arti-fit 1.0.2__tar.gz

arti_fit-1.0.2/.gitignore

@@ -0,0 +1,14 @@
+__pycache__/
+*.py[cod]
+*.egg-info/
+.pytest_cache/
+.tmp*/
+.venv/
+.uv-cache/
+build/
+dist/
+site/
+.coverage
+htmlcov/
+*.arti.st
+*.lock.json

arti_fit-1.0.2/AI_ASSISTANCE.md

@@ -0,0 +1,12 @@
+# AI Assistance
+
+ARTI's concepts, mechanism definitions, architecture, and final engineering
+decisions are directed by Thiocy.
+
+AI-assisted development tools have been used during implementation, testing,
+documentation, code review, and release auditing. Generated suggestions are
+reviewed and accepted or rejected by the human maintainer. AI systems are not
+project authors, do not own copyright, and are not responsible for releases.
+
+The human maintainer remains responsible for the correctness, security,
+licensing, and published claims of the project.

arti_fit-1.0.2/AUTHORS.md

@@ -0,0 +1,8 @@
+# Authors
+
+ARTI was initiated and designed by Thiocy.
+
+- GitHub: [@Thiocy](https://github.com/Thiocy)
+
+Additional contributors retain attribution through the project history and
+their accepted contributions.

arti_fit-1.0.2/CITATION.cff

@@ -0,0 +1,12 @@
+cff-version: 1.2.0
+message: "If you use ARTI in research or software, please cite this project."
+title: "ARTI: Composable Latent Tensor Layers for PyTorch"
+type: software
+authors:
+  - name: Thiocy
+    website: "https://github.com/Thiocy"
+version: 1.0.2
+date-released: 2026-07-11
+repository-code: "https://github.com/ragnarok-io/ARTI"
+url: "https://github.com/ragnarok-io/ARTI"
+license: MIT

arti_fit-1.0.2/CONTRIBUTING.md

@@ -0,0 +1,12 @@
+# Contributing
+
+ARTI is a tensor-first, domain-independent neural-network library.
+
+```bash
+uv sync --extra dev
+uv run --extra dev pytest
+uv build
+```
+
+Keep changes scoped, document public APIs, and add focused tests for shapes,
+masks, gradients, devices, and serialization where applicable.

arti_fit-1.0.2/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 ARTI contributors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

arti_fit-1.0.2/PKG-INFO

@@ -0,0 +1,251 @@
+Metadata-Version: 2.4
+Name: arti-fit
+Version: 1.0.2
+Summary: AI x RT: dynamic latent tensor representation layers for PyTorch.
+Author: Thiocy
+License: MIT
+License-File: AUTHORS.md
+License-File: LICENSE
+Keywords: latent-tensor,neural-networks,pytorch,representation-learning
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: MIT License
+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
+Requires-Dist: numpy>=1.24
+Requires-Dist: safetensors>=0.4
+Requires-Dist: tomli>=2.0; python_version < '3.11'
+Requires-Dist: torch>=2.2
+Provides-Extra: all
+Requires-Dist: accelerate>=0.33; extra == 'all'
+Requires-Dist: build>=1.2; extra == 'all'
+Requires-Dist: diffusers>=0.30; extra == 'all'
+Requires-Dist: equinox>=0.11.0; extra == 'all'
+Requires-Dist: hatchling>=1.25; extra == 'all'
+Requires-Dist: jax>=0.4.30; extra == 'all'
+Requires-Dist: mkdocs-material>=9.5; extra == 'all'
+Requires-Dist: mkdocs>=1.6; extra == 'all'
+Requires-Dist: mkdocstrings[python]>=0.25; extra == 'all'
+Requires-Dist: numpy>=2.0; extra == 'all'
+Requires-Dist: optax>=0.2.0; extra == 'all'
+Requires-Dist: peft>=0.12; extra == 'all'
+Requires-Dist: pillow>=10.0; extra == 'all'
+Requires-Dist: pytest>=8.0; extra == 'all'
+Requires-Dist: safetensors>=0.4; extra == 'all'
+Requires-Dist: torch>=2.2; extra == 'all'
+Requires-Dist: transformers>=4.51; extra == 'all'
+Provides-Extra: dev
+Requires-Dist: build>=1.2; extra == 'dev'
+Requires-Dist: hatchling>=1.25; extra == 'dev'
+Requires-Dist: numpy>=2.0; extra == 'dev'
+Requires-Dist: pytest>=8.0; extra == 'dev'
+Requires-Dist: torch>=2.2; extra == 'dev'
+Provides-Extra: docs
+Requires-Dist: mkdocs-material>=9.5; extra == 'docs'
+Requires-Dist: mkdocs>=1.6; extra == 'docs'
+Requires-Dist: mkdocstrings[python]>=0.25; extra == 'docs'
+Provides-Extra: font
+Requires-Dist: pillow>=10.0; extra == 'font'
+Provides-Extra: jax
+Requires-Dist: equinox>=0.11.0; extra == 'jax'
+Requires-Dist: jax>=0.4.30; extra == 'jax'
+Requires-Dist: optax>=0.2.0; extra == 'jax'
+Provides-Extra: peft
+Requires-Dist: accelerate>=0.33; extra == 'peft'
+Requires-Dist: peft>=0.12; extra == 'peft'
+Requires-Dist: transformers>=4.51; extra == 'peft'
+Provides-Extra: qwen
+Requires-Dist: accelerate>=0.33; extra == 'qwen'
+Requires-Dist: safetensors>=0.4; extra == 'qwen'
+Requires-Dist: transformers>=4.51; extra == 'qwen'
+Provides-Extra: sd
+Requires-Dist: accelerate>=0.33; extra == 'sd'
+Requires-Dist: diffusers>=0.30; extra == 'sd'
+Requires-Dist: peft>=0.12; extra == 'sd'
+Requires-Dist: pillow>=10.0; extra == 'sd'
+Requires-Dist: safetensors>=0.4; extra == 'sd'
+Requires-Dist: transformers>=4.51; extra == 'sd'
+Provides-Extra: torch
+Requires-Dist: torch>=2.2; extra == 'torch'
+Description-Content-Type: text/markdown
+
+# ARTI
+
+**AI x RT: composable latent tensor layers for PyTorch.**
+
+ARTI is a domain-independent neural-network library for transforming hidden
+tensors at runtime. Its layers work with ordinary tensors and can optionally
+use coordinates, masks, visibility, latent recall, and compact workspaces.
+
+```text
+hidden tensor -> ARTI layer or block -> transformed latent tensor
+```
+
+ARTI does not define a tokenizer, task head, data schema, or business model.
+Applications remain responsible for encoding their context into tensors.
+
+Version 1.0.2 is a **Stable Candidate**. The supported 1.x surface is frozen
+for final compatibility verification, but this release does not yet carry an
+LTS commitment. See [Stability](STABILITY.md) and [Security](SECURITY.md).
+
+## Install
+
+Add ARTI to a project with [uv](https://docs.astral.sh/uv/):
+
+```bash
+uv add arti-fit
+```
+
+ARTI requires Python 3.10 or newer and PyTorch 2.2 or newer. The consuming
+project chooses the appropriate CPU or CUDA build of PyTorch.
+
+The PyPI distribution is named `arti-fit`; the Python import remains `arti`.
+
+Optional integrations can be installed as needed:
+
+```bash
+uv sync --extra jax
+uv sync --extra qwen
+uv sync --extra peft
+uv sync --extra sd
+```
+
+## Use ARTI As A Layer
+
+The smallest API behaves like a normal PyTorch layer:
+
+```python
+import arti
+import torch
+
+layer = arti.nn.Layer(dim=32)
+x = torch.randn(4, 16, 32)
+mask = torch.ones(4, 16, dtype=torch.bool)
+
+out = layer(x, mask=mask)
+
+assert out.y.shape == (4, 16, 32)
+assert out.pooled.shape == (4, 32)
+print(out.diagnostics.keys())
+```
+
+For `[B, D]` inputs, ARTI treats each row as a single token and restores the
+original rank on output.
+
+Capabilities are opt-in. Enable only the structure carried by the data:
+
+```python
+recall_layer = arti.nn.Layer(dim=32, profile="recall")
+multisource = arti.nn.Layer(dim=32, profile="multisource", coord_dim=4)
+```
+
+## Compose Recall, Half, And Fold
+
+ARTI mechanisms are also available as standalone modules:
+
+```python
+import arti
+
+recall = arti.ARTILatentRecallField(hidden_dim=64, slots=8)
+half = arti.nn.Half()
+fold = arti.nn.Fold(k=16, dim=64)
+```
+
+The common Recall branch pattern is deliberately small:
+
+```python
+delta = recall(h, mask, recall=None)[0]
+h = h + half(delta)
+workspace = fold(h, mask=mask)
+```
+
+`Recall` proposes latent traces, `Half` applies feature-strength survival, and
+`Fold` compacts surviving information into a fixed-size workspace. Each module
+can be used independently.
+
+## Attach To An Existing Model
+
+ARTI can discover and attach Recall branches without changing the model class:
+
+```python
+import arti
+
+model = arti.ARTI.attach(
+    model,
+    recall={
+        "layers": "model.layers.*",
+        "rank": 16,
+        "slots": 8,
+    },
+)
+
+print(model.arti.summary())
+model.arti.save("assistant.recall.arti.st")
+```
+
+Attachment configuration supports explicit layer paths, per-layer dimensions,
+independent Recall lines, Half switches, resource previews, and reversible
+removal. Transformers, PEFT, and Diffusers are optional integration boundaries;
+the core package remains PyTorch-first.
+
+## Save And Load Weights
+
+ARTI uses SafeTensors with JSON integrity sidecars:
+
+```python
+saved = arti.save(layer, "layer.arti.st")
+loaded = arti.load("layer.arti.st", model=fresh_layer)
+
+print(saved.weights_sha256)
+print(loaded.missing_keys, loaded.unexpected_keys)
+```
+
+ARTI 1.x reads compatible format-version 1 artifacts produced by the pre-public
+0.x line. Legacy `.pt` migration uses PyTorch's restricted tensor-only loader:
+
+```python
+arti.migrate_pt("legacy-state.pt", "layer.arti.st")
+```
+
+Artifact hashes detect modification relative to their lock files; they are not
+publisher signatures. Obtain models and weights from trusted sources.
+
+## Public Modules
+
+- `arti.nn`: `Layer`, `Half`, `Fold`, `Pulse`, `RecallRefiner`, and visual workspace modules.
+- `arti`: complete ARTI layers, residual blocks, reference models, attachment, serialization, and diagnostics.
+- `arti.torch`: backend-explicit aliases for PyTorch applications.
+- `arti.jax`: optional functional JAX backend with JIT and gradient support.
+- `arti.functional`: mask, visibility, pooling, coordinate-frame, and activation helpers.
+
+Experimental and legacy APIs are identified in their docstrings and are not
+frozen at the same level as the supported core surface.
+
+## Develop
+
+```bash
+git clone https://github.com/ragnarok-io/ARTI.git
+cd ARTI
+uv sync --extra dev
+uv run --extra dev pytest
+uv build
+```
+
+The test suite covers tensor shapes, masks, gradients, serialization, malformed
+artifacts, public API imports, and optional backend boundaries. Contribution
+guidance is in [CONTRIBUTING.md](CONTRIBUTING.md).
+
+## Citation And Authorship
+
+ARTI was initiated and designed by [Thiocy](https://github.com/Thiocy).
+Citation metadata is provided in [CITATION.cff](CITATION.cff). The project also
+documents [authorship](AUTHORS.md) and [AI assistance](AI_ASSISTANCE.md).
+
+## License
+
+[MIT](LICENSE)

arti_fit-1.0.2/README.md

@@ -0,0 +1,175 @@
+# ARTI
+
+**AI x RT: composable latent tensor layers for PyTorch.**
+
+ARTI is a domain-independent neural-network library for transforming hidden
+tensors at runtime. Its layers work with ordinary tensors and can optionally
+use coordinates, masks, visibility, latent recall, and compact workspaces.
+
+```text
+hidden tensor -> ARTI layer or block -> transformed latent tensor
+```
+
+ARTI does not define a tokenizer, task head, data schema, or business model.
+Applications remain responsible for encoding their context into tensors.
+
+Version 1.0.2 is a **Stable Candidate**. The supported 1.x surface is frozen
+for final compatibility verification, but this release does not yet carry an
+LTS commitment. See [Stability](STABILITY.md) and [Security](SECURITY.md).
+
+## Install
+
+Add ARTI to a project with [uv](https://docs.astral.sh/uv/):
+
+```bash
+uv add arti-fit
+```
+
+ARTI requires Python 3.10 or newer and PyTorch 2.2 or newer. The consuming
+project chooses the appropriate CPU or CUDA build of PyTorch.
+
+The PyPI distribution is named `arti-fit`; the Python import remains `arti`.
+
+Optional integrations can be installed as needed:
+
+```bash
+uv sync --extra jax
+uv sync --extra qwen
+uv sync --extra peft
+uv sync --extra sd
+```
+
+## Use ARTI As A Layer
+
+The smallest API behaves like a normal PyTorch layer:
+
+```python
+import arti
+import torch
+
+layer = arti.nn.Layer(dim=32)
+x = torch.randn(4, 16, 32)
+mask = torch.ones(4, 16, dtype=torch.bool)
+
+out = layer(x, mask=mask)
+
+assert out.y.shape == (4, 16, 32)
+assert out.pooled.shape == (4, 32)
+print(out.diagnostics.keys())
+```
+
+For `[B, D]` inputs, ARTI treats each row as a single token and restores the
+original rank on output.
+
+Capabilities are opt-in. Enable only the structure carried by the data:
+
+```python
+recall_layer = arti.nn.Layer(dim=32, profile="recall")
+multisource = arti.nn.Layer(dim=32, profile="multisource", coord_dim=4)
+```
+
+## Compose Recall, Half, And Fold
+
+ARTI mechanisms are also available as standalone modules:
+
+```python
+import arti
+
+recall = arti.ARTILatentRecallField(hidden_dim=64, slots=8)
+half = arti.nn.Half()
+fold = arti.nn.Fold(k=16, dim=64)
+```
+
+The common Recall branch pattern is deliberately small:
+
+```python
+delta = recall(h, mask, recall=None)[0]
+h = h + half(delta)
+workspace = fold(h, mask=mask)
+```
+
+`Recall` proposes latent traces, `Half` applies feature-strength survival, and
+`Fold` compacts surviving information into a fixed-size workspace. Each module
+can be used independently.
+
+## Attach To An Existing Model
+
+ARTI can discover and attach Recall branches without changing the model class:
+
+```python
+import arti
+
+model = arti.ARTI.attach(
+    model,
+    recall={
+        "layers": "model.layers.*",
+        "rank": 16,
+        "slots": 8,
+    },
+)
+
+print(model.arti.summary())
+model.arti.save("assistant.recall.arti.st")
+```
+
+Attachment configuration supports explicit layer paths, per-layer dimensions,
+independent Recall lines, Half switches, resource previews, and reversible
+removal. Transformers, PEFT, and Diffusers are optional integration boundaries;
+the core package remains PyTorch-first.
+
+## Save And Load Weights
+
+ARTI uses SafeTensors with JSON integrity sidecars:
+
+```python
+saved = arti.save(layer, "layer.arti.st")
+loaded = arti.load("layer.arti.st", model=fresh_layer)
+
+print(saved.weights_sha256)
+print(loaded.missing_keys, loaded.unexpected_keys)
+```
+
+ARTI 1.x reads compatible format-version 1 artifacts produced by the pre-public
+0.x line. Legacy `.pt` migration uses PyTorch's restricted tensor-only loader:
+
+```python
+arti.migrate_pt("legacy-state.pt", "layer.arti.st")
+```
+
+Artifact hashes detect modification relative to their lock files; they are not
+publisher signatures. Obtain models and weights from trusted sources.
+
+## Public Modules
+
+- `arti.nn`: `Layer`, `Half`, `Fold`, `Pulse`, `RecallRefiner`, and visual workspace modules.
+- `arti`: complete ARTI layers, residual blocks, reference models, attachment, serialization, and diagnostics.
+- `arti.torch`: backend-explicit aliases for PyTorch applications.
+- `arti.jax`: optional functional JAX backend with JIT and gradient support.
+- `arti.functional`: mask, visibility, pooling, coordinate-frame, and activation helpers.
+
+Experimental and legacy APIs are identified in their docstrings and are not
+frozen at the same level as the supported core surface.
+
+## Develop
+
+```bash
+git clone https://github.com/ragnarok-io/ARTI.git
+cd ARTI
+uv sync --extra dev
+uv run --extra dev pytest
+uv build
+```
+
+The test suite covers tensor shapes, masks, gradients, serialization, malformed
+artifacts, public API imports, and optional backend boundaries. Contribution
+guidance is in [CONTRIBUTING.md](CONTRIBUTING.md).
+
+## Citation And Authorship
+
+ARTI was initiated and designed by [Thiocy](https://github.com/Thiocy).
+Citation metadata is provided in [CITATION.cff](CITATION.cff). The project also
+documents [authorship](AUTHORS.md) and [AI assistance](AI_ASSISTANCE.md).
+
+## License
+
+[MIT](LICENSE)

arti_fit-1.0.2/SECURITY.md

@@ -0,0 +1,32 @@
+# Security Policy
+
+## Supported Version
+
+Security fixes are provided for the current public 1.x release line.
+
+## Trust Boundaries
+
+- `.arti.st` uses SafeTensors and validates its sidecar hashes. Hashes detect
+  accidental or unauthorised modification relative to the supplied lock, but
+  they are not a publisher signature. Obtain artifacts from a trusted source.
+- Legacy `.pt` migration and ARTI artifact loaders use PyTorch's restricted
+  `weights_only=True` loader. Full pickled Python objects are not supported.
+- Declarative pretrained loading rejects `trust_remote_code=True`. Review and
+  instantiate models requiring remote code yourself before passing the model
+  object to ARTI.
+- CLI model references in `module:attribute` form intentionally import and run
+  local Python factories. Treat such references and their Python environment as
+  executable code, not as passive configuration.
+- Downloaded models and optional dependencies remain external supply-chain
+  inputs. Pin revisions and dependencies according to your deployment policy.
+- ARTI rejects oversized CLI sample tensors and malformed artifact metadata,
+  but callers remain responsible for bounding arbitrary model inputs, sequence
+  lengths, batch sizes, and user-selected model dimensions.
+- NaN and infinity values in activation configuration are rejected. NaN or
+  infinity in model data follows normal PyTorch propagation semantics.
+
+## Reporting
+
+Report suspected vulnerabilities privately to the repository maintainers.
+Do not include secrets, private model weights, or sensitive user data in a
+public issue.

arti_fit-1.0.2/STABILITY.md

@@ -0,0 +1,67 @@
+# Stability Policy
+
+ARTI 1.0.2 is published as a Stable Candidate. This label is a release stage,
+not a separate package version and not an LTS promise.
+
+## Supported 1.x Surface
+
+The following contracts are frozen across compatible 1.x releases:
+
+- Core tensor contracts for `[B, D]` and `[B, N, D]` inputs.
+- `arti.nn` core layers: `Layer`, `Half`, `Fold`, `Pulse`, and
+  `RecallRefiner`.
+- Root core layers: `ARTILayer`, `ARTIResidualBlock`, `ARTISequenceBlock`,
+  `ARTIPooledBlock`, and `ARTIClassifier`.
+- `ARTI.attach`, attachment configuration, save/load, and reversible removal.
+- The `arti.st` format-version 1 reader and writer.
+- Mask, coordinate, visibility, output, and diagnostics tensor contracts.
+
+Modules and parameters explicitly documented as alpha, experimental, or
+legacy are not frozen at the same level. They will still follow the deprecation
+process below.
+
+## Compatibility
+
+- Patch releases fix defects without intentionally breaking supported APIs.
+- Minor 1.x releases may add optional parameters and APIs with compatible
+  defaults.
+- Breaking supported APIs requires a new major release.
+- ARTI 1.x reads valid format-version 1 `arti.st` artifacts produced by the
+  pre-public 0.x line. Artifact format compatibility is independent of the
+  package version.
+- Serialized artifacts must not depend on Python pickle for normal `.arti.st`
+  loading.
+
+## Deprecation
+
+A supported API is deprecated before removal. Deprecation must include a
+runtime warning, a documented replacement, and coverage in the release notes.
+Removal occurs no earlier than the next major release. Security or correctness
+issues may require an exception, which must be documented.
+
+## Support Matrix
+
+| Component | Stable Candidate support |
+| --- | --- |
+| Python | 3.10, 3.11, 3.12 |
+| PyTorch | 2.2 or newer |
+| CPU | Supported |
+| NVIDIA CUDA | Supported through CUDA-enabled PyTorch |
+| JAX | Optional functional backend; smaller surface than PyTorch |
+| Transformers, PEFT, Diffusers | Optional integrations |
+| Artifact format | `arti.st` format version 1 |
+
+## Promotion To Stable
+
+The Stable label requires the public CI matrix to pass from a clean checkout,
+wheel and sdist installation checks to pass, the supported API inventory to be
+reviewed, and at least one release-candidate feedback cycle to complete without
+an unresolved compatibility defect.
+
+## Future LTS
+
+An LTS release will be declared only with a published maintenance window of at
+least 12 months, security and critical-defect support, a documented Python and
+PyTorch support window, and continued read compatibility for supported
+`arti.st` artifacts. Until that declaration, no ARTI release carries an LTS
+commitment.