mammofm 0.1.0__tar.gz

mammofm-0.1.0/LICENSE.txt

@@ -0,0 +1,202 @@
+
+                                 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.

mammofm-0.1.0/LICENSE_WEIGHTS.txt

@@ -0,0 +1,80 @@
+Custom Academic License for Model Weights
+(Based on CC BY-NC 4.0)
+
+1. Overview
+This license governs the use of the accompanying model weights, code, and associated materials (“Model”).
+This license is based on the Creative Commons Attribution-NonCommercial 4.0 International License, with additional restrictions specific to machine learning models.
+By accessing, downloading, or using the Model, you agree to the terms below.
+2. Permitted Use
+The Model may be used solely for non-commercial academic research purposes.
+
+Permitted users include:
+- Academic institutions
+- Non-profit research organizations
+Allowed activities:
+- Research use
+- Evaluation and benchmarking
+- Fine-tuning or linear probing for academic research only
+
+3. Prohibited Use
+The following uses are strictly prohibited:
+
+3.1 Commercial Use
+- Any commercial use
+- Any use by for-profit entities without prior written approval
+- Use in products, services, or revenue-generating activities
+
+3.2 Industry Research Without Approval
+- Any use by industry, including internal research, evaluation, or development, without explicit prior written permission
+
+3.3 Clinical and Medical Use
+- Any clinical deployment
+- Any diagnostic or treatment use
+- Any use in medical decision-making systems
+
+4. Restrictions on Model Use
+
+4.1 Redistribution
+- Redistribution of the Model weights, in whole or in part, is not permitted
+
+4.2 Hosting and Serving
+- Hosting, serving, or making the Model available through an API, SaaS platform, web service, or other remote access mechanism is not permitted
+
+4.3 Distillation and Model Extraction
+- Distillation, compression, extraction, imitation, or transfer of knowledge from the Model to create, train, initialize, or improve another model is not permitted
+
+5. Derivative Works
+Derivative models are allowed only under the following conditions:
+- They must be used solely for non-commercial academic research
+- They must not be shared publicly or redistributed
+- They must inherit this same license in full
+- They must not be used to circumvent restrictions in this license
+
+6. Attribution
+Any use of the Model must include:
+- Proper citation of the associated publication(s)
+- Acknowledgment of the original authors and institution(s)
+Example acknowledgment:
+“This work uses [MODEL NAME] developed by [AUTHORS], [INSTITUTION].”
+7. Access Control
+The Licensor reserves the right to:
+- Restrict access to the Model
+- Require user registration or approval
+- Revoke access in case of misuse or violation of this license
+
+8. No Warranty
+The Model is provided “as is”, without warranty of any kind, express or implied, including but not limited to warranties of merchantability, fitness for a particular purpose, or non-infringement.
+The authors and licensors make no guarantees regarding performance, safety, legality, or correctness.
+
+9. Limitation of Liability
+In no event shall the authors, licensors, or affiliated institutions be liable for any claim, damages, or other liability arising from, out of, or in connection with the Model or its use or misuse.
+
+10. Termination
+Any violation of this license results in immediate termination of all rights granted under this license.
+Upon termination, the user must cease all use of the Model and destroy all copies in their possession or control.
+
+11. Contact for Permissions
+For commercial use, industry research, or other exceptions, contact:
+contact:
+Kayhan Batmanghelich
+email: batman@bu.edu

mammofm-0.1.0/MANIFEST.in

@@ -0,0 +1,17 @@
+include pyproject.toml LICENSE.txt LICENSE_WEIGHTS.txt
+include docs/python-package.md
+exclude README.md
+recursive-include src/mammofm *.py *.json LICENSE* NOTICE*
+prune src/codebase
+prune src/preprocessing
+prune src/scripts
+prune src/nltk_data
+prune src/zs_prompts
+prune Test
+prune tests
+prune doc
+prune .git
+prune .idea
+prune .codex
+prune .claude
+global-exclude *.pyc *.pyo .DS_Store .env *.pth *.pt *.tar *.zip *.dcm *.dicom *.png

mammofm-0.1.0/PKG-INFO

@@ -0,0 +1,312 @@
+Metadata-Version: 2.4
+Name: mammofm
+Version: 0.1.0
+Summary: Research-only mammography diagnosis and image features with Mammo-FM
+Author: Batman Lab
+Project-URL: Homepage, https://github.com/batmanlab/Mammo-FM
+Project-URL: Documentation, https://github.com/batmanlab/Mammo-FM/blob/main/docs/python-package.md
+Project-URL: Checkpoints, https://huggingface.co/batmanLab/Mammo-FM
+Keywords: mammography,Mammo-FM,medical-imaging,research
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Science/Research
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE.txt
+License-File: LICENSE_WEIGHTS.txt
+License-File: src/mammofm/_vendor/LICENSE-EFFICIENTNET.txt
+Requires-Dist: torch<3,>=2.6
+Requires-Dist: numpy<3,>=1.24
+Requires-Dist: Pillow<13,>=9
+Requires-Dist: opencv-python-headless<5,>=4.8
+Requires-Dist: pydicom<4,>=2.4
+Requires-Dist: huggingface-hub<1,>=0.27
+Requires-Dist: transformers<5,>=4.48
+Requires-Dist: safetensors>=0.4
+Provides-Extra: dicom
+Requires-Dist: pylibjpeg<3,>=2; extra == "dicom"
+Requires-Dist: pylibjpeg-libjpeg<3,>=2; extra == "dicom"
+Requires-Dist: pylibjpeg-openjpeg<3,>=2; extra == "dicom"
+Provides-Extra: dev
+Requires-Dist: build>=1.2; extra == "dev"
+Requires-Dist: twine>=6; extra == "dev"
+Dynamic: license-file
+
+# mammofm Python package
+
+Local, single-image inference with Mammo-FM. This first package provides
+`mammofm.diagnosis` and `mammofm.features`. Prognosis and report generation are
+not exposed by this release; use the repository's separate workflows for now.
+
+**Research only.** Outputs are uncalibrated model scores, not clinical diagnoses
+or validated patient-level disease probabilities. Read the repository's
+`LICENSE.txt` and `LICENSE_WEIGHTS.txt`; packaging does not change their terms.
+Images stay on your machine. The package has no report-generation API key,
+telemetry, image upload, or remote inference service.
+
+## Contents
+
+- [Installation](#installation)
+- [Diagnosis](#diagnosis)
+- [Supported tasks and missing models](#supported-tasks-and-missing-models)
+- [Image features](#image-features)
+- [DICOM and PNG preprocessing](#dicom-and-png-preprocessing)
+- [Checkpoints, caching, and offline use](#checkpoints-caching-and-offline-use)
+- [API reference](#api-reference)
+- [Testing and publishing](#testing-and-publishing)
+
+## Installation
+
+Python 3.10+ and PyTorch 2.6+ are required. CPU inference is supported; a compatible
+CUDA-enabled PyTorch installation is recommended for repeated inference. Install
+the appropriate PyTorch build for your machine before installing this package.
+
+From the repository root, before the first PyPI publication:
+
+```bash
+python -m pip install .
+# Include decoders for compressed mammography DICOMs:
+python -m pip install '.[dicom]'
+```
+
+The distribution name is `mammofm`. **Once the maintainer publishes it to PyPI**, users can run:
+
+```bash
+python -m pip install 'mammofm[dicom]'
+# For preprocessed PNGs and uncompressed DICOMs only:
+python -m pip install mammofm
+```
+
+These are the intended public installation commands, not a claim that a PyPI
+release has already been published. No model weights or patient data are bundled
+in the wheel or source distribution.
+
+On a shared CPU host with a low process/thread limit, set these before starting
+Python to bound numerical and tokenizer worker pools:
+
+```bash
+export OMP_NUM_THREADS=2 MKL_NUM_THREADS=2 OPENBLAS_NUM_THREADS=2
+export TOKENIZERS_PARALLELISM=false TORCHINDUCTOR_COMPILE_THREADS=1
+```
+
+You can additionally set `cv2.setNumThreads(2)` in your application. The package
+does not change process-wide thread settings. The latter PyTorch setting avoids
+an import-time compiler worker pool in some Transformers versions even though
+Mammo-FM uses eager, noncompiled inference.
+
+## Diagnosis
+
+```python
+import json
+import mammofm as mf
+
+output = mf.diagnosis('/data/mammogram.dcm', 'vindr')
+print(json.dumps(output, indent=2, allow_nan=False))
+
+# A preprocessed PNG uses the same API.
+output = mf.diagnosis('/data/mammogram.png', 'nlbreast')
+```
+
+`output` is a Python dictionary that can be directly serialized to JSON. Its
+top-level prediction groups are `zero_shot`, `linear_probe`, and `full_finetune`.
+Each maps task names to positive-class probabilities (Python floats) or `None`
+(`null` in JSON) when that specific model has not been released.
+
+```python
+output['zero_shot']['mass']
+output['linear_probe']['mass']
+output['full_finetune']['mass']
+output['unavailable']  # Reasons for unavailable fields; never substitute zeros.
+output['metadata']     # Input processing, pinned checkpoint identities, semantics.
+```
+
+Only binary positive-class scores are returned, not hard labels or a clinical
+threshold. Classifiers use sigmoid. Zero-shot preserves the official evaluator:
+normalize individual projected prompt embeddings, average within negative and
+positive classes, compute cosine similarity to each mean, and apply a two-class
+softmax. **The checkpoint's learned logit scale is not multiplied into these
+scores**, matching the original evaluation.
+
+To avoid unnecessary models or downloads:
+
+```python
+output = mf.diagnosis('/data/mammogram.png', 'rsna', methods=['linear_probe'])
+print(mf.availability('rsna'))  # Does not load or download any weights.
+```
+
+Missing published models produce a warning and explicit null fields. Use
+`strict=True` to raise `MissingModelError` before any model download if any
+requested task/method is unavailable. Download failures, invalid inputs,
+checkpoint mismatches, and actual inference failures always raise errors rather
+than masquerading as a missing model.
+
+## Supported tasks and missing models
+
+Registry pinned to HF revision `5e07ca4c4e88cd118b0c2cfaacadddfe69f30026`:
+
+| Dataset argument | Task key | Zero-shot | Linear probe | Full finetune |
+| --- | --- | --- | --- | --- |
+| `rsna` | `cancer` | Available | Available | Available |
+| `rsna` | `invasive_cancer` | Not released | Not released | Not released |
+| `vindr` | `abnormality` | Available, historical surrogate | Available | Available |
+| `vindr` | `mass` | Available | Available | Available |
+| `vindr` | `suspicious_calcification` | Available | Available | Available |
+| `nlbreast` | `cancer` | Available | Available | Available |
+
+NL-Breast aliases such as `nl_breast` and `NL-Breast` are accepted. Unknown
+diagnosis datasets are rejected. RSNA cancer is **not** an invasive-cancer
+substitute. CMMD and LUMINA diagnosis are deferred and rejected by this release;
+only RSNA, VinDr, and NL-Breast are enabled.
+
+VinDr abnormality means **BI-RADS 4/5 versus 1/2**, with BI-RADS 3 excluded in
+evaluation. It does not mean any finding or pathology-confirmed cancer. Its
+historical zero-shot surrogate uses the original `cancer` / `no cancer` prompts;
+this limitation is also recorded in the output metadata.
+
+## Image features
+
+```python
+import numpy as np
+import mammofm as mf
+
+output = mf.features('/data/mammogram.dcm', 'rsna')
+spatial = output['before_pooling']
+pooled = output['after_pooling']
+print(spatial.shape)  # (1, 2048, 48, 29) at the standard image size
+print(pooled.shape)   # (1, 2048)
+np.savez('/outputs/image_features.npz', before_pooling=spatial, after_pooling=pooled)
+
+# An unknown or omitted dataset uses general min-max DICOM preprocessing.
+output = mf.features('/data/mammogram.png', 'my_dataset')
+output = mf.features('/data/mammogram.dcm')
+```
+
+The arrays are full float32 CPU NumPy arrays, retaining the batch dimension.
+They are the final image-encoder spatial map and its global average-pooled
+vector, **before the CLIP projection and L2 normalization**. Feature arrays are
+not automatically converted to enormous JSON lists. `metadata` identifies the
+checkpoint and preprocessing. Use `.tolist()` explicitly if JSON is required.
+
+Both feature extraction and zero-shot use the requested
+[`Mammo-FM_ASU_Trained_CLIP.tar`](https://huggingface.co/batmanLab/Mammo-FM/blob/5e07ca4c4e88cd118b0c2cfaacadddfe69f30026/Mammo-FM_ASU_Trained_CLIP.tar).
+At this pinned revision it is byte-identical to the Batmanlab CLIP release:
+SHA-256 `58ae959de52d563acc1d43f50571498eddcd4fb7ca599e706540fdd319e3599b`.
+The classifier paths load their own complete released encoder+head weights;
+fine-tuned encoders are not replaced with the ASU encoder.
+
+## DICOM and PNG preprocessing
+
+`img` is a local path to one single-frame grayscale DICOM or an already
+preprocessed 8-bit PNG. RGB PNGs are also accepted. Color/multiframe DICOMs,
+16-bit PNGs, and transparent PNGs are rejected rather than silently altered.
+
+| Dataset | DICOM intensity route |
+| --- | --- |
+| RSNA, VinDr | Original CPU-style min-max scaling |
+| NL-Breast | DICOM rescale + window center/width; 1st/99th-percentile fallback |
+| Unknown/omitted dataset in `features` | General min-max preprocessing |
+
+DICOMs use the unified converter's five-pixel border removal, MONOCHROME1
+correction, original breast-cropping heuristic, and `INTER_AREA` resize to
+1520H x 912W. Conversion is **in memory**; no temporary PNG or DICOM metadata is
+written. PNGs skip this contrast/cropping pipeline. PNG input should already
+have the intended breast preprocessing; being a PNG does not establish that.
+
+All inputs then receive the model's RGB conversion, branch-specific resize
+when needed (`INTER_LINEAR`), per-image min-max normalization and
+`(x - 0.3089279) / 0.25053555408335154`. Constant images are handled without NaNs.
+
+All enabled branches use **1520H x 912W**. For arbitrary PNG sizes, this resize
+is a package standardization step; the original non-CMMD loaders expected
+already-preprocessed PNG dimensions and did not apply it. Image preprocessing
+does not update detection boxes, perform anatomical segmentation or anonymize
+burned-in text. No patient identifiers or source paths are included in results.
+
+The shared image operations preserve the original CPU conventions, not a
+guarantee of byte-identical behavior for every legacy GPU/decoder path.
+
+## Checkpoints, caching, and offline use
+
+The first relevant inference downloads weights from the public
+[`batmanLab/Mammo-FM`](https://huggingface.co/batmanLab/Mammo-FM) repository. The
+ASU joint checkpoint is about 2.15 GB; each classifier is downloaded separately
+only when requested. No HF token is required for this pinned public release.
+
+```python
+model = mf.MammoFM(device='cuda', cache_dir='/models/huggingface')
+result = model.diagnosis('/data/image.png', 'vindr')
+features = model.features('/data/image.png', 'vindr')
+model.clear_cache()  # Release this client's model references; keep disk cache.
+```
+
+For offline use, first download the published assets, preserving their HF
+relative paths. Then use:
+
+```python
+model = mf.MammoFM(device='cpu', model_dir='/models/Mammo-FM', local_files_only=True)
+output = model.diagnosis('/data/image.png', 'nlbreast')
+```
+
+The base checkpoint sits directly under `model_dir`; classifiers remain under
+`linear_probe_mammo_FM_image_encoder/...` and `fullfinetune_mammo_FM_image_encoder/...`.
+Zero-shot also requires `config.json`, `tokenizer.json`, `tokenizer_config.json`,
+and `special_tokens_map.json` under `text_encoders/modernbert_mlm_5epochs/`.
+It does **not** require that directory's separate MLM initialization weights:
+fine-tuned text weights are already in the joint CLIP checkpoint.
+
+`MAMMOFM_MODEL_DIR` and `MAMMOFM_CACHE_DIR` provide optional environment defaults.
+Otherwise, downloads use the standard Hugging Face cache, not the codebase.
+Every asset is pinned to a revision and SHA-256 checked before loading. Model
+loading is strict; there is no silent random-weight or ImageNet-weight fallback.
+
+## API reference
+
+```python
+mf.diagnosis(img, dataset, *, methods=None, strict=False, device=None,
+             cache_dir=None, model_dir=None, local_files_only=False)
+mf.features(img, dataset=None, *, device=None, cache_dir=None,
+            model_dir=None, local_files_only=False)
+mf.availability(dataset=None)
+mf.clear_cache()
+
+mf.MammoFM(*, device=None, cache_dir=None, model_dir=None,
+           local_files_only=False, prompt_batch_size=32)
+```
+
+`device=None` selects CUDA when available, otherwise CPU. `methods` selects
+`zero_shot`, `linear_probe`, and/or `full_finetune`. All three are requested by
+default. A reusable client caches its encoder, CLIP model, prompt prototypes,
+and at most two classifier models in memory. For parallel workers, construct
+one client per worker; clients are not promised to be thread-safe.
+
+## Testing and publishing
+
+Synthetic tests do not need private data, downloaded checkpoints or GPUs:
+
+```bash
+python -m unittest discover -s tests/package -p 'test_*.py'
+```
+
+Build and check release artifacts from a clean checkout:
+
+```bash
+python -m pip install '.[dev]'
+python -m build
+python -m twine check dist/*
+```
+
+Before publishing, inspect both archives: they must contain only the inference
+package, public prompts, documentation, packaging metadata and license notices.
+Do not include weights, images, private CSVs, test-inference artifacts, secrets,
+editor/assistant files, or the original training code tree. Run an installed-wheel
+smoke test outside the repository so accidental checkout imports cannot pass.
+
+PyPI publication is a separate maintainer action requiring project ownership
+and an authenticated upload or trusted-publishing configuration. Do not embed
+upload tokens in the repository. Only after release review and publication:
+
+```bash
+python -m twine upload dist/*
+```