Automated deployment from GitHub (source commit: 95e226)

docs/ADRs/001_design_of_huggingface_space_dockerfile.md

@@ -0,0 +1,32 @@
+# Design of HuggingFace Space Dockerfile
+
+- Status: Ongoing
+- Deciders: Martin Lellep (@PellelNitram)
+- Drivers: Martin Lellep (@PellelNitram)
+- PRD: None
+- Date: 2025-10-04
+
+## Context
+
+*Explain the background and the context in which the decision is being made. Include any relevant information about the problem, constraints, or goals.*
+
+## Decisions
+
+*State the decision that has been made. Be clear and concise.*
+
+- In the future, download models at build time into the Docker image from Github release page. In the
+  very far future, pull them from HuggingFace at run-time.
+- Add `xournalpp` binary to Docker image so that the `xopp` file can be exported as PDF prior to
+  execution of the HTR pipeline.
+
+## Consequences
+
+*Describe the consequences of the decision. Include both positive and negative outcomes, as well as any trade-offs.*
+
+## Alternatives Considered
+
+*List and briefly describe other options that were considered and why they were not chosen.*
+
+## References
+
+*Include links or references to any supporting documentation, discussions, or resources.*

docs/ADRs/002_use_HuggingFace_ecosystem_for_ML.md

@@ -0,0 +1,52 @@
+# ADR 002 – Use Hugging Face Ecosystem for Machine Learning
+
+- Date: 2025-11-03
+- Status: Accepted
+- PRD: None
+- Drivers: Martin Lellep ([@PellelNitram](https://github.com/PellelNitram/))
+- Deciders: Martin Lellep ([@PellelNitram](https://github.com/PellelNitram/))
+
+## Context
+
+The project originally used plain PyTorch with local files for datasets and model storage.
+This made it hard to share and version models; once trained, they essentially lived on a hard drive with no central management or deployment integration and I had to rely on Google Drive and Dropbox links.
+Also, training models on different machines is cumbersome because one needs to manually download the dataset every time.
+
+## Decision
+
+Adopt the **[Hugging Face ecosystem](https://huggingface.co/)** for all machine learning–related components, including:
+
+* Model Hub: hosting and versioning trained models
+* Dataset Hub: storing and sharing datasets
+* `transformers` and `datasets` libraries: for training and data handling
+* Trainer API: for standard training workflows
+
+Note: We need to agree on a good naming scheme for storing and retrieving models and datasets efficiently on HuggingFace.
+This will be the subject of a future ADR.
+
+## Rationale
+
+Hugging Face offers a free, community-maintained platform that is now the industry standard for open ML projects.
+It provides built-in versioning and sharing, making it easy to pull models directly in demos or end-user environments.
+The same applies to retrieving properly versioned datasets for training, benchmarking, and various demo use cases (e.g., providing sample data in Gradio applications).
+
+## Consequences
+
+### Pros
+
+* Centralized and versioned model/dataset hosting
+* Easier sharing, collaboration, and reproducibility
+* Straightforward integration in deployments by letting HuggingFace download the model automatically
+* Large and active community support
+* One can either fully integrate the model by subclassing `PreTrainedModel` or use it as plain artifact storage of
+  the binary weights file
+
+### Cons
+
+* Requires learning new APIs and conventions
+* Custom training routines may need workarounds
+* Invest time to learn how to convert a PyTorch model into a HF model, incl pre and post processing code
+
+## Alternatives
+
+Continuing with plain PyTorch and local storage would have been simpler but lacked any versioning, reproducibility, or sharing capabilities.

docs/ADRs/template.md

@@ -0,0 +1,33 @@
+# ADR 002 – Use Hugging Face Ecosystem for Machine Learning
+
+- Date: YYYY-MM-DD
+- Status: Accepted or Ongoing or Superseeded by [ADR]()
+- PRD: None
+- Drivers: Name ([Link to Github handle](https://github.com/))
+- Deciders: Name ([Link to Github handle](https://github.com/))
+
+## Context
+
+*(Add text here.)*
+
+## Decision
+
+*(Add text here.)*
+
+## Rationale
+
+*(Add text here.)*
+
+## Consequences
+
+### Pros
+
+*(Add bulletpoints here).*
+
+### Cons
+
+*(Add bulletpoints here).*
+
+## Alternatives
+
+*(Add text here.)*

mkdocs.yml

@@ -26,6 +26,9 @@ nav:
   - Getting Started as Developer:
     - Installation: 'installation_developer.md'
     - Developer Guide: 'developer_guide.md'
+    - ADRs:
+      - 001 Design of HuggingFace Space Dockerfile: 'ADRs/001_design_of_huggingface_space_dockerfile.md'
+      - 002 Use HuggingFace ecosystem for ML: 'ADRs/002_use_HuggingFace_ecosystem_for_ML.md'
     # - Data Collection: 'data_collection.md' # Unclear if even needed
     # - Developing New Models: 'developing_new_models.md' # Very unclear what to write as I haven't built anything yet
   - Contributing: 'contributing.md'