a | n/a | yes | NumPy | reference |
132
+ | elementwise | chain_relu((x*2+1)/3) | `(16384,)` | TensorStudio | 0.0471 | 0.0469 | 0.0448 | 0.0486 | 0.0013 | 0.4607 | n/a | n/a | 3.6296 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
133
+ | elementwise | chain_relu((x*2+1)/3) | `(16384,)` | NumPy | 0.0217 | 0.0226 | 0.0216 | 0.0264 | 0.0019 | 0.4607 | n/a | n/a | 3.6296 | no | n/a | n/a | yes | NumPy | NumPy baseline |
134
+ | elementwise | chain_relu((x*2+1)/3) | `(16384,)` | JAX CPU dispatch | 0.1708 | 0.1711 | 0.1685 | 0.1742 | 0.0023 | 0.4607 | n/a | n/a | 3.6296 | no | n/a | n/a | yes | NumPy | reference |
135
+ | activations | relu | `(1,)` | TensorStudio | 0.0017 | 0.0018 | 0.0017 | 0.0021 | 0.0002 | 1.2054 | n/a | n/a | 25.0949 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
136
+ | activations | relu | `(1,)` | NumPy | 0.0021 | 0.0020 | 0.0018 | 0.0022 | 0.0002 | 1.2054 | n/a | n/a | 25.0949 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
137
+ | activations | relu | `(1,)` | JAX CPU dispatch | 0.0435 | 0.0436 | 0.0431 | 0.0441 | 0.0004 | 1.2054 | n/a | n/a | 25.0949 | yes | n/a | n/a | yes | TensorStudio | reference |
138
+ | activations | sigmoid | `(1,)` | TensorStudio | 0.0018 | 0.0019 | 0.0018 | 0.0021 | 0.0001 | 2.6209 | n/a | n/a | 40.3871 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
139
+ | activations | sigmoid | `(1,)` | NumPy | 0.0047 | 0.0049 | 0.0046 | 0.0055 | 0.0003 | 2.6209 | n/a | n/a | 40.3871 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
140
+ | activations | sigmoid | `(1,)` | JAX CPU dispatch | 0.0721 | 0.0723 | 0.0714 | 0.0739 | 0.0009 | 2.6209 | n/a | n/a | 40.3871 | yes | n/a | n/a | yes | TensorStudio | reference |
141
+ | activations | tanh | `(1,)` | TensorStudio | 0.0018 | 0.0019 | 0.0018 | 0.0020 | 0.0001 | 0.6913 | n/a | n/a | 16.5592 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
142
+ | activations | tanh | `(1,)` | NumPy | 0.0012 | 0.0014 | 0.0012 | 0.0018 | 0.0002 | 0.6913 | n/a | n/a | 16.5592 | no | n/a | n/a | yes | NumPy | NumPy baseline |
143
+ | activations | tanh | `(1,)` | JAX CPU dispatch | 0.0299 | 0.0301 | 0.0292 | 0.0319 | 0.0009 | 0.6913 | n/a | n/a | 16.5592 | no | n/a | n/a | yes | NumPy | reference |
144
+ | activations | exp | `(1,)` | TensorStudio | 0.0024 | 0.0023 | 0.0019 | 0.0026 | 0.0002 | 0.5332 | n/a | n/a | 12.4549 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
145
+ | activations | exp | `(1,)` | NumPy | 0.0013 | 0.0014 | 0.0013 | 0.0018 | 0.0002 | 0.5332 | n/a | n/a | 12.4549 | no | n/a | n/a | yes | NumPy | NumPy baseline |
146
+ | activations | exp | `(1,)` | JAX CPU dispatch | 0.0299 | 0.0302 | 0.0293 | 0.0320 | 0.0010 | 0.5332 | n/a | n/a | 12.4549 | no | n/a | n/a | yes | NumPy | reference |
147
+ | activations | log | `(1,)` | TensorStudio | 0.0020 | 0.0020 | 0.0018 | 0.0022 | 0.0002 | 0.6746 | n/a | n/a | 14.9008 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
148
+ | activations | log | `(1,)` | NumPy | 0.0014 | 0.0013 | 0.0013 | 0.0014 | 0.0000 | 0.6746 | n/a | n/a | 14.9008 | no | n/a | n/a | yes | NumPy | NumPy baseline |
149
+ | activations | log | `(1,)` | JAX CPU dispatch | 0.0299 | 0.0300 | 0.0294 | 0.0310 | 0.0006 | 0.6746 | n/a | n/a | 14.9008 | no | n/a | n/a | yes | NumPy | reference |
150
+ | activations | relu | `(8,)` | TensorStudio | 0.0017 | 0.0018 | 0.0017 | 0.0020 | 0.0001 | 1.0598 | n/a | n/a | 25.9586 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
151
+ | activations | relu | `(8,)` | NumPy | 0.0018 | 0.0019 | 0.0018 | 0.0022 | 0.0002 | 1.0598 | n/a | n/a | 25.9586 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
152
+ | activations | relu | `(8,)` | JAX CPU dispatch | 0.0451 | 0.0458 | 0.0442 | 0.0491 | 0.0017 | 1.0598 | n/a | n/a | 25.9586 | yes | n/a | n/a | yes | TensorStudio | reference |
153
+ | activations | sigmoid | `(8,)` | TensorStudio | 0.0018 | 0.0019 | 0.0018 | 0.0020 | 0.0001 | 2.5449 | n/a | n/a | 49.5471 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
154
+ | activations | sigmoid | `(8,)` | NumPy | 0.0047 | 0.0047 | 0.0045 | 0.0048 | 0.0001 | 2.5449 | n/a | n/a | 49.5471 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
155
+ | activations | sigmoid | `(8,)` | JAX CPU dispatch | 0.0916 | 0.0945 | 0.0737 | 0.1350 | 0.0224 | 2.5449 | n/a | n/a | 49.5471 | yes | n/a | n/a | yes | TensorStudio | reference |
156
+ | activations | tanh | `(8,)` | TensorStudio | 0.0018 | 0.0019 | 0.0018 | 0.0024 | 0.0002 | 0.7568 | n/a | n/a | 15.8006 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
157
+ | activations | tanh | `(8,)` | NumPy | 0.0014 | 0.0014 | 0.0012 | 0.0017 | 0.0001 | 0.7568 | n/a | n/a | 15.8006 | no | n/a | n/a | yes | NumPy | NumPy baseline |
158
+ | activations | tanh | `(8,)` | JAX CPU dispatch | 0.0289 | 0.0291 | 0.0288 | 0.0296 | 0.0003 | 0.7568 | n/a | n/a | 15.8006 | no | n/a | n/a | yes | NumPy | reference |
159
+ | activations | exp | `(8,)` | TensorStudio | 0.0018 | 0.0020 | 0.0018 | 0.0026 | 0.0003 | 0.7478 | n/a | n/a | 16.5986 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
160
+ | activations | exp | `(8,)` | NumPy | 0.0014 | 0.0014 | 0.0012 | 0.0017 | 0.0002 | 0.7478 | n/a | n/a | 16.5986 | no | n/a | n/a | yes | NumPy | NumPy baseline |
161
+ | activations | exp | `(8,)` | JAX CPU dispatch | 0.0300 | 0.0301 | 0.0298 | 0.0305 | 0.0003 | 0.7478 | n/a | n/a | 16.5986 | no | n/a | n/a | yes | NumPy | reference |
162
+ | activations | log | `(8,)` | TensorStudio | 0.0019 | 0.0019 | 0.0018 | 0.0020 | 0.0001 | 0.7768 | n/a | n/a | 15.8175 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
163
+ | activations | log | `(8,)` | NumPy | 0.0015 | 0.0015 | 0.0013 | 0.0016 | 0.0001 | 0.7768 | n/a | n/a | 15.8175 | no | n/a | n/a | yes | NumPy | NumPy baseline |
164
+ | activations | log | `(8,)` | JAX CPU dispatch | 0.0295 | 0.0297 | 0.0286 | 0.0316 | 0.0010 | 0.7768 | n/a | n/a | 15.8175 | no | n/a | n/a | yes | NumPy | reference |
165
+ | activations | relu | `(32,)` | TensorStudio | 0.0018 | 0.0018 | 0.0017 | 0.0020 | 0.0001 | 1.0232 | n/a | n/a | 23.8213 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
166
+ | activations | relu | `(32,)` | NumPy | 0.0018 | 0.0019 | 0.0018 | 0.0021 | 0.0001 | 1.0232 | n/a | n/a | 23.8213 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
167
+ | activations | relu | `(32,)` | JAX CPU dispatch | 0.0430 | 0.0433 | 0.0427 | 0.0446 | 0.0007 | 1.0232 | n/a | n/a | 23.8213 | yes | n/a | n/a | yes | TensorStudio | reference |
168
+ | activations | sigmoid | `(32,)` | TensorStudio | 0.0020 | 0.0020 | 0.0020 | 0.0021 | 0.0001 | 2.4028 | n/a | n/a | 37.3031 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
169
+ | activations | sigmoid | `(32,)` | NumPy | 0.0047 | 0.0049 | 0.0047 | 0.0052 | 0.0002 | 2.4028 | n/a | n/a | 37.3031 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
170
+ | activations | sigmoid | `(32,)` | JAX CPU dispatch | 0.0737 | 0.0735 | 0.0725 | 0.0742 | 0.0006 | 2.4028 | n/a | n/a | 37.3031 | yes | n/a | n/a | yes | TensorStudio | reference |
171
+ | activations | tanh | `(32,)` | TensorStudio | 0.0022 | 0.0024 | 0.0021 | 0.0028 | 0.0003 | 0.5848 | n/a | n/a | 13.1374 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
172
+ | activations | tanh | `(32,)` | NumPy | 0.0013 | 0.0014 | 0.0013 | 0.0017 | 0.0002 | 0.5848 | n/a | n/a | 13.1374 | no | n/a | n/a | yes | NumPy | NumPy baseline |
173
+ | activations | tanh | `(32,)` | JAX CPU dispatch | 0.0295 | 0.0301 | 0.0290 | 0.0328 | 0.0014 | 0.5848 | n/a | n/a | 13.1374 | no | n/a | n/a | yes | NumPy | reference |
174
+ | activations | exp | `(32,)` | TensorStudio | 0.0021 | 0.0021 | 0.0020 | 0.0024 | 0.0001 | 0.6041 | n/a | n/a | 13.9319 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
175
+ | activations | exp | `(32,)` | NumPy | 0.0013 | 0.0013 | 0.0013 | 0.0015 | 0.0001 | 0.6041 | n/a | n/a | 13.9319 | no | n/a | n/a | yes | NumPy | NumPy baseline |
176
+ | activations | exp | `(32,)` | JAX CPU dispatch | 0.0297 | 0.0300 | 0.0289 | 0.0324 | 0.0013 | 0.6041 | n/a | n/a | 13.9319 | no | n/a | n/a | yes | NumPy | reference |
177
+ | activations | log | `(32,)` | TensorStudio | 0.0021 | 0.0022 | 0.0019 | 0.0025 | 0.0003 | 0.6153 | n/a | n/a | 14.4325 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
178
+ | activations | log | `(32,)` | NumPy | 0.0013 | 0.0014 | 0.0013 | 0.0018 | 0.0002 | 0.6153 | n/a | n/a | 14.4325 | no | n/a | n/a | yes | NumPy | NumPy baseline |
179
+ | activations | log | `(32,)` | JAX CPU dispatch | 0.0300 | 0.0301 | 0.0293 | 0.0310 | 0.0005 | 0.6153 | n/a | n/a | 14.4325 | no | n/a | n/a | yes | NumPy | reference |
180
+ | activations | relu | `(128,)` | TensorStudio | 0.0019 | 0.0019 | 0.0018 | 0.0020 | 0.0001 | 0.9802 | n/a | n/a | 22.8329 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
181
+ | activations | relu | `(128,)` | NumPy | 0.0019 | 0.0019 | 0.0019 | 0.0021 | 0.0001 | 0.9802 | n/a | n/a | 22.8329 | no | n/a | n/a | yes | NumPy | NumPy baseline |
182
+ | activations | relu | `(128,)` | JAX CPU dispatch | 0.0441 | 0.0440 | 0.0429 | 0.0453 | 0.0008 | 0.9802 | n/a | n/a | 22.8329 | no | n/a | n/a | yes | NumPy | reference |
183
+ | activations | sigmoid | `(128,)` | TensorStudio | 0.0028 | 0.0028 | 0.0026 | 0.0031 | 0.0002 | 1.7769 | n/a | n/a | 25.6381 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
184
+ | activations | sigmoid | `(128,)` | NumPy | 0.0050 | 0.0051 | 0.0046 | 0.0060 | 0.0005 | 1.7769 | n/a | n/a | 25.6381 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
185
+ | activations | sigmoid | `(128,)` | JAX CPU dispatch | 0.0721 | 0.0728 | 0.0713 | 0.0761 | 0.0017 | 1.7769 | n/a | n/a | 25.6381 | yes | n/a | n/a | yes | TensorStudio | reference |
186
+ | activations | tanh | `(128,)` | TensorStudio | 0.0030 | 0.0030 | 0.0029 | 0.0031 | 0.0001 | 0.5311 | n/a | n/a | 10.0826 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
187
+ | activations | tanh | `(128,)` | NumPy | 0.0016 | 0.0017 | 0.0015 | 0.0020 | 0.0002 | 0.5311 | n/a | n/a | 10.0826 | no | n/a | n/a | yes | NumPy | NumPy baseline |
188
+ | activations | tanh | `(128,)` | JAX CPU dispatch | 0.0301 | 0.0302 | 0.0295 | 0.0311 | 0.0005 | 0.5311 | n/a | n/a | 10.0826 | no | n/a | n/a | yes | NumPy | reference |
189
+ | activations | exp | `(128,)` | TensorStudio | 0.0027 | 0.0027 | 0.0025 | 0.0029 | 0.0001 | 0.5038 | n/a | n/a | 11.1650 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
190
+ | activations | exp | `(128,)` | NumPy | 0.0014 | 0.0014 | 0.0013 | 0.0018 | 0.0002 | 0.5038 | n/a | n/a | 11.1650 | no | n/a | n/a | yes | NumPy | NumPy baseline |
191
+ | activations | exp | `(128,)` | JAX CPU dispatch | 0.0301 | 0.0299 | 0.0292 | 0.0303 | 0.0004 | 0.5038 | n/a | n/a | 11.1650 | no | n/a | n/a | yes | NumPy | reference |
192
+ | activations | log | `(128,)` | TensorStudio | 0.0028 | 0.0029 | 0.0027 | 0.0032 | 0.0002 | 0.5353 | n/a | n/a | 10.9066 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
193
+ | activations | log | `(128,)` | NumPy | 0.0015 | 0.0016 | 0.0015 | 0.0017 | 0.0001 | 0.5353 | n/a | n/a | 10.9066 | no | n/a | n/a | yes | NumPy | NumPy baseline |
194
+ | activations | log | `(128,)` | JAX CPU dispatch | 0.0304 | 0.0302 | 0.0288 | 0.0312 | 0.0008 | 0.5353 | n/a | n/a | 10.9066 | no | n/a | n/a | yes | NumPy | reference |
195
+ | activations | relu | `(1024,)` | TensorStudio | 0.0051 | 0.0043 | 0.0024 | 0.0061 | 0.0016 | 0.4881 | n/a | n/a | 16.0759 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
196
+ | activations | relu | `(1024,)` | NumPy | 0.0025 | 0.0025 | 0.0025 | 0.0025 | 0.0000 | 0.4881 | n/a | n/a | 16.0759 | no | n/a | n/a | yes | NumPy | NumPy baseline |
197
+ | activations | relu | `(1024,)` | JAX CPU dispatch | 0.0819 | 0.0770 | 0.0564 | 0.0834 | 0.0103 | 0.4881 | n/a | n/a | 16.0759 | no | n/a | n/a | yes | NumPy | reference |
198
+ | activations | sigmoid | `(1024,)` | TensorStudio | 0.0092 | 0.0095 | 0.0092 | 0.0104 | 0.0005 | 0.7810 | n/a | n/a | 14.6484 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
199
+ | activations | sigmoid | `(1024,)` | NumPy | 0.0072 | 0.0076 | 0.0072 | 0.0088 | 0.0006 | 0.7810 | n/a | n/a | 14.6484 | no | n/a | n/a | yes | NumPy | NumPy baseline |
200
+ | activations | sigmoid | `(1024,)` | JAX CPU dispatch | 0.1353 | 0.1360 | 0.1346 | 0.1377 | 0.0013 | 0.7810 | n/a | n/a | 14.6484 | no | n/a | n/a | yes | NumPy | reference |
201
+ | activations | tanh | `(1024,)` | TensorStudio | 0.0139 | 0.0140 | 0.0136 | 0.0146 | 0.0004 | 0.2809 | n/a | n/a | 2.1942 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
202
+ | activations | tanh | `(1024,)` | NumPy | 0.0039 | 0.0039 | 0.0038 | 0.0040 | 0.0001 | 0.2809 | n/a | n/a | 2.1942 | no | n/a | n/a | yes | NumPy | NumPy baseline |
203
+ | activations | tanh | `(1024,)` | JAX CPU dispatch | 0.0305 | 0.0319 | 0.0292 | 0.0386 | 0.0034 | 0.2809 | n/a | n/a | 2.1942 | no | n/a | n/a | yes | NumPy | reference |
204
+ | activations | exp | `(1024,)` | TensorStudio | 0.0089 | 0.0090 | 0.0081 | 0.0099 | 0.0007 | 0.2969 | n/a | n/a | 3.5192 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
205
+ | activations | exp | `(1024,)` | NumPy | 0.0027 | 0.0027 | 0.0026 | 0.0027 | 0.0000 | 0.2969 | n/a | n/a | 3.5192 | no | n/a | n/a | yes | NumPy | NumPy baseline |
206
+ | activations | exp | `(1024,)` | JAX CPU dispatch | 0.0315 | 0.0313 | 0.0309 | 0.0316 | 0.0003 | 0.2969 | n/a | n/a | 3.5192 | no | n/a | n/a | yes | NumPy | reference |
207
+ | activations | log | `(1024,)` | TensorStudio | 0.0086 | 0.0087 | 0.0084 | 0.0092 | 0.0003 | 0.3322 | n/a | n/a | 3.6510 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
208
+ | activations | log | `(1024,)` | NumPy | 0.0029 | 0.0031 | 0.0028 | 0.0036 | 0.0003 | 0.3322 | n/a | n/a | 3.6510 | no | n/a | n/a | yes | NumPy | NumPy baseline |
209
+ | activations | log | `(1024,)` | JAX CPU dispatch | 0.0313 | 0.0316 | 0.0308 | 0.0333 | 0.0009 | 0.3322 | n/a | n/a | 3.6510 | no | n/a | n/a | yes | NumPy | reference |
210
+ | activations | relu | `(4096,)` | TensorStudio | 0.0053 | 0.0053 | 0.0050 | 0.0058 | 0.0003 | 0.7256 | n/a | n/a | 15.5506 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
211
+ | activations | relu | `(4096,)` | NumPy | 0.0038 | 0.0039 | 0.0038 | 0.0044 | 0.0002 | 0.7256 | n/a | n/a | 15.5506 | no | n/a | n/a | yes | NumPy | NumPy baseline |
212
+ | activations | relu | `(4096,)` | JAX CPU dispatch | 0.0820 | 0.0823 | 0.0808 | 0.0844 | 0.0014 | 0.7256 | n/a | n/a | 15.5506 | no | n/a | n/a | yes | NumPy | reference |
213
+ | activations | sigmoid | `(4096,)` | TensorStudio | 0.0324 | 0.0324 | 0.0322 | 0.0327 | 0.0002 | 0.3669 | n/a | n/a | 3.2790 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
214
+ | activations | sigmoid | `(4096,)` | NumPy | 0.0119 | 0.0125 | 0.0116 | 0.0138 | 0.0009 | 0.3669 | n/a | n/a | 3.2790 | no | n/a | n/a | yes | NumPy | NumPy baseline |
215
+ | activations | sigmoid | `(4096,)` | JAX CPU dispatch | 0.1063 | 0.1031 | 0.0864 | 0.1159 | 0.0101 | 0.3669 | n/a | n/a | 3.2790 | no | n/a | n/a | yes | NumPy | reference |
216
+ | activations | tanh | `(4096,)` | TensorStudio | 0.0661 | 0.0690 | 0.0603 | 0.0876 | 0.0101 | 0.1742 | n/a | n/a | 0.4973 | no | n/a | n/a | no | NumPy | loss vs NumPy |
217
+ | activations | tanh | `(4096,)` | NumPy | 0.0115 | 0.0136 | 0.0106 | 0.0196 | 0.0034 | 0.1742 | n/a | n/a | 0.4973 | no | n/a | n/a | no | NumPy | NumPy baseline |
218
+ | activations | tanh | `(4096,)` | JAX CPU dispatch | 0.0329 | 0.0336 | 0.0322 | 0.0369 | 0.0017 | 0.1742 | n/a | n/a | 0.4973 | no | n/a | n/a | no | NumPy | reference |
219
+ | activations | exp | `(4096,)` | TensorStudio | 0.0302 | 0.0316 | 0.0281 | 0.0381 | 0.0037 | 0.2002 | n/a | n/a | 1.1157 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
220
+ | activations | exp | `(4096,)` | NumPy | 0.0060 | 0.0067 | 0.0059 | 0.0090 | 0.0012 | 0.2002 | n/a | n/a | 1.1157 | no | n/a | n/a | yes | NumPy | NumPy baseline |
221
+ | activations | exp | `(4096,)` | JAX CPU dispatch | 0.0336 | 0.0336 | 0.0333 | 0.0340 | 0.0002 | 0.2002 | n/a | n/a | 1.1157 | no | n/a | n/a | yes | NumPy | reference |
222
+ | activations | log | `(4096,)` | TensorStudio | 0.0303 | 0.0303 | 0.0301 | 0.0305 | 0.0002 | 0.2441 | n/a | n/a | 1.1674 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
223
+ | activations | log | `(4096,)` | NumPy | 0.0074 | 0.0076 | 0.0074 | 0.0082 | 0.0003 | 0.2441 | n/a | n/a | 1.1674 | no | n/a | n/a | yes | NumPy | NumPy baseline |
224
+ | activations | log | `(4096,)` | JAX CPU dispatch | 0.0353 | 0.0357 | 0.0346 | 0.0368 | 0.0009 | 0.2441 | n/a | n/a | 1.1674 | no | n/a | n/a | yes | NumPy | reference |
225
+ | activations | relu | `(16384,)` | TensorStudio | 0.0136 | 0.0152 | 0.0132 | 0.0217 | 0.0032 | 0.6651 | n/a | n/a | 6.7101 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
226
+ | activations | relu | `(16384,)` | NumPy | 0.0090 | 0.0094 | 0.0090 | 0.0111 | 0.0008 | 0.6651 | n/a | n/a | 6.7101 | no | n/a | n/a | yes | NumPy | NumPy baseline |
227
+ | activations | relu | `(16384,)` | JAX CPU dispatch | 0.0910 | 0.0915 | 0.0892 | 0.0945 | 0.0020 | 0.6651 | n/a | n/a | 6.7101 | no | n/a | n/a | yes | NumPy | reference |
228
+ | activations | sigmoid | `(16384,)` | TensorStudio | 0.1241 | 0.1238 | 0.1205 | 0.1257 | 0.0019 | 0.2796 | n/a | n/a | 1.2600 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
229
+ | activations | sigmoid | `(16384,)` | NumPy | 0.0347 | 0.0353 | 0.0339 | 0.0371 | 0.0012 | 0.2796 | n/a | n/a | 1.2600 | no | n/a | n/a | yes | NumPy | NumPy baseline |
230
+ | activations | sigmoid | `(16384,)` | JAX CPU dispatch | 0.1563 | 0.1550 | 0.1498 | 0.1574 | 0.0028 | 0.2796 | n/a | n/a | 1.2600 | no | n/a | n/a | yes | NumPy | reference |
231
+ | activations | tanh | `(16384,)` | TensorStudio | 0.2519 | 0.2483 | 0.2386 | 0.2566 | 0.0080 | 0.1508 | n/a | n/a | 0.1509 | no | n/a | n/a | no | NumPy | loss vs NumPy |
232
+ | activations | tanh | `(16384,)` | NumPy | 0.0380 | 0.0424 | 0.0377 | 0.0537 | 0.0062 | 0.1508 | n/a | n/a | 0.1509 | no | n/a | n/a | no | NumPy | NumPy baseline |
233
+ | activations | tanh | `(16384,)` | JAX CPU dispatch | 0.0380 | 0.0393 | 0.0377 | 0.0430 | 0.0020 | 0.1508 | n/a | n/a | 0.1509 | no | n/a | n/a | no | NumPy | reference |
234
+ | activations | exp | `(16384,)` | TensorStudio | 0.1050 | 0.1040 | 0.1007 | 0.1075 | 0.0025 | 0.1777 | n/a | n/a | 0.3935 | no | n/a | n/a | no | NumPy | loss vs NumPy |
235
+ | activations | exp | `(16384,)` | NumPy | 0.0187 | 0.0190 | 0.0184 | 0.0205 | 0.0008 | 0.1777 | n/a | n/a | 0.3935 | no | n/a | n/a | no | NumPy | NumPy baseline |
236
+ | activations | exp | `(16384,)` | JAX CPU dispatch | 0.0413 | 0.0412 | 0.0390 | 0.0438 | 0.0018 | 0.1777 | n/a | n/a | 0.3935 | no | n/a | n/a | no | NumPy | reference |
237
+ | activations | log | `(16384,)` | TensorStudio | 0.1131 | 0.1126 | 0.1082 | 0.1174 | 0.0030 | 0.2184 | n/a | n/a | 0.3879 | no | n/a | n/a | no | NumPy | loss vs NumPy |
238
+ | activations | log | `(16384,)` | NumPy | 0.0247 | 0.0253 | 0.0246 | 0.0271 | 0.0009 | 0.2184 | n/a | n/a | 0.3879 | no | n/a | n/a | no | NumPy | NumPy baseline |
239
+ | activations | log | `(16384,)` | JAX CPU dispatch | 0.0439 | 0.0447 | 0.0432 | 0.0478 | 0.0017 | 0.2184 | n/a | n/a | 0.3879 | no | n/a | n/a | no | NumPy | reference |
240
+ | reductions | sum | `(1,)` | TensorStudio | 0.0022 | 0.0023 | 0.0021 | 0.0026 | 0.0002 | 1.0375 | n/a | n/a | 5.2655 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
241
+ | reductions | sum | `(1,)` | NumPy | 0.0023 | 0.0023 | 0.0022 | 0.0025 | 0.0001 | 1.0375 | n/a | n/a | 5.2655 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
242
+ | reductions | sum | `(1,)` | JAX CPU dispatch | 0.0115 | 0.0115 | 0.0112 | 0.0117 | 0.0002 | 1.0375 | n/a | n/a | 5.2655 | yes | n/a | n/a | yes | TensorStudio | reference |
243
+ | reductions | mean | `(1,)` | TensorStudio | 0.0021 | 0.0024 | 0.0021 | 0.0033 | 0.0004 | 4.1243 | n/a | n/a | 5.2052 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
244
+ | reductions | mean | `(1,)` | NumPy | 0.0088 | 0.0087 | 0.0084 | 0.0092 | 0.0003 | 4.1243 | n/a | n/a | 5.2052 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
245
+ | reductions | mean | `(1,)` | JAX CPU dispatch | 0.0111 | 0.0112 | 0.0108 | 0.0119 | 0.0004 | 4.1243 | n/a | n/a | 5.2052 | yes | n/a | n/a | yes | TensorStudio | reference |
246
+ | reductions | sum | `(8,)` | TensorStudio | 0.0022 | 0.0022 | 0.0022 | 0.0023 | 0.0001 | 1.0613 | n/a | n/a | 5.6130 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
247
+ | reductions | sum | `(8,)` | NumPy | 0.0023 | 0.0025 | 0.0023 | 0.0028 | 0.0002 | 1.0613 | n/a | n/a | 5.6130 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
248
+ | reductions | sum | `(8,)` | JAX CPU dispatch | 0.0121 | 0.0122 | 0.0119 | 0.0127 | 0.0003 | 1.0613 | n/a | n/a | 5.6130 | yes | n/a | n/a | yes | TensorStudio | reference |
249
+ | reductions | mean | `(8,)` | TensorStudio | 0.0022 | 0.0023 | 0.0022 | 0.0025 | 0.0001 | 4.0185 | n/a | n/a | 5.6748 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
250
+ | reductions | mean | `(8,)` | NumPy | 0.0087 | 0.0095 | 0.0086 | 0.0116 | 0.0012 | 4.0185 | n/a | n/a | 5.6748 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
251
+ | reductions | mean | `(8,)` | JAX CPU dispatch | 0.0123 | 0.0122 | 0.0115 | 0.0128 | 0.0004 | 4.0185 | n/a | n/a | 5.6748 | yes | n/a | n/a | yes | TensorStudio | reference |
252
+ | reductions | sum | `(32,)` | TensorStudio | 0.0025 | 0.0025 | 0.0023 | 0.0027 | 0.0001 | 0.9256 | n/a | n/a | 5.3569 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
253
+ | reductions | sum | `(32,)` | NumPy | 0.0023 | 0.0023 | 0.0022 | 0.0026 | 0.0001 | 0.9256 | n/a | n/a | 5.3569 | no | n/a | n/a | yes | NumPy | NumPy baseline |
254
+ | reductions | sum | `(32,)` | JAX CPU dispatch | 0.0131 | 0.0131 | 0.0126 | 0.0139 | 0.0004 | 0.9256 | n/a | n/a | 5.3569 | no | n/a | n/a | yes | NumPy | reference |
255
+ | reductions | mean | `(32,)` | TensorStudio | 0.0024 | 0.0025 | 0.0023 | 0.0027 | 0.0002 | 3.7243 | n/a | n/a | 4.8115 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
256
+ | reductions | mean | `(32,)` | NumPy | 0.0089 | 0.0089 | 0.0085 | 0.0093 | 0.0003 | 3.7243 | n/a | n/a | 4.8115 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
257
+ | reductions | mean | `(32,)` | JAX CPU dispatch | 0.0115 | 0.0115 | 0.0110 | 0.0122 | 0.0004 | 3.7243 | n/a | n/a | 4.8115 | yes | n/a | n/a | yes | TensorStudio | reference |
258
+ | reductions | sum | `(128,)` | TensorStudio | 0.0032 | 0.0031 | 0.0028 | 0.0037 | 0.0003 | 0.7673 | n/a | n/a | 4.1468 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
259
+ | reductions | sum | `(128,)` | NumPy | 0.0024 | 0.0027 | 0.0023 | 0.0038 | 0.0006 | 0.7673 | n/a | n/a | 4.1468 | no | n/a | n/a | yes | NumPy | NumPy baseline |
260
+ | reductions | sum | `(128,)` | JAX CPU dispatch | 0.0132 | 0.0133 | 0.0131 | 0.0138 | 0.0002 | 0.7673 | n/a | n/a | 4.1468 | no | n/a | n/a | yes | NumPy | reference |
261
+ | reductions | mean | `(128,)` | TensorStudio | 0.0030 | 0.0030 | 0.0029 | 0.0031 | 0.0001 | 2.8256 | n/a | n/a | 4.2652 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
262
+ | reductions | mean | `(128,)` | NumPy | 0.0085 | 0.0085 | 0.0084 | 0.0088 | 0.0002 | 2.8256 | n/a | n/a | 4.2652 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
263
+ | reductions | mean | `(128,)` | JAX CPU dispatch | 0.0128 | 0.0130 | 0.0122 | 0.0145 | 0.0008 | 2.8256 | n/a | n/a | 4.2652 | yes | n/a | n/a | yes | TensorStudio | reference |
264
+ | reductions | sum | `(1024,)` | TensorStudio | 0.0085 | 0.0088 | 0.0083 | 0.0095 | 0.0005 | 0.3132 | n/a | n/a | 2.2445 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
265
+ | reductions | sum | `(1024,)` | NumPy | 0.0027 | 0.0027 | 0.0026 | 0.0027 | 0.0000 | 0.3132 | n/a | n/a | 2.2445 | no | n/a | n/a | yes | NumPy | NumPy baseline |
266
+ | reductions | sum | `(1024,)` | JAX CPU dispatch | 0.0191 | 0.0185 | 0.0147 | 0.0205 | 0.0020 | 0.3132 | n/a | n/a | 2.2445 | no | n/a | n/a | yes | NumPy | reference |
267
+ | reductions | mean | `(1024,)` | TensorStudio | 0.0084 | 0.0086 | 0.0083 | 0.0097 | 0.0005 | 1.1476 | n/a | n/a | 2.1492 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
268
+ | reductions | mean | `(1024,)` | NumPy | 0.0096 | 0.0096 | 0.0086 | 0.0110 | 0.0009 | 1.1476 | n/a | n/a | 2.1492 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
269
+ | reductions | mean | `(1024,)` | JAX CPU dispatch | 0.0180 | 0.0167 | 0.0141 | 0.0185 | 0.0019 | 1.1476 | n/a | n/a | 2.1492 | yes | n/a | n/a | yes | TensorStudio | reference |
270
+ | reductions | sum | `(4096,)` | TensorStudio | 0.0284 | 0.0284 | 0.0282 | 0.0289 | 0.0003 | 0.1262 | n/a | n/a | 0.7003 | no | n/a | n/a | no | NumPy | loss vs NumPy |
271
+ | reductions | sum | `(4096,)` | NumPy | 0.0036 | 0.0038 | 0.0036 | 0.0045 | 0.0004 | 0.1262 | n/a | n/a | 0.7003 | no | n/a | n/a | no | NumPy | NumPy baseline |
272
+ | reductions | sum | `(4096,)` | JAX CPU dispatch | 0.0199 | 0.0186 | 0.0150 | 0.0213 | 0.0023 | 0.1262 | n/a | n/a | 0.7003 | no | n/a | n/a | no | NumPy | reference |
273
+ | reductions | mean | `(4096,)` | TensorStudio | 0.0278 | 0.0282 | 0.0273 | 0.0302 | 0.0011 | 0.3866 | n/a | n/a | 0.6819 | no | n/a | n/a | no | NumPy | loss vs NumPy |
274
+ | reductions | mean | `(4096,)` | NumPy | 0.0108 | 0.0112 | 0.0100 | 0.0127 | 0.0012 | 0.3866 | n/a | n/a | 0.6819 | no | n/a | n/a | no | NumPy | NumPy baseline |
275
+ | reductions | mean | `(4096,)` | JAX CPU dispatch | 0.0190 | 0.0182 | 0.0148 | 0.0199 | 0.0018 | 0.3866 | n/a | n/a | 0.6819 | no | n/a | n/a | no | NumPy | reference |
276
+ | reductions | sum | `(16384,)` | TensorStudio | 0.1155 | 0.1149 | 0.1124 | 0.1169 | 0.0017 | 0.0623 | n/a | n/a | 0.2400 | no | n/a | n/a | no | NumPy | loss vs NumPy |
277
+ | reductions | sum | `(16384,)` | NumPy | 0.0072 | 0.0078 | 0.0070 | 0.0107 | 0.0014 | 0.0623 | n/a | n/a | 0.2400 | no | n/a | n/a | no | NumPy | NumPy baseline |
278
+ | reductions | sum | `(16384,)` | JAX CPU dispatch | 0.0277 | 0.0283 | 0.0268 | 0.0299 | 0.0012 | 0.0623 | n/a | n/a | 0.2400 | no | n/a | n/a | no | NumPy | reference |
279
+ | reductions | mean | `(16384,)` | TensorStudio | 0.1029 | 0.1178 | 0.1022 | 0.1694 | 0.0261 | 0.1284 | n/a | n/a | 0.2503 | no | n/a | n/a | no | NumPy | loss vs NumPy |
280
+ | reductions | mean | `(16384,)` | NumPy | 0.0132 | 0.0144 | 0.0132 | 0.0179 | 0.0018 | 0.1284 | n/a | n/a | 0.2503 | no | n/a | n/a | no | NumPy | NumPy baseline |
281
+ | reductions | mean | `(16384,)` | JAX CPU dispatch | 0.0258 | 0.0260 | 0.0253 | 0.0271 | 0.0006 | 0.1284 | n/a | n/a | 0.2503 | no | n/a | n/a | no | NumPy | reference |
282
+ | reductions | sum_axis1 | `(16, 16)` | TensorStudio | 0.0027 | 0.0029 | 0.0027 | 0.0035 | 0.0003 | 1.0159 | n/a | n/a | 4.8244 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
283
+ | reductions | sum_axis1 | `(16, 16)` | NumPy | 0.0028 | 0.0029 | 0.0027 | 0.0032 | 0.0002 | 1.0159 | n/a | n/a | 4.8244 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
284
+ | reductions | sum_axis1 | `(16, 16)` | JAX CPU dispatch | 0.0131 | 0.0131 | 0.0123 | 0.0141 | 0.0006 | 1.0159 | n/a | n/a | 4.8244 | yes | n/a | n/a | yes | TensorStudio | reference |
285
+ | reductions | mean_axis0 | `(16, 16)` | TensorStudio | 0.0027 | 0.0029 | 0.0027 | 0.0038 | 0.0005 | 6.9616 | n/a | n/a | 4.3981 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
286
+ | reductions | mean_axis0 | `(16, 16)` | NumPy | 0.0188 | 0.0173 | 0.0107 | 0.0224 | 0.0044 | 6.9616 | n/a | n/a | 4.3981 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
287
+ | reductions | mean_axis0 | `(16, 16)` | JAX CPU dispatch | 0.0119 | 0.0121 | 0.0110 | 0.0137 | 0.0009 | 6.9616 | n/a | n/a | 4.3981 | yes | n/a | n/a | yes | TensorStudio | reference |
288
+ | reductions | sum_axis1 | `(64, 64)` | TensorStudio | 0.0056 | 0.0057 | 0.0055 | 0.0060 | 0.0002 | 0.7324 | n/a | n/a | 3.6234 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
289
+ | reductions | sum_axis1 | `(64, 64)` | NumPy | 0.0041 | 0.0042 | 0.0041 | 0.0047 | 0.0002 | 0.7324 | n/a | n/a | 3.6234 | no | n/a | n/a | yes | NumPy | NumPy baseline |
290
+ | reductions | sum_axis1 | `(64, 64)` | JAX CPU dispatch | 0.0205 | 0.0190 | 0.0158 | 0.0208 | 0.0020 | 0.7324 | n/a | n/a | 3.6234 | no | n/a | n/a | yes | NumPy | reference |
291
+ | reductions | mean_axis0 | `(64, 64)` | TensorStudio | 0.0037 | 0.0038 | 0.0037 | 0.0041 | 0.0002 | 3.4748 | n/a | n/a | 5.0602 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
292
+ | reductions | mean_axis0 | `(64, 64)` | NumPy | 0.0129 | 0.0129 | 0.0120 | 0.0135 | 0.0005 | 3.4748 | n/a | n/a | 5.0602 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
293
+ | reductions | mean_axis0 | `(64, 64)` | JAX CPU dispatch | 0.0188 | 0.0174 | 0.0130 | 0.0203 | 0.0026 | 3.4748 | n/a | n/a | 5.0602 | yes | n/a | n/a | yes | TensorStudio | reference |
294
+ | reductions | sum_axis1 | `(128, 128)` | TensorStudio | 0.0162 | 0.0162 | 0.0160 | 0.0164 | 0.0001 | 0.4971 | n/a | n/a | 1.5494 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
295
+ | reductions | sum_axis1 | `(128, 128)` | NumPy | 0.0081 | 0.0081 | 0.0080 | 0.0081 | 0.0000 | 0.4971 | n/a | n/a | 1.5494 | no | n/a | n/a | yes | NumPy | NumPy baseline |
296
+ | reductions | sum_axis1 | `(128, 128)` | JAX CPU dispatch | 0.0251 | 0.0227 | 0.0175 | 0.0255 | 0.0033 | 0.4971 | n/a | n/a | 1.5494 | no | n/a | n/a | yes | NumPy | reference |
297
+ | reductions | mean_axis0 | `(128, 128)` | TensorStudio | 0.0066 | 0.0066 | 0.0066 | 0.0066 | 0.0000 | 2.6388 | n/a | n/a | 2.1855 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
298
+ | reductions | mean_axis0 | `(128, 128)` | NumPy | 0.0174 | 0.0167 | 0.0150 | 0.0182 | 0.0014 | 2.6388 | n/a | n/a | 2.1855 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
299
+ | reductions | mean_axis0 | `(128, 128)` | JAX CPU dispatch | 0.0144 | 0.0162 | 0.0128 | 0.0207 | 0.0035 | 2.6388 | n/a | n/a | 2.1855 | yes | n/a | n/a | yes | TensorStudio | reference |
300
+ | reductions | sum_axis1 | `(256, 256)` | TensorStudio | 0.0648 | 0.0656 | 0.0642 | 0.0680 | 0.0015 | 0.3345 | n/a | n/a | 1.1645 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
301
+ | reductions | sum_axis1 | `(256, 256)` | NumPy | 0.0217 | 0.0227 | 0.0215 | 0.0259 | 0.0017 | 0.3345 | n/a | n/a | 1.1645 | no | n/a | n/a | yes | NumPy | NumPy baseline |
302
+ | reductions | sum_axis1 | `(256, 256)` | JAX CPU dispatch | 0.0754 | 0.0763 | 0.0728 | 0.0810 | 0.0027 | 0.3345 | n/a | n/a | 1.1645 | no | n/a | n/a | yes | NumPy | reference |
303
+ | reductions | mean_axis0 | `(256, 256)` | TensorStudio | 0.0184 | 0.0186 | 0.0183 | 0.0194 | 0.0004 | 1.2350 | n/a | n/a | 1.3719 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
304
+ | reductions | mean_axis0 | `(256, 256)` | NumPy | 0.0227 | 0.0250 | 0.0223 | 0.0297 | 0.0032 | 1.2350 | n/a | n/a | 1.3719 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
305
+ | reductions | mean_axis0 | `(256, 256)` | JAX CPU dispatch | 0.0252 | 0.0255 | 0.0246 | 0.0269 | 0.0009 | 1.2350 | n/a | n/a | 1.3719 | yes | n/a | n/a | yes | TensorStudio | reference |
306
+ | matmul | matmul | `(16, 16)` | TensorStudio | 0.0034 | 0.0034 | 0.0034 | 0.0034 | 0.0000 | 1.0573 | n/a | n/a | 2.6977 | yes | n/a | n/a | yes | TensorStudio | win vs NumPy |
307
+ | matmul | matmul | `(16, 16)` | NumPy | 0.0036 | 0.0036 | 0.0028 | 0.0044 | 0.0007 | 1.0573 | n/a | n/a | 2.6977 | yes | n/a | n/a | yes | TensorStudio | NumPy baseline |
308
+ | matmul | matmul | `(16, 16)` | JAX CPU dispatch | 0.0091 | 0.0113 | 0.0090 | 0.0152 | 0.0027 | 1.0573 | n/a | n/a | 2.6977 | yes | n/a | n/a | yes | TensorStudio | reference |
309
+ | matmul | matmul | `(64, 64)` | TensorStudio | 0.0345 | 0.0368 | 0.0343 | 0.0463 | 0.0047 | 0.3155 | n/a | n/a | 1.4296 | no | n/a | n/a | yes | NumPy | loss vs NumPy |
310
+ | matmul | matmul | `(64, 64)` | NumPy | 0.0109 | 0.0109 | 0.0108 | 0.0110 | 0.0001 | 0.3155 | n/a | n/a | 1.4296 | no | n/a | n/a | yes | NumPy | NumPy baseline |
311
+ | matmul | matmul | `(64, 64)` | JAX CPU dispatch | 0.0494 | 0.0428 | 0.0232 | 0.0519 | 0.0107 | 0.3155 | n/a | n/a | 1.4296 | no | n/a | n/a | yes | NumPy | reference |
312
+ | matmul | matmul | `(128, 128)` | TensorStudio | 0.2620 | 0.2663 | 0.2610 | 0.2799 | 0.0072 | 1.3408 | n/a | n/a | 0.4619 | yes | n/a | n/a | no | JAX CPU dispatch | win vs NumPy |
313
+ | matmul | matmul | `(128, 128)` | NumPy | 0.3513 | 0.3511 | 0.3144 | 0.3951 | 0.0259 | 1.3408 | n/a | n/a | 0.4619 | yes | n/a | n/a | no | JAX CPU dispatch | NumPy baseline |
314
+ | matmul | matmul | `(128, 128)` | JAX CPU dispatch | 0.1210 | 0.1196 | 0.1126 | 0.1284 | 0.0062 | 1.3408 | n/a | n/a | 0.4619 | yes | n/a | n/a | no | JAX CPU dispatch | reference |
315
+ | matmul | matmul | `(256, 256)` | TensorStudio | 2.2228 | 2.2646 | 2.2090 | 2.3544 | 0.0617 | 0.2021 | n/a | n/a | 0.1261 | no | n/a | n/a | no | JAX CPU dispatch | loss vs NumPy |
316
+ | matmul | matmul | `(256, 256)` | NumPy | 0.4492 | 0.4439 | 0.4029 | 0.4722 | 0.0254 | 0.2021 | n/a | n/a | 0.1261 | no | n/a | n/a | no | JAX CPU dispatch | NumPy baseline |
317
+ | matmul | matmul | `(256, 256)` | JAX CPU dispatch | 0.2803 | 0.2786 | 0.2731 | 0.2821 | 0.0032 | 0.2021 | n/a | n/a | 0.1261 | no | n/a | n/a | no | JAX CPU dispatch | reference |
318
+ | conv2d | conv2d_3x3_padding1 | `(1, 1, 8, 8)` | TensorStudio | 0.1517 | 0.1518 | 0.1497 | 0.1545 | 0.0016 | 8.5189 | n/a | n/a | 0.7899 | yes | n/a | n/a | no | JAX CPU dispatch | win vs NumPy |
319
+ | conv2d | conv2d_3x3_padding1 | `(1, 1, 8, 8)` | NumPy | 1.2925 | 1.3066 | 1.2823 | 1.3362 | 0.0242 | 8.5189 | n/a | n/a | 0.7899 | yes | n/a | n/a | no | JAX CPU dispatch | NumPy baseline |
320
+ | conv2d | conv2d_3x3_padding1 | `(1, 1, 8, 8)` | JAX CPU dispatch | 0.1198 | 0.1211 | 0.1164 | 0.1270 | 0.0043 | 8.5189 | n/a | n/a | 0.7899 | yes | n/a | n/a | no | JAX CPU dispatch | reference |
321
+ | conv2d | conv2d_3x3_padding1 | `(4, 3, 16, 16)` | TensorStudio | 7.5418 | 7.5529 | 7.5164 | 7.6132 | 0.0360 | 8.4226 | n/a | n/a | 0.0411 | yes | n/a | n/a | no | JAX CPU dispatch | win vs NumPy |
322
+ | conv2d | conv2d_3x3_padding1 | `(4, 3, 16, 16)` | NumPy | 63.5214 | 63.7343 | 62.8244 | 64.6576 | 0.6405 | 8.4226 | n/a | n/a | 0.0411 | yes | n/a | n/a | no | JAX CPU dispatch | NumPy baseline |
323
+ | conv2d | conv2d_3x3_padding1 | `(4, 3, 16, 16)` | JAX CPU dispatch | 0.3099 | 0.3194 | 0.3036 | 0.3439 | 0.0156 | 8.4226 | n/a | n/a | 0.0411 | yes | n/a | n/a | no | JAX CPU dispatch | reference |
324
+ | pooling | max_pool2d_2x2 | `(1, 1, 16, 16)` | TensorStudio | 0.0144 | 0.0149 | 0.0143 | 0.0168 | 0.0010 | 12.2930 | n/a | n/a | n/a | yes | n/a | n/a | n/a | TensorStudio | win vs NumPy |
325
+ | pooling | max_pool2d_2x2 | `(1, 1, 16, 16)` | NumPy | 0.1765 | 0.1762 | 0.1734 | 0.1806 | 0.0026 | 12.2930 | n/a | n/a | n/a | yes | n/a | n/a | n/a | TensorStudio | NumPy baseline |
326
+ | pooling | avg_pool2d_2x2 | `(1, 1, 16, 16)` | TensorStudio | 0.0156 | 0.0152 | 0.0140 | 0.0166 | 0.0011 | 39.5541 | n/a | n/a | n/a | yes | n/a | n/a | n/a | TensorStudio | win vs NumPy |
327
+ | pooling | avg_pool2d_2x2 | `(1, 1, 16, 16)` | NumPy | 0.6162 | 0.6166 | 0.5966 | 0.6521 | 0.0196 | 39.5541 | n/a | n/a | n/a | yes | n/a | n/a | n/a | TensorStudio | NumPy baseline |
328
+ | pooling | max_pool2d_2x2 | `(4, 3, 32, 32)` | TensorStudio | 0.5568 | 0.5904 | 0.5423 | 0.6611 | 0.0485 | 15.2546 | n/a | n/a | n/a | yes | n/a | n/a | n/a | TensorStudio | win vs NumPy |
329
+ | pooling | max_pool2d_2x2 | `(4, 3, 32, 32)` | NumPy | 8.4943 | 8.5757 | 8.3747 | 8.8608 | 0.1833 | 15.2546 | n/a | n/a | n/a | yes | n/a | n/a | n/a | TensorStudio | NumPy baseline |
330
+ | pooling | avg_pool2d_2x2 | `(4, 3, 32, 32)` | TensorStudio | 0.5530 | 0.5529 | 0.5463 | 0.5626 | 0.0055 | 53.3335 | n/a | n/a | n/a | yes | n/a | n/a | n/a | TensorStudio | win vs NumPy |
331
+ | pooling | avg_pool2d_2x2 | `(4, 3, 32, 32)` | NumPy | 29.4954 | 29.7615 | 29.3748 | 30.6629 | 0.4739 | 53.3335 | n/a | n/a | n/a | yes | n/a | n/a | n/a | TensorStudio | NumPy baseline |
332
+ | autograd | scalar_backward | `(1,)` | TensorStudio | 0.0148 | 0.0159 | 0.0143 | 0.0197 | 0.0020 | n/a | n/a | n/a | n/a | n/a | n/a | n/a | n/a | TensorStudio | no NumPy reference |
333
+ | autograd | elementwise_backward | `(1,)` | TensorStudio | 0.0307 | 0.0319 | 0.0301 | 0.0368 | 0.0025 | n/a | n/a | n/a | n/a | n/a | n/a | n/a | n/a | TensorStudio | no NumPy reference |
334
+ | autograd | scalar_backward | `(128,)` | TensorStudio | 0.1458 | 0.1640 | 0.1390 | 0.2102 | 0.0285 | n/a | n/a | n/a | n/a | n/a | n/a | n/a | n/a | TensorStudio | no NumPy reference |
335
+ | autograd | elementwise_backward | `(128,)` | TensorStudio | 0.3263 | 0.3526 | 0.3238 | 0.4602 | 0.0538 | n/a | n/a | n/a | n/a | n/a | n/a | n/a | n/a | TensorStudio | no NumPy reference |
336
+ | autograd | scalar_backward | `(1024,)` | TensorStudio | 1.0264 | 1.0239 | 1.0159 | 1.0312 | 0.0057 | n/a | n/a | n/a | n/a | n/a | n/a | n/a | n/a | TensorStudio | no NumPy reference |
337
+ | autograd | elementwise_backward | `(1024,)` | TensorStudio | 2.4146 | 2.4096 | 2.3658 | 2.4579 | 0.0302 | n/a | n/a | n/a | n/a | n/a | n/a | n/a | n/a | TensorStudio | no NumPy reference |
338
+ | training_loop | tiny_linear_regression_10_steps | `(32,)` | TensorStudio | 1.5981 | 1.5986 | 1.5944 | 1.6036 | 0.0034 | 0.1010 | n/a | n/a | n/a | no | n/a | n/a | n/a | NumPy | loss vs NumPy |
339
+ | training_loop | tiny_linear_regression_10_steps | `(32,)` | NumPy | 0.1614 | 0.1753 | 0.1602 | 0.2106 | 0.0198 | 0.1010 | n/a | n/a | n/a | no | n/a | n/a | n/a | NumPy | NumPy baseline |
340
+
341
+ Speedup columns are `competitor median / TensorStudio median`; values above 1.0 mean
342
+ TensorStudio was faster for that specific case.
343
+ Win columns say whether that same speedup is above 1.0 for the competitor.
tensorstudio-1.3.4/docs/Autograde/index.md → tensorstudio-1.3.6/docs/autograd/coverage.md RENAMED
@@ -40,7 +40,7 @@ Calling `backward()`:
40
40
 
41
41
  ## Supported Gradient Operations
42
42
 
43
- `1.3.4` supports gradients for:
43
+ `1.3.6` supports gradients for:
44
44
 
45
45
  - `add`
46
46
  - `sub`
tensorstudio-1.3.6/docs/autograd/gradient-notes.md ADDED
@@ -0,0 +1,46 @@
1
+ # Gradient Notes
2
+
3
+ TensorStudio uses eager reverse-mode autodiff. Each differentiable native op can
4
+ attach a small backward function to its output tensor.
5
+
6
+ ## Scalar Backward
7
+
8
+ Calling `backward()` without an explicit gradient is valid for scalar outputs:
9
+
10
+ ```python
11
+ import tensorstudio as ts
12
+
13
+ x = ts.tensor([1.0, 2.0, 3.0], requires_grad=True)
14
+ loss = (x * x).sum()
15
+ loss.backward()
16
+ print(x.grad.tolist())
17
+ ```
18
+
19
+ For non-scalar outputs, pass an explicit gradient tensor.
20
+
21
+ ## Broadcasting Gradients
22
+
23
+ When a forward operation broadcasts an input, the backward pass sums gradients
24
+ back to the original input shape.
25
+
26
+ ```python
27
+ x = ts.ones((2, 3), requires_grad=True)
28
+ b = ts.ones((3,), requires_grad=True)
29
+ (x + b).sum().backward()
30
+ print(b.grad.tolist()) # [2.0, 2.0, 2.0]
31
+ ```
32
+
33
+ ## Selection Gradients
34
+
35
+ `where(condition, a, b)` does not differentiate through the boolean condition.
36
+ Gradients flow to `a` where the condition is true and to `b` where it is false.
37
+
38
+ `maximum(a, b)` and `minimum(a, b)` split gradients evenly when values are tied.
39
+ This matches the tie-handling used by TensorStudio's extrema reductions and
40
+ avoids unstable left-or-right-only behavior.
41
+
42
+ ## Non-Differentiable Outputs
43
+
44
+ Integer and boolean tensors do not carry autograd history. Arg reductions return
45
+ `int64` indices and are intentionally non-differentiable.
46
+