pyqitnn 0.3.6__tar.gz

pyqitnn-0.3.6/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.

pyqitnn-0.3.6/MANIFEST.in

@@ -0,0 +1,4 @@
+include README.md
+include pyproject.toml
+recursive-include csrc *.cu *.h
+recursive-include src *.cpp

pyqitnn-0.3.6/PKG-INFO

@@ -0,0 +1,443 @@
+Metadata-Version: 2.1
+Name: pyqitnn
+Version: 0.3.6
+Summary: Quantum-inspired ternary neural network layers for PyTorch
+License: Apache-2.0
+Project-URL: Homepage, https://github.com/kaifczxc-lab/PyQITNN
+Keywords: quantum,ternary,neural-network,pytorch,cuda
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Science/Research
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: C++
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Operating System :: Microsoft :: Windows
+Classifier: Operating System :: POSIX :: Linux
+Classifier: Environment :: GPU :: NVIDIA CUDA
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Provides-Extra: tokenizers
+Requires-Dist: tokenizers<0.23,>=0.22; extra == "tokenizers"
+
+# PyQITNN
+
+A PyTorch library implementing quantum-inspired ternary neural network layers.
+Runs on standard NVIDIA GPUs; no quantum hardware is required.
+
+```python
+import torch
+import pyqitnn
+
+model = pyqitnn.QITNNSimplexTransformerLM(
+    vocab_size=256,
+    dim=64,
+    ffn_dim=128,
+    seq_len=128,
+    layers=2,
+    precision_mode="fp32",
+    device="cuda:0",
+)
+tokens = torch.randint(0, 256, (2, 128), device="cuda:0")
+targets = torch.randint(0, 256, (2, 128), device="cuda:0")
+logits, loss = model(tokens, targets=targets)
+loss.backward()
+```
+
+---
+
+## What this is
+
+Every linear projection stores three amplitude vectors `(a_neg, a_zero, a_pos)` instead of
+one weight matrix. A Born-rule normalization converts amplitudes to ternary probabilities
+`(P-, P0, P+)`. The result propagates through the network as a full 2D centered simplex
+state `[x | y]`, not a collapsed scalar.
+
+This gives the network two independent degrees of freedom per output coordinate - the minimal
+complete representation of a ternary probability state.
+
+These are classical amplitudes computed on a GPU. The "quantum-inspired" part is the geometry
+and the normalization rule, not the hardware.
+
+---
+
+## Installation
+
+### Prerequisites
+
+- NVIDIA GPU with CUDA support
+- Python 3.10+ at the source level
+- PyTorch 2.0+ with CUDA support
+- Local CUDA toolkit when building from source
+
+### Current tested setup
+
+- Windows 11 x86_64
+- Python 3.13
+- PyTorch 2.10.0+cu126
+- Local CUDA toolkit 13.1
+
+The package contains a compiled CUDA extension. Prebuilt wheels are platform- and
+Python-version-specific. If no wheel matches your environment, build from source inside
+a CUDA-enabled PyTorch environment.
+
+### Install
+
+Install a CUDA-enabled PyTorch build first. Example for CUDA 12.6:
+
+```bash
+pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126
+```
+
+Then install PyQITNN without allowing pip to replace your existing Torch build:
+
+```bash
+pip install pyqitnn --no-deps
+```
+
+If you want optional BPE/subword tokenizer support:
+
+```bash
+pip install pyqitnn[tokenizers] --no-deps
+```
+
+### Verify
+
+```python
+import pyqitnn
+
+status = pyqitnn.bridge_status()
+print(pyqitnn.__version__)        # e.g. 0.3.6
+print(status["native_found"])     # True
+print(status["native_loadable"])  # True
+```
+
+---
+
+## Quickstart
+
+### Single QTS layer
+
+```python
+import torch
+import pyqitnn
+
+layer = pyqitnn.QITNNLinear(in_dim=64, out_dim=32, device="cuda:0")
+
+x = torch.randn(4, 64, device="cuda:0")
+out = layer(x)         # shape: [4, 64] -> packed [x | y] simplex state
+print(out.shape)       # torch.Size([4, 64])
+```
+
+### Full transformer
+
+```python
+import torch
+import pyqitnn
+
+model = pyqitnn.QITNNSimplexTransformerLM(
+    vocab_size=256,    # byte-level
+    dim=64,            # logical feature width
+    ffn_dim=128,       # FFN intermediate width
+    seq_len=128,       # max sequence length
+    layers=2,          # transformer blocks
+    precision_mode="fp32",  # explicit trusted baseline
+    device="cuda:0",
+)
+
+tokens = torch.randint(0, 256, (2, 128), device="cuda:0")
+targets = torch.randint(0, 256, (2, 128), device="cuda:0")
+
+logits, loss = model(tokens, targets=targets)
+loss.backward()
+```
+
+### Training loop
+
+```python
+import torch
+
+opt = torch.optim.AdamW(model.parameters(), lr=3e-4)
+
+for step in range(1000):
+    opt.zero_grad(set_to_none=True)
+    _, loss = model(tokens, targets=targets)
+    loss.backward()
+    opt.step()
+
+    # Entropy-floor prior. Call only after optimizer.step().
+    model.apply_qitnn_prior(
+        step_qk=5e-5,
+        step_vo=5e-5,
+        step_ff=5e-5,
+        entropy_floor=1.0840643,
+    )
+```
+
+### Mixed precision toggle
+
+Core `pyqitnn` modules stay on dense `fp32` unless you opt in explicitly.
+
+```python
+model = pyqitnn.QITNNSimplexTransformerLM(
+    vocab_size=256,
+    dim=64,
+    ffn_dim=128,
+    seq_len=128,
+    layers=2,
+    precision_mode="fp32",
+    device="cuda:0",
+)
+```
+
+Set `precision_mode="qts_fp32_rest_bf16"` to enable the conservative mixed path:
+
+- visible activations use CUDA autocast (`bf16`)
+- the native extension accepts `bf16/fp16` activations directly; it no longer relies on Python-side `float32` staging for mixed mode
+- QITNN master weights stay in `fp32`
+- sensitive math stays in `fp32`: Born normalization, backnorm, entropy/prior, and the attention softmax path
+- training script / CLI: use `TrainConfig(precision_mode="qts_fp32_rest_bf16")` or `--precision-mode qts_fp32_rest_bf16`
+- to force the trusted baseline explicitly from CLI, use `TrainConfig(precision_mode="fp32")`, `--precision-mode fp32`, or legacy `--no-mixed-precision`
+- legacy compatibility still exists: `mixed_precision=True` maps to `qts_fp32_rest_bf16`
+
+`BasicQITNN_Transformer.py` currently ships with `TrainConfig.precision_mode="qts_fp32_rest_bf16"` as its standalone trainer default. The lower-level `pyqitnn` modules still default to trusted `fp32` if you omit both `precision_mode` and legacy `mixed_precision`.
+
+### Generation
+
+```python
+import torch
+
+prompt = torch.tensor([[72, 101, 108, 108, 111]], device="cuda:0")  # "Hello"
+output = model.generate(prompt, max_new_tokens=64, temperature=0.7, top_k=12)
+
+text = bytes(output[0].cpu().tolist()).decode("utf-8", errors="replace")
+print(text)
+```
+
+### Tokenizer Modes
+
+PyQITNN's QTS math is tokenizer-agnostic. Switching from byte tokens to BPE/subword
+tokens does **not** change `forward3`, `backnorm3`, `centered_simplex`, `attention2`,
+the 2D simplex state, or the ternary/Born-rule parameterization. It only changes
+how raw text is mapped to token ids and what `vocab_size` the embedding/head use.
+
+- `byte`: built in, fixed `vocab_size=256`, no extra dependency
+- `bpe`: optional, uses HuggingFace `tokenizers`
+
+```python
+import pyqitnn
+
+bpe = pyqitnn.train_bpe_tokenizer(
+    ["hello simplex transformer", "born rule ternary attention"],
+    vocab_size=320,
+    min_frequency=1,
+)
+
+model = pyqitnn.QITNNSimplexTransformerLM(
+    vocab_size=bpe.vocab_size,
+    dim=64,
+    ffn_dim=128,
+    seq_len=128,
+    layers=2,
+    precision_mode="fp32",
+    device="cuda:0",
+)
+```
+
+### Trainer data formats
+
+The training script accepts plain text as well as structured JSON corpora.
+
+- `text`: raw file contents
+- `json`: parse JSON and extract text fields
+- `jsonl` / `ndjson`: parse one JSON record per line
+- `auto`: use file extension to choose between text and JSON parsing
+
+For JSON inputs, the trainer can either collect all string leaves recursively or prefer
+specific fields such as `text,content`.
+
+---
+
+## Architecture Overview
+
+```text
+tokens -> embedding + pos_emb -> [x | y]
+                                    |
+                          +---------+---------+
+                          | QITNNSimplexBlock |  x N layers
+                          |                   |
+                          |  LN -> Q,K,V (QTS)|
+                          |  -> attention2    |
+                          |  -> O (QTS)       |
+                          |  + residual       |
+                          |                   |
+                          |  LN -> ff1 (QTS)  |
+                          |  -> gelu(x) | y   |
+                          |  -> ff2 (QTS)     |
+                          |  + residual       |
+                          +---------+---------+
+                                    |
+                          final LN -> head -> logits
+```
+
+Every QTS projection replaces a standard `nn.Linear` with three amplitude matrices and
+Born-rule normalization. The hidden state is always packed as `[x | y]`, where `x` is the
+polarity channel and `y` is the centered zero-state channel.
+
+---
+
+## Core Math
+
+Each QTS projection computes:
+
+```text
+C_neg  = input @ A_neg
+C_zero = input @ A_zero
+C_pos  = input @ A_pos
+
+Z = C_neg^2 + C_zero^2 + C_pos^2
+
+u = (C_pos^2 - C_neg^2) / Z        # polarity, range [-1, +1]
+v = C_zero^2 / Z                   # zero-state probability, range [0, 1]
+```
+
+Then the centered simplex transform maps `(u, v)` to `(x, y)`:
+
+```text
+x = u
+y = sqrt(3) * v - 1/sqrt(3)
+```
+
+The three pure qutrit states become vertices of an equilateral triangle:
+
+| State  | (u, v)  | (x, y)           |
+|--------|---------|------------------|
+| \|-1\> | (-1, 0) | (-1, -1/sqrt(3)) |
+| \|0\>  | (0, 1)  | (0, 2/sqrt(3))   |
+| \|+1\> | (1, 0)  | (1, -1/sqrt(3))  |
+
+Full math derivations are in the [reference](docs/reference.md).
+
+---
+
+## Stability And Training Notes
+
+There are three separate stabilization mechanisms:
+
+### 1. `ent_lambda`
+
+Adds entropy pressure inside the backward path of `forward3`.
+Use it when you want the optimization itself to discourage collapsed ternary distributions.
+
+### 2. `prior_()` / `apply_qitnn_prior()`
+
+A post-step entropy-floor correction.
+Silent when a triplet is healthy. Only nudges it when entropy drops below the floor.
+
+### 3. Zero-branch learning-rate boost
+
+The zero branch often benefits from a somewhat higher effective learning rate.
+The optimizer helpers and training script expose separate handling for `a_zero`.
+
+A healthy training regime is not "perfectly uniform all the time".
+The goal is to avoid hard collapse while still allowing the model to specialize.
+
+---
+
+## AdamW Configuration
+
+QTS amplitude parameters need `weight_decay=0`. Standard weight decay fights the
+ternary structure and collapses the distribution. Use the trit-floor prior instead.
+
+```python
+# Separate QTS params from standard params.
+qts_ids = set()
+qts_params = []
+for _, _, layer in model.iter_qitnn_layers():
+    for p in (layer.a_neg, layer.a_zero, layer.a_pos):
+        qts_ids.add(id(p))
+        qts_params.append(p)
+
+other_params = [p for p in model.parameters() if id(p) not in qts_ids]
+
+opt = torch.optim.AdamW([
+    {"params": qts_params, "lr": 3e-4, "weight_decay": 0.0},
+    {"params": other_params, "lr": 3e-4, "weight_decay": 0.01},
+])
+```
+
+---
+
+## Known Limitations
+
+**Hardware:**
+- Only `cuda:0` is supported. Multi-GPU requires changes to the CUDA backend.
+- Supported precision modes today are `fp32` and `qts_fp32_rest_bf16`.
+- `precision_mode="fp32"` keeps the original all-`fp32` path.
+- `precision_mode="qts_fp32_rest_bf16"` enables a conservative CUDA `bf16` path for activations while keeping sensitive QITNN math in `fp32`.
+- Legacy `mixed_precision=True` is still accepted as a compatibility alias for `qts_fp32_rest_bf16`.
+- Do not call `.half()` or `.bfloat16()` on the model. Mixed mode expects fp32 master weights.
+
+**Architecture:**
+- Single-head attention only. Multi-head QTS attention is not implemented.
+- No dropout. Regularization comes from the entropy prior.
+- No gradient checkpointing. Memory scales linearly with layers.
+- `seq_len` is fixed at construction time and cannot be changed.
+- BPE/subword tokenization is supported at the Python/trainer layer and does not alter the QTS math path.
+- Byte mode remains the simplest baseline and the default install path.
+
+**Numerical:**
+- cuBLAS GEMM results may differ from `torch.mm` by up to about `1e-2` on large matrices.
+  This is expected FP32 accumulation error and does not affect training.
+- Attention backward error grows with sequence length due to FP32 accumulation.
+  For `seq_len <= 256`, max error is typically below `0.05`.
+- Very small `init_std` (`< 1e-5`) can create dead zones where `Z ~ 0` and gradients
+  vanish. The default `init_std=0.02` avoids this.
+- `prior_()` modifies tensors in-place. Call it only after `optimizer.step()` and
+  outside any autograd context.
+
+**Platform:**
+- Primary development is on Windows. Linux builds are less exercised.
+- macOS is not supported because CUDA is required.
+
+---
+
+## Tests
+
+The repository ships a stress test covering correctness, stability, and convergence:
+
+```bash
+python stress_test.py
+```
+
+This checks Born-rule invariants, finite-difference gradient correctness, attention
+forward/backward vs PyTorch SDPA, prior effectiveness, checkpoint roundtrip,
+determinism, memory stability, and more.
+
+---
+
+## Links
+
+- [Full API Reference](https://github.com/kaifczxc-lab/PyQITNN/blob/SiritoriProjects/docs/reference.md)
+- [Basic QITNN Transformer](https://github.com/kaifczxc-lab/PyQITNN/blob/SiritoriProjects/BasicQITNN_Transformer.py)
+- [GitHub Repository](https://github.com/kaifczxc-lab/PyQITNN)
+- [QITNN Architecture Analysis](https://github.com/kaifczxc-lab/qitnn/blob/SiritoriProjects/Analysis-QITNN.md)
+- [Original Devlog (Discord, GPU Mode)](https://discord.com/channels/1189498204333543425/1466534042768904356/1476227907327098931)
+
+---
+
+## Disclaimer
+
+This is an experimental library implementing a novel neural network architecture.
+The core math, architecture design, debugging, and system integration are the author's
+original work, developed with AI assistance for implementation.
+
+The CUDA kernels are optimized for NVIDIA consumer GPUs; RTX 3060 Ti was the primary
+development target. They work on other architectures but have not been extensively
+benchmarked outside that hardware.
+
+No guarantees of correctness, performance, or suitability for production use.
+Constructive feedback is welcome.