Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Version numbers follow the pattern: <spaCy-version>-<release-number>

[Unreleased]

Streaming deadlock with korapxmltool on large corpora containing many small documents (e.g. Wikipedia article/discussion dumps), which manifested as near-idle CPU and a progress bar frozen at 0 for hours. The pipe now honors korapxmltool's # eot/# eof document-delimiter protocol: each document is annotated, emitted, and flushed immediately, so the worker pool can deliver results and release its bounded in-flight buffer slots. Previously # eot/ # eof were swallowed as CoNLL-U comments, output was block-buffered with no flush, and nothing was delivered until the process exited at stdin EOF — which deadlocked once the in-flight buffer filled. Inputs without protocol markers (a CoNLL-U file piped in directly) are still processed in SPACY_CHUNK_SIZE-bounded blocks and now stream their output.

Initial release of conllu-spacy-docker
Multi-language support for 70+ languages via spaCy models
CoNLL-U input/output format support
On-demand model fetching with caching in /local/models
Optional GermaLemma integration for enhanced German lemmatization
Morphological features extraction in CoNLL-U format
Optional dependency parsing (HEAD/DEPREL columns)
Command-line options: -h, -m MODEL, -L, -V, -d, -g
Environment variables for configuration (batch size, chunk size, timeouts, etc.)
Model preloading via preload-models.sh script
Three Docker image variants:
- Standard (662 MB) - with GermaLemma
- Slim (490 MB) - without GermaLemma
- With-models (1.22 GB) - includes pre-installed de_core_news_lg model
Optimized Docker image using COPY --chown to avoid layer duplication
CI/CD pipelines for GitLab and GitHub
Progress indicators for model downloads
Non-root user execution for security
List available/installed models with -L flag
Display version information with -V flag