trainkernel 0.1.0__tar.gz

trainkernel-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 [yyyy] [name of copyright owner]
+
+   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.

trainkernel-0.1.0/NOTICE

@@ -0,0 +1,6 @@
+TrainKernel
+
+Copyright 2026 Arya Somu.
+
+This distribution includes third-party dependencies that remain under their
+respective licenses. See LICENSE for the project license.

trainkernel-0.1.0/PKG-INFO

@@ -0,0 +1,169 @@
+Metadata-Version: 2.4
+Name: trainkernel
+Version: 0.1.0
+Summary: Compile a complete static PyTorch training step into one CUDA kernel
+Author: Arya Somu
+License-Expression: Apache-2.0
+Project-URL: Homepage, https://github.com/arpjw/trainkernel
+Project-URL: Documentation, https://github.com/arpjw/trainkernel#readme
+Project-URL: Issues, https://github.com/arpjw/trainkernel/issues
+Project-URL: Source, https://github.com/arpjw/trainkernel
+Keywords: pytorch,cuda,compiler,training,gpu
+Classifier: Development Status :: 3 - Alpha
+Classifier: Environment :: GPU :: NVIDIA CUDA
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Science/Research
+Classifier: Operating System :: POSIX :: Linux
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+License-File: NOTICE
+Requires-Dist: torch>=2.8
+Requires-Dist: ninja>=1.11
+Provides-Extra: dev
+Requires-Dist: build>=1.2; extra == "dev"
+Requires-Dist: pytest>=8; extra == "dev"
+Requires-Dist: twine>=5; extra == "dev"
+Dynamic: license-file
+
+# TrainKernel
+
+TrainKernel compiles a supported static PyTorch training step into one
+cooperative CUDA megakernel. The generated kernel can include the forward pass,
+loss, gradients, and SGD update, so a captured step has no per-operator Python
+launch overhead.
+
+The current backend schedules operators in topological order, uses grid barriers for dependencies, inlines single-consumer pointwise operations, and reuses temporary storage after its last reader. BF16 and FP16 matrix multiplications use WMMA tensor cores. FP32 uses a tiled SIMT implementation for accuracy.
+
+This is a scoped v0.1 research release for static functional training on NVIDIA
+SM80 GPUs. It is not a general PyTorch compiler and does not claim universal
+speedups.
+
+| A100 workload | TrainKernel | Fastest measured Inductor | TrainKernel / Inductor |
+| --- | ---: | ---: | ---: |
+| tiny online MLP | 15.28 μs | 22.85 μs | **0.669, 1.49x faster** |
+| online MLP | 20.32 μs | 24.24 μs | **0.838, 1.19x faster** |
+| small MLP | 66.04 μs | 22.92 μs | 2.894, slower |
+| medium MLP | 255.32 μs | 42.19 μs | 6.052, slower |
+
+These are complete forward, backward, and update steps captured as CUDA Graphs.
+The [held-out report](https://github.com/arpjw/trainkernel/blob/main/results/a100-optimization/README.md) publishes all seeds,
+raw timings, correctness checks, baseline failures, and the two regressions.
+
+An independent release run adds a one-hot least-squares classification head.
+Across three seeds, TrainKernel's in-place path measured 13.07 us versus
+18.63 us for the fastest measured Inductor configuration at the median, or
+1.43x faster. The [release evidence](https://github.com/arpjw/trainkernel/blob/main/results/release-validation/README.md)
+retains all raw samples and correctness failures. Its cold-cache setup took
+53.62 seconds and required an estimated 7.82 million updates to amortize, so
+this is a steady-state small-workload result rather than a claim about
+short-lived jobs.
+
+## Install
+
+Linux, Python 3.10+, PyTorch 2.8+, an NVIDIA SM80+ GPU, a CUDA toolkit with `nvcc`, and a C++ compiler are required for generated kernel execution. CPU capture and IR checks run without CUDA.
+
+```bash
+python -m pip install trainkernel
+```
+
+For development:
+
+```bash
+python -m venv --system-site-packages .venv
+source .venv/bin/activate
+python -m pip install -e '.[dev]'
+python -m pytest -q
+python -m examples.online_training
+```
+
+The backend targets SM80+, including A100 and H100. Release evidence is A100-only; H100 has not been tested and Hopper TMA/WGMMA paths are not implemented.
+
+## Compile a training step
+
+```python
+import torch
+import trainkernel
+
+def train_step(weights, x, target):
+    def loss(w):
+        return ((torch.relu(x @ w) - target) ** 2).mean()
+    gradients, value = torch.func.grad_and_value(loss)(weights)
+    return value, gradients, weights - 1e-3 * gradients
+
+compiled = trainkernel.compile_training(train_step, (weights, x, target))
+loss, gradients, weights = compiled(weights, x, target)
+```
+
+`compile_training` returns a typed `CompiledTrainingStep`. It verifies that
+gradients match state structure and derives the update mapping, including nested
+state trees. The supplied state must be detached from autograd because the
+training function computes its own gradients. `compiled.capture(...)` creates a
+CUDA Graph after warmup, restores the supplied state after warmup and capture,
+and returns a reusable `CapturedProgram`; replay outputs borrow captured buffers
+with fixed shapes and addresses. See the [100-update example](https://github.com/arpjw/trainkernel/blob/main/examples/online_training.py).
+
+Parameters may be nested dictionaries or tuples. Inputs and operations must stay on one device; resolve lazy negative or conjugate input views before compiling. Updated parameters are returned explicitly. The compiler proves direct last-read state commits where safe and stages all other updates transactionally; input mutation and arbitrary aliasing remain unsupported. Different shapes, strides, devices, or dtypes require another compilation.
+
+Inspect the compiler without a GPU:
+
+```python
+program, output_tree = trainkernel.capture(train_step, cpu_examples)
+from trainkernel.codegen import generate
+cuda_source = generate(program).source
+```
+
+## Benchmark
+
+```bash
+bash scripts/run_train_loop_suite.sh results/a100-optimization/heldout
+```
+
+For a clean release-candidate check, including the useful classification-head
+workload and package build:
+
+```bash
+bash scripts/run_release_validation.sh results/release-validation
+```
+
+The [training-loop harness](https://github.com/arpjw/trainkernel/blob/main/benchmarks/train_loop.py) uses FP32 master weights, BF16 compute, FP32 loss, and FP32 SGD updates. It checks every loss, gradient, and updated parameter for 100 steps and separately checks a captured 20-step replay. Timed graphs include state feedback over four resident batches. They compare pure and in-place megakernels, captured eager PyTorch, and default and eager-rounding Inductor variants. CUDA event timings exclude compilation, capture, reset, and warmup, and retain raw samples. Numerical failures remain visible for all baselines. The [methodology](https://github.com/arpjw/trainkernel/blob/main/docs/a100-methodology.md) explains the held-out seeds and fixed tolerance policy; the [result](https://github.com/arpjw/trainkernel/blob/main/results/a100-optimization/README.md) includes both wins and regressions. The older [native-BF16 fixed-step harness](https://github.com/arpjw/trainkernel/blob/main/benchmarks/train_mlp.py) remains available for the historical report.
+
+## Supported subset
+
+- FP32, BF16, and FP16 strided tensors with static shapes.
+- Matrix multiplication and `addmm`, broadcast arithmetic, ReLU and its backward, scalar powers, sums and means, dtype casts, fills, and supported storage views.
+- Pure tensor functions, including explicit differentiation through `torch.func.grad_and_value`.
+
+Unsupported operations produce an error. There is no silent eager fallback. Attention, dropout/RNG, convolutions, dynamic shapes, arbitrary input mutation, distributed collectives, and asynchronous task-queue scheduling are not implemented.
+
+Temporary values use FP32 workspace storage with explicit rounding at low-precision boundaries. Fusion reduces materialization but does not keep an entire model on-chip. The tested larger GEMMs run faster in Inductor; the report keeps those regressions visible.
+
+## Design and evidence
+
+See [compiler design](https://github.com/arpjw/trainkernel/blob/main/docs/design.md), [A100 optimization evidence](https://github.com/arpjw/trainkernel/blob/main/results/a100-optimization/README.md), [independent release validation](https://github.com/arpjw/trainkernel/blob/main/results/release-validation/README.md), and [GPU access record](https://github.com/arpjw/trainkernel/blob/main/docs/infrastructure.md). CPU tests validate IR semantics, views, reductions, diagnostics, and arena lifetimes. The independent release suite passed 116 tests that compile and execute generated CUDA, compare training outputs across precisions, and exercise CUDA Graph replay.
+
+## Related work and roadmap
+
+Most adjacent public megakernel systems currently focus on inference:
+[HazyResearch Megakernels](https://github.com/HazyResearch/Megakernels) targets
+low-latency Llama inference, [MegaBake](https://github.com/megabake/megabake)
+captures PyTorch inference workloads for SM90, [AutoMegaKernel](https://github.com/RightNow-AI/AutoMegaKernel)
+generates Llama-family inference kernels, and [Mirage Persistent Kernel](https://arxiv.org/abs/2512.22219)
+targets multi-GPU model inference. TrainKernel instead explores compiling the
+stateful forward, loss, backward, and optimizer update of a small training step
+on A100 into one kernel. It does not claim priority for megakernel compilation
+in general.
+
+The next engineering priorities are a measured shape-aware dispatcher that can
+delegate GEMM-heavy steps to Inductor, an experimental `torch.compile` backend
+after AOTAutograd, AdamW, and additional real online-learning workloads. These
+are roadmap items, not v0.1 capabilities.
+
+## License
+
+Apache License 2.0. See [LICENSE](https://github.com/arpjw/trainkernel/blob/main/LICENSE) and [NOTICE](https://github.com/arpjw/trainkernel/blob/main/NOTICE).

trainkernel-0.1.0/README.md

@@ -0,0 +1,136 @@
+# TrainKernel
+
+TrainKernel compiles a supported static PyTorch training step into one
+cooperative CUDA megakernel. The generated kernel can include the forward pass,
+loss, gradients, and SGD update, so a captured step has no per-operator Python
+launch overhead.
+
+The current backend schedules operators in topological order, uses grid barriers for dependencies, inlines single-consumer pointwise operations, and reuses temporary storage after its last reader. BF16 and FP16 matrix multiplications use WMMA tensor cores. FP32 uses a tiled SIMT implementation for accuracy.
+
+This is a scoped v0.1 research release for static functional training on NVIDIA
+SM80 GPUs. It is not a general PyTorch compiler and does not claim universal
+speedups.
+
+| A100 workload | TrainKernel | Fastest measured Inductor | TrainKernel / Inductor |
+| --- | ---: | ---: | ---: |
+| tiny online MLP | 15.28 μs | 22.85 μs | **0.669, 1.49x faster** |
+| online MLP | 20.32 μs | 24.24 μs | **0.838, 1.19x faster** |
+| small MLP | 66.04 μs | 22.92 μs | 2.894, slower |
+| medium MLP | 255.32 μs | 42.19 μs | 6.052, slower |
+
+These are complete forward, backward, and update steps captured as CUDA Graphs.
+The [held-out report](https://github.com/arpjw/trainkernel/blob/main/results/a100-optimization/README.md) publishes all seeds,
+raw timings, correctness checks, baseline failures, and the two regressions.
+
+An independent release run adds a one-hot least-squares classification head.
+Across three seeds, TrainKernel's in-place path measured 13.07 us versus
+18.63 us for the fastest measured Inductor configuration at the median, or
+1.43x faster. The [release evidence](https://github.com/arpjw/trainkernel/blob/main/results/release-validation/README.md)
+retains all raw samples and correctness failures. Its cold-cache setup took
+53.62 seconds and required an estimated 7.82 million updates to amortize, so
+this is a steady-state small-workload result rather than a claim about
+short-lived jobs.
+
+## Install
+
+Linux, Python 3.10+, PyTorch 2.8+, an NVIDIA SM80+ GPU, a CUDA toolkit with `nvcc`, and a C++ compiler are required for generated kernel execution. CPU capture and IR checks run without CUDA.
+
+```bash
+python -m pip install trainkernel
+```
+
+For development:
+
+```bash
+python -m venv --system-site-packages .venv
+source .venv/bin/activate
+python -m pip install -e '.[dev]'
+python -m pytest -q
+python -m examples.online_training
+```
+
+The backend targets SM80+, including A100 and H100. Release evidence is A100-only; H100 has not been tested and Hopper TMA/WGMMA paths are not implemented.
+
+## Compile a training step
+
+```python
+import torch
+import trainkernel
+
+def train_step(weights, x, target):
+    def loss(w):
+        return ((torch.relu(x @ w) - target) ** 2).mean()
+    gradients, value = torch.func.grad_and_value(loss)(weights)
+    return value, gradients, weights - 1e-3 * gradients
+
+compiled = trainkernel.compile_training(train_step, (weights, x, target))
+loss, gradients, weights = compiled(weights, x, target)
+```
+
+`compile_training` returns a typed `CompiledTrainingStep`. It verifies that
+gradients match state structure and derives the update mapping, including nested
+state trees. The supplied state must be detached from autograd because the
+training function computes its own gradients. `compiled.capture(...)` creates a
+CUDA Graph after warmup, restores the supplied state after warmup and capture,
+and returns a reusable `CapturedProgram`; replay outputs borrow captured buffers
+with fixed shapes and addresses. See the [100-update example](https://github.com/arpjw/trainkernel/blob/main/examples/online_training.py).
+
+Parameters may be nested dictionaries or tuples. Inputs and operations must stay on one device; resolve lazy negative or conjugate input views before compiling. Updated parameters are returned explicitly. The compiler proves direct last-read state commits where safe and stages all other updates transactionally; input mutation and arbitrary aliasing remain unsupported. Different shapes, strides, devices, or dtypes require another compilation.
+
+Inspect the compiler without a GPU:
+
+```python
+program, output_tree = trainkernel.capture(train_step, cpu_examples)
+from trainkernel.codegen import generate
+cuda_source = generate(program).source
+```
+
+## Benchmark
+
+```bash
+bash scripts/run_train_loop_suite.sh results/a100-optimization/heldout
+```
+
+For a clean release-candidate check, including the useful classification-head
+workload and package build:
+
+```bash
+bash scripts/run_release_validation.sh results/release-validation
+```
+
+The [training-loop harness](https://github.com/arpjw/trainkernel/blob/main/benchmarks/train_loop.py) uses FP32 master weights, BF16 compute, FP32 loss, and FP32 SGD updates. It checks every loss, gradient, and updated parameter for 100 steps and separately checks a captured 20-step replay. Timed graphs include state feedback over four resident batches. They compare pure and in-place megakernels, captured eager PyTorch, and default and eager-rounding Inductor variants. CUDA event timings exclude compilation, capture, reset, and warmup, and retain raw samples. Numerical failures remain visible for all baselines. The [methodology](https://github.com/arpjw/trainkernel/blob/main/docs/a100-methodology.md) explains the held-out seeds and fixed tolerance policy; the [result](https://github.com/arpjw/trainkernel/blob/main/results/a100-optimization/README.md) includes both wins and regressions. The older [native-BF16 fixed-step harness](https://github.com/arpjw/trainkernel/blob/main/benchmarks/train_mlp.py) remains available for the historical report.
+
+## Supported subset
+
+- FP32, BF16, and FP16 strided tensors with static shapes.
+- Matrix multiplication and `addmm`, broadcast arithmetic, ReLU and its backward, scalar powers, sums and means, dtype casts, fills, and supported storage views.
+- Pure tensor functions, including explicit differentiation through `torch.func.grad_and_value`.
+
+Unsupported operations produce an error. There is no silent eager fallback. Attention, dropout/RNG, convolutions, dynamic shapes, arbitrary input mutation, distributed collectives, and asynchronous task-queue scheduling are not implemented.
+
+Temporary values use FP32 workspace storage with explicit rounding at low-precision boundaries. Fusion reduces materialization but does not keep an entire model on-chip. The tested larger GEMMs run faster in Inductor; the report keeps those regressions visible.
+
+## Design and evidence
+
+See [compiler design](https://github.com/arpjw/trainkernel/blob/main/docs/design.md), [A100 optimization evidence](https://github.com/arpjw/trainkernel/blob/main/results/a100-optimization/README.md), [independent release validation](https://github.com/arpjw/trainkernel/blob/main/results/release-validation/README.md), and [GPU access record](https://github.com/arpjw/trainkernel/blob/main/docs/infrastructure.md). CPU tests validate IR semantics, views, reductions, diagnostics, and arena lifetimes. The independent release suite passed 116 tests that compile and execute generated CUDA, compare training outputs across precisions, and exercise CUDA Graph replay.
+
+## Related work and roadmap
+
+Most adjacent public megakernel systems currently focus on inference:
+[HazyResearch Megakernels](https://github.com/HazyResearch/Megakernels) targets
+low-latency Llama inference, [MegaBake](https://github.com/megabake/megabake)
+captures PyTorch inference workloads for SM90, [AutoMegaKernel](https://github.com/RightNow-AI/AutoMegaKernel)
+generates Llama-family inference kernels, and [Mirage Persistent Kernel](https://arxiv.org/abs/2512.22219)
+targets multi-GPU model inference. TrainKernel instead explores compiling the
+stateful forward, loss, backward, and optimizer update of a small training step
+on A100 into one kernel. It does not claim priority for megakernel compilation
+in general.
+
+The next engineering priorities are a measured shape-aware dispatcher that can
+delegate GEMM-heavy steps to Inductor, an experimental `torch.compile` backend
+after AOTAutograd, AdamW, and additional real online-learning workloads. These
+are roadmap items, not v0.1 capabilities.
+
+## License
+
+Apache License 2.0. See [LICENSE](https://github.com/arpjw/trainkernel/blob/main/LICENSE) and [NOTICE](https://github.com/arpjw/trainkernel/blob/main/NOTICE).

trainkernel-0.1.0/pyproject.toml

@@ -0,0 +1,45 @@
+[build-system]
+requires = ["setuptools>=77"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "trainkernel"
+version = "0.1.0"
+description = "Compile a complete static PyTorch training step into one CUDA kernel"
+readme = "README.md"
+license = "Apache-2.0"
+license-files = ["LICENSE", "NOTICE"]
+authors = [{name = "Arya Somu"}]
+requires-python = ">=3.10"
+dependencies = ["torch>=2.8", "ninja>=1.11"]
+keywords = ["pytorch", "cuda", "compiler", "training", "gpu"]
+classifiers = [
+  "Development Status :: 3 - Alpha",
+  "Environment :: GPU :: NVIDIA CUDA",
+  "Intended Audience :: Developers",
+  "Intended Audience :: Science/Research",
+  "Operating System :: POSIX :: Linux",
+  "Programming Language :: Python :: 3",
+  "Programming Language :: Python :: 3.10",
+  "Programming Language :: Python :: 3.11",
+  "Programming Language :: Python :: 3.12",
+  "Topic :: Scientific/Engineering :: Artificial Intelligence",
+]
+
+[project.urls]
+Homepage = "https://github.com/arpjw/trainkernel"
+Documentation = "https://github.com/arpjw/trainkernel#readme"
+Issues = "https://github.com/arpjw/trainkernel/issues"
+Source = "https://github.com/arpjw/trainkernel"
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+
+[project.optional-dependencies]
+dev = ["build>=1.2", "pytest>=8", "twine>=5"]
+
+[tool.setuptools.packages.find]
+include = ["trainkernel*"]
+
+[tool.setuptools.package-data]
+trainkernel = ["py.typed"]

trainkernel-0.1.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

trainkernel-0.1.0/tests/test_benchmark_integrity.py

@@ -0,0 +1,89 @@
+import pytest
+import torch
+
+from benchmarks.train_loop import (
+    _amortization,
+    _bootstrap_speedup_ci,
+    _close_training,
+    _trajectory,
+    make_batches,
+    step_master,
+)
+
+
+def test_amortization_includes_both_methods_setup_and_handles_no_win():
+    timings = {"mega": {"median_ms": 1.0}, "reference": {"median_ms": 2.0}}
+    compilation = {"mega": 2.0, "reference": 0.5}
+    capture = {"mega": 0.2, "reference": 0.1}
+    result = _amortization("mega", "reference", timings, compilation, capture)
+    assert result["incremental_setup_seconds"] == pytest.approx(1.6)
+    assert result["break_even_steps"] == pytest.approx(1600.0)
+
+    timings["mega"]["median_ms"] = 3.0
+    slower = _amortization("mega", "reference", timings, compilation, capture)
+    assert slower["break_even_steps"] is None
+    assert slower["break_even_defined"] is False
+
+
+def test_training_policy_checks_loss_dtype_and_all_outputs():
+    params, batches = make_batches(2, 3, 4, 2, torch.bfloat16, torch.device("cpu"), 1, seed=17)
+    master = tuple(p.float() for p in params)
+    expected = step_master(master, *batches[0], 1e-3)
+    bad_loss = (expected[0].bfloat16(), expected[1], expected[2])
+    with pytest.raises(AssertionError):
+        _close_training(bad_loss, expected)
+    bad_gradients = (expected[0], tuple(g for g in expected[1][:-1]), expected[2])
+    with pytest.raises(AssertionError):
+        _close_training(bad_gradients, expected)
+
+
+def test_teacher_targets_produce_finite_long_trajectory():
+    params, batches = make_batches(4, 8, 16, 4, torch.bfloat16, torch.device("cpu"), 4, seed=211)
+    master = tuple(p.float() for p in params)
+    losses, _ = _trajectory(lambda p, x, y: step_master(p, x, y, 1e-3), master, batches, 100)
+    assert len(losses) == 100
+    assert all(torch.isfinite(torch.tensor(losses)))
+
+
+def test_classification_head_targets_are_one_hot_and_finite():
+    params, batches = make_batches(8, 6, 10, 4, torch.bfloat16, torch.device("cpu"), 3, seed=211, task="classification-head")
+    assert all(target.shape == (8, 4) for _, target in batches)
+    assert all(torch.equal(target.sum(dim=-1), torch.ones(8, dtype=torch.bfloat16)) for _, target in batches)
+    master = tuple(p.float() for p in params)
+    losses, _ = _trajectory(lambda p, x, y: step_master(p, x, y, 1e-3), master, batches, 20)
+    assert all(torch.isfinite(torch.tensor(losses)))
+
+
+def test_bootstrap_is_deterministic_and_lower_ratio_is_faster():
+    a = [1.0, 1.1, 0.9, 1.05]
+    b = [2.0, 2.1, 1.9, 2.05]
+    first = _bootstrap_speedup_ci(a, b)
+    second = _bootstrap_speedup_ci(a, b)
+    assert first == second
+    assert first["median_ratio"] < 1.0
+    assert first["draws"] == 2000 and first["seed"] == 0
+
+
+def test_report_mode_continues_after_baseline_mismatch():
+    params, batches = make_batches(2, 3, 4, 2, torch.bfloat16, torch.device("cpu"), 1, seed=307)
+    master = tuple(p.float() for p in params)
+    def bad(p, x, y):
+        loss, grads, updated = step_master(p, x, y, 1e-3)
+        return loss + 1.0, grads, updated
+    losses, _, report = _trajectory(bad, master, batches, 4, enforce=False, report=True)
+    assert len(losses) == 4
+    assert report["passed"] is False
+    assert report["first_failure_step"] == 0
+    with pytest.raises(AssertionError):
+        _trajectory(bad, master, batches, 4)
+
+
+def test_report_mode_measures_nonzero_error_for_passing_perturbation():
+    params, batches = make_batches(2, 3, 4, 2, torch.bfloat16, torch.device("cpu"), 1, seed=101)
+    master = tuple(p.float() for p in params)
+    def near(p, x, y):
+        loss, grads, updated = step_master(p, x, y, 1e-3)
+        return loss, grads, tuple(value + 1e-8 for value in updated)
+    _, _, report = _trajectory(near, master, batches, 2, enforce=False, report=True)
+    assert report["passed"] is True
+    assert report["max_absolute_error"] > 0.0