mflux-teacache 0.1.0__tar.gz

mflux_teacache-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 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 type. We recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Denis Ineshin
+
+   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.

mflux_teacache-0.1.0/NOTICE

@@ -0,0 +1,41 @@
+mflux-teacache
+Copyright 2026 Denis Ineshin and the mflux-teacache contributors
+
+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
+
+------------------------------------------------------------------------------
+Third-party attributions
+------------------------------------------------------------------------------
+
+This product includes software derived from mflux
+(https://github.com/mflux-community/mflux), licensed under MIT License:
+
+  MIT License
+
+  Copyright (c) 2026 Filip Strand
+
+  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.
+
+  src/mflux/extras/teacache/z_image.py (warn_ineffective_options) copies a
+  block of statements from main() in mflux's
+  src/mflux/models/z_image/cli/z_image_generate.py.

mflux_teacache-0.1.0/PKG-INFO

@@ -0,0 +1,158 @@
+Metadata-Version: 2.4
+Name: mflux-teacache
+Version: 0.1.0
+Summary: TeaCache step skipping for mflux's command line, powered by mlx-teacache.
+Keywords: mflux,mlx,teacache,apple-silicon,diffusion,z-image
+Author: Denis Ineshin
+Author-email: Denis Ineshin <denis.ineshin@gmail.com>
+License-Expression: Apache-2.0
+License-File: LICENSE
+License-File: NOTICE
+Classifier: Intended Audience :: Developers
+Classifier: Operating System :: MacOS
+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
+Requires-Dist: mlx-teacache[mflux]>=0.13.2
+Requires-Dist: mflux>=0.22
+Maintainer: Denis Ineshin
+Maintainer-email: Denis Ineshin <denis.ineshin@gmail.com>
+Requires-Python: >=3.10
+Project-URL: Homepage, https://github.com/mflux-community/mflux-plugins/tree/main/teacache
+Project-URL: Repository, https://github.com/mflux-community/mflux-plugins
+Project-URL: Issues, https://github.com/mflux-community/mflux-plugins/issues
+Project-URL: Documentation, https://github.com/mflux-community/mflux-plugins/tree/main/teacache#readme
+Description-Content-Type: text/markdown
+
+# mflux-teacache
+
+Adds the command `mflux-generate-z-image-teacache` to [mflux](https://github.com/mflux-community/mflux). It takes
+every option of `mflux-generate-z-image` and runs mflux's own loading and generation steps, with TeaCache step
+skipping from [mlx-teacache](https://github.com/IonDen/mlx-teacache) turned on in between.
+
+TeaCache skips transformer steps when the result would barely change, and reuses an earlier result. The image is
+close to the plain command's output but not identical. For a supported model, run `mflux-generate-z-image-teacache`
+wherever you would run `mflux-generate-z-image`; the other options stay the same.
+
+mflux-teacache lives in the `teacache/` folder of
+[mflux-plugins](https://github.com/mflux-community/mflux-plugins), the plugin repository of the mflux-community
+organization. Denis Ineshin maintains it. The step skipping itself is in
+[mlx-teacache](https://github.com/IonDen/mlx-teacache), a separate Python library by the same author; this package
+connects it to mflux's command line.
+[Versions](https://github.com/mflux-community/mflux-plugins/tree/main/teacache#versions) lists the mflux releases it
+installs with.
+
+## Install
+
+The recommended setup puts mflux and this package in one tool environment:
+
+```bash
+uv tool install mflux --with-executables-from mflux-teacache
+```
+
+You get mflux's commands and `mflux-generate-z-image-teacache` side by side, and `mflux-capabilities` lists all of
+them. In a virtual environment that already has mflux:
+
+```bash
+pip install mflux-teacache
+```
+
+`uv tool install mflux-teacache` on its own works too, but it builds a second mflux environment and puts only this
+package's command on your path.
+
+## Usage
+
+```bash
+mflux-generate-z-image-teacache --prompt "a lighthouse on a cliff at dusk" --steps 50 --seed 42 -q 8
+```
+
+Every `mflux-generate-z-image` option works the same way. After each image the command prints a line such as
+`TeaCache: skipped <k> of <n> steps (threshold 0.12)`.
+
+`--teacache-threshold VALUE` replaces the model's calibrated threshold. It must be greater than 0 and at most 1.
+A higher value skips more steps and moves the image further from a plain mflux run.
+
+mflux also has a `--step-cache-ratio` flag (alias `--teacache-ratio`), on the Qwen-Image 2.1 command. It is a
+different feature: a fixed share of steps reused by mflux's own step cache. That step cache is not an option on the
+Z-Image command.
+
+`--compute-precision` is accepted with a warning: in mflux 0.22 the only value is `float16`, and TeaCache's skip
+decisions and image quality have not been measured with it. It changes only mflux's attention and feed-forward math;
+TeaCache's cached results keep the model's own precision.
+
+The command checks your request before it loads any weights. When TeaCache can't run it, the command stops with
+exit code 2 and a message that points to the plain mflux command. That happens for:
+
+- a model TeaCache doesn't support on this command, such as `--model z-image-turbo` (the message lists the
+  supported models);
+- a distilled few-step model, where there is nothing to skip;
+- a threshold outside 0 to 1;
+- mflux's step cache. This command has no step-cache option; a later command for a model that has one refuses it,
+  also when `-C` restores it;
+- too few steps. The first and last steps always run. 2 or fewer active steps are refused. With 3, the run is accepted
+  but skips nothing, and TeaCache prints a warning saying so. An image-to-image run counts only the steps it actually
+  runs, so a high strength in `--image PATH STRENGTH` can leave too few (strength 1.0 leaves none).
+
+A checkpoint other than the one TeaCache was calibrated on (a finetune, a third-party repo) runs with a warning,
+because the skips may not suit it.
+
+If something fails after the model has loaded, the command prints the error's traceback and exits with code 1.
+
+## Supported models
+
+| Command | `--model` | Default threshold |
+|---|---|---|
+| `mflux-generate-z-image-teacache` | `z-image` (the default) | 0.12 |
+
+Z-Image Turbo is not supported: the command refuses it as an unsupported model. Other model families get a
+command once their mflux command has the same load-and-generate steps.
+
+## What it changes
+
+The command patches the loaded model in memory. For Z-Image it replaces the model's prediction step with
+mlx-teacache's gated version. Nothing on disk changes, and mflux's own commands behave exactly as before. mflux's
+shell completions list only mflux's own commands.
+
+Each image records what TeaCache did, next to mflux's usual metadata: `teacache_threshold`,
+`teacache_skipped_steps`, `teacache_active_steps`, `teacache_variant` and `teacache_version` (the mlx-teacache
+version). Replaying a TeaCache image with plain mflux and `-C` produces the plain run, without TeaCache. Replaying
+it with this command uses the model's default threshold unless you pass `--teacache-threshold` again, because the
+`teacache_*` keys are not read back.
+
+For measured speedups and image comparisons, see the
+[mlx-teacache README](https://github.com/IonDen/mlx-teacache#readme).
+
+## Versions
+
+- Python 3.10 or newer, on Apple Silicon.
+- mflux 0.22 or newer. The upper limit comes from mlx-teacache: its `[mflux]` extra allows mflux below 0.25. A newer
+  mflux needs a newer mlx-teacache first; until then the installer keeps or picks an older mflux. mflux 0.21 and
+  earlier set the Z-Image scheduler default outside the parser this command reuses, so the command would not get it.
+- mlx-teacache 0.13.2 or newer, the first release that lists mflux 0.22 as checked. This package adds no upper limit
+  of its own. On an mflux newer than the newest one mlx-teacache was checked against, the command runs and prints a
+  warning.
+
+mflux's other options, `--float32` included, pass through unchanged.
+
+What changed in each release is in
+[CHANGELOG.md](https://github.com/mflux-community/mflux-plugins/blob/main/teacache/CHANGELOG.md).
+
+## Development
+
+From the `teacache/` folder of an mflux-plugins checkout:
+
+```bash
+uv sync --locked
+uv run pytest
+uv run ruff check . && uv run ruff format --check .
+uv run ty check src
+```
+
+The tests use stand-ins for the model and never download weights. `AGENTS.md` describes the package layout and its
+rules.
+
+## License
+
+Apache-2.0. `NOTICE` lists the code this package copies from mflux, which is MIT-licensed.

mflux_teacache-0.1.0/README.md

@@ -0,0 +1,130 @@
+# mflux-teacache
+
+Adds the command `mflux-generate-z-image-teacache` to [mflux](https://github.com/mflux-community/mflux). It takes
+every option of `mflux-generate-z-image` and runs mflux's own loading and generation steps, with TeaCache step
+skipping from [mlx-teacache](https://github.com/IonDen/mlx-teacache) turned on in between.
+
+TeaCache skips transformer steps when the result would barely change, and reuses an earlier result. The image is
+close to the plain command's output but not identical. For a supported model, run `mflux-generate-z-image-teacache`
+wherever you would run `mflux-generate-z-image`; the other options stay the same.
+
+mflux-teacache lives in the `teacache/` folder of
+[mflux-plugins](https://github.com/mflux-community/mflux-plugins), the plugin repository of the mflux-community
+organization. Denis Ineshin maintains it. The step skipping itself is in
+[mlx-teacache](https://github.com/IonDen/mlx-teacache), a separate Python library by the same author; this package
+connects it to mflux's command line.
+[Versions](https://github.com/mflux-community/mflux-plugins/tree/main/teacache#versions) lists the mflux releases it
+installs with.
+
+## Install
+
+The recommended setup puts mflux and this package in one tool environment:
+
+```bash
+uv tool install mflux --with-executables-from mflux-teacache
+```
+
+You get mflux's commands and `mflux-generate-z-image-teacache` side by side, and `mflux-capabilities` lists all of
+them. In a virtual environment that already has mflux:
+
+```bash
+pip install mflux-teacache
+```
+
+`uv tool install mflux-teacache` on its own works too, but it builds a second mflux environment and puts only this
+package's command on your path.
+
+## Usage
+
+```bash
+mflux-generate-z-image-teacache --prompt "a lighthouse on a cliff at dusk" --steps 50 --seed 42 -q 8
+```
+
+Every `mflux-generate-z-image` option works the same way. After each image the command prints a line such as
+`TeaCache: skipped <k> of <n> steps (threshold 0.12)`.
+
+`--teacache-threshold VALUE` replaces the model's calibrated threshold. It must be greater than 0 and at most 1.
+A higher value skips more steps and moves the image further from a plain mflux run.
+
+mflux also has a `--step-cache-ratio` flag (alias `--teacache-ratio`), on the Qwen-Image 2.1 command. It is a
+different feature: a fixed share of steps reused by mflux's own step cache. That step cache is not an option on the
+Z-Image command.
+
+`--compute-precision` is accepted with a warning: in mflux 0.22 the only value is `float16`, and TeaCache's skip
+decisions and image quality have not been measured with it. It changes only mflux's attention and feed-forward math;
+TeaCache's cached results keep the model's own precision.
+
+The command checks your request before it loads any weights. When TeaCache can't run it, the command stops with
+exit code 2 and a message that points to the plain mflux command. That happens for:
+
+- a model TeaCache doesn't support on this command, such as `--model z-image-turbo` (the message lists the
+  supported models);
+- a distilled few-step model, where there is nothing to skip;
+- a threshold outside 0 to 1;
+- mflux's step cache. This command has no step-cache option; a later command for a model that has one refuses it,
+  also when `-C` restores it;
+- too few steps. The first and last steps always run. 2 or fewer active steps are refused. With 3, the run is accepted
+  but skips nothing, and TeaCache prints a warning saying so. An image-to-image run counts only the steps it actually
+  runs, so a high strength in `--image PATH STRENGTH` can leave too few (strength 1.0 leaves none).
+
+A checkpoint other than the one TeaCache was calibrated on (a finetune, a third-party repo) runs with a warning,
+because the skips may not suit it.
+
+If something fails after the model has loaded, the command prints the error's traceback and exits with code 1.
+
+## Supported models
+
+| Command | `--model` | Default threshold |
+|---|---|---|
+| `mflux-generate-z-image-teacache` | `z-image` (the default) | 0.12 |
+
+Z-Image Turbo is not supported: the command refuses it as an unsupported model. Other model families get a
+command once their mflux command has the same load-and-generate steps.
+
+## What it changes
+
+The command patches the loaded model in memory. For Z-Image it replaces the model's prediction step with
+mlx-teacache's gated version. Nothing on disk changes, and mflux's own commands behave exactly as before. mflux's
+shell completions list only mflux's own commands.
+
+Each image records what TeaCache did, next to mflux's usual metadata: `teacache_threshold`,
+`teacache_skipped_steps`, `teacache_active_steps`, `teacache_variant` and `teacache_version` (the mlx-teacache
+version). Replaying a TeaCache image with plain mflux and `-C` produces the plain run, without TeaCache. Replaying
+it with this command uses the model's default threshold unless you pass `--teacache-threshold` again, because the
+`teacache_*` keys are not read back.
+
+For measured speedups and image comparisons, see the
+[mlx-teacache README](https://github.com/IonDen/mlx-teacache#readme).
+
+## Versions
+
+- Python 3.10 or newer, on Apple Silicon.
+- mflux 0.22 or newer. The upper limit comes from mlx-teacache: its `[mflux]` extra allows mflux below 0.25. A newer
+  mflux needs a newer mlx-teacache first; until then the installer keeps or picks an older mflux. mflux 0.21 and
+  earlier set the Z-Image scheduler default outside the parser this command reuses, so the command would not get it.
+- mlx-teacache 0.13.2 or newer, the first release that lists mflux 0.22 as checked. This package adds no upper limit
+  of its own. On an mflux newer than the newest one mlx-teacache was checked against, the command runs and prints a
+  warning.
+
+mflux's other options, `--float32` included, pass through unchanged.
+
+What changed in each release is in
+[CHANGELOG.md](https://github.com/mflux-community/mflux-plugins/blob/main/teacache/CHANGELOG.md).
+
+## Development
+
+From the `teacache/` folder of an mflux-plugins checkout:
+
+```bash
+uv sync --locked
+uv run pytest
+uv run ruff check . && uv run ruff format --check .
+uv run ty check src
+```
+
+The tests use stand-ins for the model and never download weights. `AGENTS.md` describes the package layout and its
+rules.
+
+## License
+
+Apache-2.0. `NOTICE` lists the code this package copies from mflux, which is MIT-licensed.

mflux_teacache-0.1.0/pyproject.toml

@@ -0,0 +1,101 @@
+[build-system]
+# The backend this package is built and tested with: uv 0.11.2 (CI pins it) builds with its own uv_build 0.11.2.
+requires = ["uv_build>=0.11.2,<0.12"]
+build-backend = "uv_build"
+
+[tool.uv.build-backend]
+default-excludes = true
+# mflux PR #776 namespace contract: this distribution ships only mflux/extras/teacache/.
+# Never ship src/mflux/__init__.py (mflux core owns it) or src/mflux/extras/__init__.py
+# (mflux.extras must stay an implicit namespace package that every extras package shares).
+# The dotted module name already marks a namespace package, so no `namespace = true`.
+module-name = "mflux.extras.teacache"
+
+[project]
+name = "mflux-teacache"
+version = "0.1.0"
+description = "TeaCache step skipping for mflux's command line, powered by mlx-teacache."
+readme = "README.md"
+license = "Apache-2.0"
+license-files = ["LICENSE", "NOTICE"]
+authors = [{ name = "Denis Ineshin", email = "denis.ineshin@gmail.com" }]
+maintainers = [{ name = "Denis Ineshin", email = "denis.ineshin@gmail.com" }]
+requires-python = ">=3.10"
+keywords = ["mflux", "mlx", "teacache", "apple-silicon", "diffusion", "z-image"]
+classifiers = [
+    "Intended Audience :: Developers",
+    "Operating System :: MacOS",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Programming Language :: Python :: 3.13",
+]
+dependencies = [
+    # The supported mflux range comes from mlx-teacache's [mflux] extra, which lists the
+    # mflux versions mlx-teacache has checked. This package adds no mflux upper bound.
+    # 0.13.2 is the first release that records mflux 0.22 as checked (it follows 0.22's bfloat16 Z-Image
+    # stream and does not warn on it).
+    "mlx-teacache[mflux]>=0.13.2",
+    # A floor only: 0.22 is the first mflux whose Z-Image parser sets the scheduler default.
+    "mflux>=0.22",
+]
+
+[project.urls]
+Homepage = "https://github.com/mflux-community/mflux-plugins/tree/main/teacache"
+Repository = "https://github.com/mflux-community/mflux-plugins"
+Issues = "https://github.com/mflux-community/mflux-plugins/issues"
+Documentation = "https://github.com/mflux-community/mflux-plugins/tree/main/teacache#readme"
+
+[project.scripts]
+mflux-generate-z-image-teacache = "mflux.extras.teacache.z_image:main"
+
+[dependency-groups]
+dev = [
+    "pytest>=9.1,<10.0",
+    # Pinned exactly, matching mflux core.
+    "ruff==0.16.3",
+    "ty==0.0.72",
+]
+
+[tool.ruff]
+line-length = 120
+indent-width = 4
+target-version = "py310"
+respect-gitignore = true
+
+[tool.ruff.lint]
+select = ["BLE", "E4", "E7", "E9", "F", "I", "ICN", "LOG", "PERF", "W"]
+
+[tool.ruff.lint.per-file-ignores]
+# conftest.py sets HF_HUB_OFFLINE / HF_HOME before importing anything that reads them.
+"tests/conftest.py" = ["E402"]
+
+[tool.ruff.lint.isort]
+case-sensitive = false
+combine-as-imports = true
+known-first-party = ["mflux.extras.teacache"]
+
+[tool.ruff.format]
+quote-style = "double"
+indent-style = "space"
+
+[tool.ty.environment]
+python-version = "3.10"
+
+[tool.ty.analysis]
+# ty cannot model mflux core's pkgutil.extend_path merge: core is a regular package in
+# site-packages and this distribution adds mflux/extras/teacache/ from src/, so a static
+# resolver reports absolute mflux.extras.* imports as unresolved. The package uses relative
+# imports internally; tests/test_namespace.py checks the real import at runtime.
+allowed-unresolved-imports = ["mflux.extras.**"]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+addopts = "-ra --strict-markers"
+filterwarnings = [
+    "error",
+    "ignore:builtin type SwigPyPacked has no __module__ attribute:DeprecationWarning",
+    "ignore:builtin type SwigPyObject has no __module__ attribute:DeprecationWarning",
+    "ignore:builtin type swigvarlink has no __module__ attribute:DeprecationWarning",
+]

mflux_teacache-0.1.0/src/mflux/extras/teacache/__init__.py

@@ -0,0 +1,8 @@
+"""TeaCache step skipping for mflux's command line, powered by mlx-teacache."""
+
+from importlib.metadata import PackageNotFoundError, version
+
+try:
+    __version__ = version("mflux-teacache")
+except PackageNotFoundError:  # imported from a source tree that was never installed
+    __version__ = "0+unknown"

mflux_teacache-0.1.0/src/mflux/extras/teacache/_core.py

@@ -0,0 +1,261 @@
+"""The command flow every mflux-teacache command shares.
+
+Each model family has a small adapter module (z_image.py) that names mflux's own command steps.
+run() parses with mflux's parser, refuses anything TeaCache can't run before any weights load, then
+loads with mflux, applies TeaCache, and generates, records and saves the way the plain mflux command
+does."""
+
+import sys
+import traceback
+import warnings
+from argparse import Namespace
+from collections.abc import Callable
+from dataclasses import dataclass
+from typing import Any
+
+from mflux.callbacks.callback_manager import CallbackManager
+from mflux.cli.parser.parsers import CommandLineParser
+from mflux.models.common.config.config import Config
+from mflux.models.common.config.model_config import AVAILABLE_MODELS, ModelConfig
+from mflux.utils.exceptions import ModelConfigError, PromptFileReadError, StopImageGenerationException
+from mflux.utils.generated_image import GeneratedImage
+from mflux.utils.prompt_util import PromptUtil
+from mlx_teacache import (
+    InvalidStepWindowError,
+    TeaCacheHandle,
+    TeaCacheUncalibratedCheckpointWarning,
+    TeaCacheValueError,
+    VariantInfo,
+    __version__ as mlx_teacache_version,
+    apply_teacache,
+    check_step_window,
+    match_variant,
+)
+
+THRESHOLD_FLAG = "--teacache-threshold"
+# mflux 0.22's --compute-precision (float16 today): any non-default value is allowed for now, with this one
+# warning. Kept in one place so a later switch to a refusal is a small change.
+COMPUTE_PRECISION_WARNING = (
+    "TeaCache's skip decisions and image quality have not been measured with --compute-precision {value}; "
+    "the run uses it as asked."
+)
+
+
+@dataclass(frozen=True)
+class Adapter:
+    """One mflux command as run() needs it.
+
+    The required fields are mflux's own command steps. The hooks are optional, so an adapter names
+    only what its mirrored command does around them:
+    - before_validate(args): right after parsing and the threshold check, before validate (no current
+      adapter uses it).
+    - after_checks(args, model_config): after TeaCache's refusals, before load (Z-Image: mflux's
+      guidance and negative-prompt warnings), so a refused run prints one message.
+    - active_steps(args, model_config): the denoising steps mflux will run; None uses
+      default_active_steps (mflux's own Config.init_time_step).
+    - library_checks_checkpoint: True when mlx-teacache's VariantInfo.calibrated already tells a
+      custom checkpoint apart (Qwen-Image); it switches off this package's model-path warning so
+      the user sees one warning, not two."""
+
+    command: str
+    plain_command: str
+    pipeline: type
+    family_keys: tuple[str, ...]
+    build_parser: Callable[[], CommandLineParser]
+    validate: Callable[[Namespace], ModelConfig]
+    load: Callable[[Namespace], Any]
+    generate: Callable[[Any, Namespace, int, str], GeneratedImage]
+    latent_creator: type
+    before_validate: Callable[[Namespace], None] | None = None
+    after_checks: Callable[[Namespace, ModelConfig], None] | None = None
+    active_steps: Callable[[Namespace, ModelConfig], int] | None = None
+    library_checks_checkpoint: bool = False
+
+
+def add_teacache_arguments(parser: CommandLineParser) -> None:
+    group = parser.add_argument_group("TeaCache")
+    group.add_argument(
+        THRESHOLD_FLAG,
+        dest="teacache_threshold",
+        type=float,
+        default=None,
+        metavar="VALUE",
+        help=(
+            "How much change TeaCache tolerates before it computes a step again: greater than 0, at most 1. "
+            "Higher skips more steps and changes the image more. Default: the model's calibrated value."
+        ),
+    )
+
+
+def default_active_steps(args: Namespace, model_config: ModelConfig) -> int:
+    """The denoising steps mflux will run: --steps minus mflux's own init_time_step (an image-to-image
+    run starts later; strength 1.0 leaves none). Loads no weights and opens no image: width and height
+    keep Config's defaults."""
+    init_time_step = Config(
+        model_config=model_config,
+        num_inference_steps=args.steps,
+        image_path=getattr(args, "image_path", None),
+        image_strength=getattr(args, "image_strength", None),
+    ).init_time_step
+    return args.steps - init_time_step
+
+
+def supported_model_names(adapter: Adapter) -> tuple[str, ...]:
+    """The --model names this command can run under TeaCache, from mlx-teacache's own registry."""
+    names: list[str] = []
+    for key in adapter.family_keys:
+        variant = match_variant(AVAILABLE_MODELS[key], adapter.pipeline)
+        if variant is not None:
+            names.extend(name for name in variant.model_names if name not in names)
+    return tuple(names)
+
+
+def teacache_metadata(handle: TeaCacheHandle, committed_before: int) -> dict[str, float | int | str] | None:
+    """The image's TeaCache record, or None when this generation was not committed (interrupted, or
+    the handle was restored mid-run). Flat keys and JSON primitives, like mflux's own step-cache key,
+    because mflux writes them into the PNG and the JSON sidecar as they are."""
+    stats = handle.stats
+    last = stats.last_generation
+    if stats.generations <= committed_before or last is None:
+        return None
+    return {
+        "teacache_threshold": float(handle.rel_l1_thresh),
+        "teacache_skipped_steps": sum(1 for step in last.decisions if step.decision == "skipped"),
+        "teacache_active_steps": int(last.num_steps),
+        "teacache_variant": str(getattr(handle, "variant_id", "unknown")),
+        "teacache_version": str(mlx_teacache_version),
+    }
+
+
+def _record_teacache(image: GeneratedImage, handle: TeaCacheHandle, committed_before: int) -> None:
+    recorded = teacache_metadata(handle, committed_before)
+    if recorded is None:
+        return
+    # Merged into what mflux recorded there (Z-Image: float32, compute_precision, which -C replays), and a new
+    # dict: never mutate one the image may share (GeneratedImage.get_right_half shares it). A later family's image may
+    # carry None there.
+    image.generation_parameters = {**(image.generation_parameters or {}), **recorded}
+    print(
+        f"TeaCache: skipped {recorded['teacache_skipped_steps']} of {recorded['teacache_active_steps']} steps "
+        f"(threshold {recorded['teacache_threshold']:g})"
+    )
+
+
+def _checked_threshold(parser: CommandLineParser, adapter: Adapter, value: float | None) -> float | None:
+    if value is not None and not 0.0 < value <= 1.0:  # NaN fails both comparisons and is refused
+        parser.error(
+            f"{THRESHOLD_FLAG} must be greater than 0 and at most 1, got {value}. "
+            f"Run {adapter.plain_command} to generate without TeaCache."
+        )
+    return value
+
+
+def _refuse_what_teacache_cannot_run(
+    parser: CommandLineParser, adapter: Adapter, args: Namespace, model_config: ModelConfig
+) -> VariantInfo:
+    variant = match_variant(model_config, adapter.pipeline)
+    if variant is None:
+        shown = args.model or model_config.model_name
+        names = ", ".join(supported_model_names(adapter)) or "none"
+        parser.error(
+            f"TeaCache doesn't support {shown} yet. Supported on this command: {names}. "
+            f"Run {adapter.plain_command} to generate without TeaCache."
+        )
+    if variant.default_thresh is None:
+        parser.error(
+            f"{variant.display_name} is a distilled few-step model: TeaCache has no steps to skip there. "
+            f"Run {adapter.plain_command} instead."
+        )
+    if getattr(args, "step_cache_ratio", None) is not None:
+        parser.error(
+            "--step-cache-ratio (given on the command line or restored by --config-from-conf) also skips steps, "
+            "and mflux's step cache and TeaCache can't run together. "
+            "Remove --step-cache-ratio (or --teacache-ratio) from the command line, or the step_cache_ratio key "
+            f"from the -C file; or run {adapter.plain_command} to use mflux's step cache."
+        )
+    active_steps = (adapter.active_steps or default_active_steps)(args, model_config)
+    try:
+        check_step_window(active_num_steps=active_steps, nominal_num_inference_steps=args.steps)
+    # TeaCacheValueError: a negative count (--steps -1, or --steps 0 with --image); mflux's --steps has no range check.
+    except (InvalidStepWindowError, TeaCacheValueError) as exc:
+        # The library's texts end with or without a period; print exactly one before the hint.
+        parser.error(
+            f"{str(exc).rstrip('.')}. Use more --steps (or a lower --image strength), "
+            f"or run {adapter.plain_command} without TeaCache."
+        )
+    return variant
+
+
+def _warn_unverified_checkpoint(
+    adapter: Adapter, args: Namespace, model_config: ModelConfig, variant: VariantInfo
+) -> None:
+    """One warning, never a refusal, when the coefficients were not fitted on the checkpoint this run loads."""
+    model_path = getattr(args, "model_path", None)
+    if not variant.calibrated:
+        shown = model_path or args.model or model_config.model_name
+        warnings.warn(
+            TeaCacheUncalibratedCheckpointWarning(
+                f"{shown} is not the checkpoint {variant.display_name}'s TeaCache coefficients were calibrated "
+                "on; skip counts and image quality on it are unchecked. TeaCache still runs."
+            ),
+            stacklevel=3,
+        )
+        return
+    if adapter.library_checks_checkpoint or model_path is None or model_path == model_config.model_name:
+        return
+    warnings.warn(
+        TeaCacheUncalibratedCheckpointWarning(
+            f"{model_path} is not {model_config.model_name}, the checkpoint TeaCache's coefficients were "
+            "calibrated on; skip counts and image quality on it are unchecked."
+        ),
+        stacklevel=3,
+    )
+
+
+def _warn_unmeasured_compute_precision(args: Namespace) -> None:
+    """Reads the parsed value, so a -C sidecar that recorded a compute precision warns too."""
+    value = getattr(args, "compute_precision", None)
+    if value is not None:
+        warnings.warn(COMPUTE_PRECISION_WARNING.format(value=value), UserWarning, stacklevel=3)
+
+
+def run(adapter: Adapter) -> None:
+    parser = adapter.build_parser()
+    args = parser.parse_args()
+    threshold = _checked_threshold(parser, adapter, args.teacache_threshold)
+    if adapter.before_validate is not None:
+        adapter.before_validate(args)
+    try:
+        model_config = adapter.validate(args)
+    except ModelConfigError as exc:
+        parser.error(str(exc))
+    # TeaCache's refusals come before mflux's option warnings: a refused run prints one message.
+    variant = _refuse_what_teacache_cannot_run(parser, adapter, args, model_config)
+    if adapter.after_checks is not None:
+        adapter.after_checks(args, model_config)
+    _warn_unverified_checkpoint(adapter, args, model_config, variant)
+    _warn_unmeasured_compute_precision(args)
+
+    model = adapter.load(args)
+    memory_saver = None
+    try:
+        handle = apply_teacache(model, rel_l1_thresh=threshold)
+        memory_saver = CallbackManager.register_callbacks(args=args, model=model, latent_creator=adapter.latent_creator)
+        for seed in args.seed:
+            committed_before = handle.stats.generations
+            image = adapter.generate(model, args, seed, PromptUtil.read_prompt(args))
+            _record_teacache(image, handle, committed_before)
+            image.save(path=args.output.format(seed=seed), export_json_metadata=args.metadata)
+    except (StopImageGenerationException, PromptFileReadError) as exc:
+        print(exc)
+    except Exception:  # noqa: BLE001 -- any failure after load: keep the traceback for a report, exit 1
+        traceback.print_exc()
+        print(
+            f"{adapter.command}: error: the run failed (traceback above). "
+            f"Run {adapter.plain_command} to generate without TeaCache.",
+            file=sys.stderr,
+        )
+        raise SystemExit(1) from None
+    finally:
+        if memory_saver:
+            print(memory_saver.memory_stats())

mflux_teacache-0.1.0/src/mflux/extras/teacache/z_image.py

@@ -0,0 +1,78 @@
+"""mflux-generate-z-image-teacache: mflux's Z-Image command with TeaCache step skipping."""
+
+import warnings
+from argparse import Namespace
+
+from mflux.cli.parser.parsers import CommandLineParser
+from mflux.models.common.config.model_config import ModelConfig
+from mflux.models.z_image.cli import z_image_generate
+from mflux.models.z_image.cli.z_image_generate import CONDITIONAL_OPTIONS, ZImageCommand
+from mflux.models.z_image.variants.z_image import ZImage
+
+from ._core import Adapter, add_teacache_arguments, run
+
+__all__ = [
+    "ADAPTER",
+    "CONDITIONAL_OPTIONS",
+    "DESCRIPTION",
+    "PIPELINE",
+    "build_parser",
+    "main",
+    "warn_ineffective_options",
+]
+
+# The class ZImageCommand.load() returns; mlx-teacache's match_variant is asked about it.
+PIPELINE = ZImage
+DESCRIPTION = "Generate an image using Z-Image with TeaCache step skipping."
+
+
+def build_parser() -> CommandLineParser:
+    parser = z_image_generate.build_parser()
+    parser.description = DESCRIPTION
+    add_teacache_arguments(parser)
+    return parser
+
+
+# The function below is copied statement for statement from mflux's z_image_generate.main(),
+# which runs it before load(). tests/test_mflux_drift.py fails when it stops matching. It goes
+# away once mflux's Z-Image command has a Command.prepare(args) step the plugin can call.
+def warn_ineffective_options(args: Namespace, model_config: ModelConfig) -> None:
+    # Warn on the EFFECTIVE behavior, not the flag value: --model may resolve to a
+    # guidance-distilled variant (guidance forced to 0.0 regardless of the flag), and on
+    # CFG-capable models the negative prompt is only encoded at guidance > 1.0, where an
+    # omitted --guidance defaults to 0.0.
+    if not model_config.supports_guidance:
+        CommandLineParser.warn_ignored_options(
+            {
+                "--guidance": CONDITIONAL_OPTIONS["--guidance"]["reason"],
+                "--negative-prompt": CONDITIONAL_OPTIONS["--negative-prompt"]["reason"],
+            }
+        )
+    elif CommandLineParser._option_was_provided("--negative-prompt") and (
+        args.guidance is None or args.guidance <= 1.0
+    ):
+        warnings.warn(
+            f"--negative-prompt has no effect: {CONDITIONAL_OPTIONS['--negative-prompt']['reason']}"
+            " Pass --guidance above 1.0 to enable it.",
+            stacklevel=1,
+        )
+
+
+# Z-Image needs one optional hook (after_checks); active_steps keeps the Config-based default, and
+# mlx-teacache has no checkpoint check for Z-Image, so this package's own model-path warning applies.
+ADAPTER = Adapter(
+    command="mflux-generate-z-image-teacache",
+    plain_command="mflux-generate-z-image",
+    pipeline=PIPELINE,
+    family_keys=(z_image_generate.DEFAULT_MODEL, *z_image_generate.FAMILY_MODELS),
+    build_parser=build_parser,
+    validate=ZImageCommand.validate,
+    load=ZImageCommand.load,
+    generate=ZImageCommand.generate,
+    latent_creator=ZImageCommand.latent_creator,
+    after_checks=warn_ineffective_options,
+)
+
+
+def main() -> None:
+    run(ADAPTER)