- start_period 60s → 120s: Zenodo download on cold start can exceed 60s on slow connections
- ocr_cache volume comment: documents what the cache stores for future operators
- .env.example: add token generation command to prevent weak placeholder in production
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add aria-expanded + aria-controls to expand button (WCAG 4.1.2)
- Add id="training-history-rows" to tbody for aria-controls target
- Replace title= tooltip on FAILED badge with details/summary for keyboard
and touch accessibility; add training_error_detail_label i18n key
- Use motion-safe:animate-pulse on RUNNING badge for prefers-reduced-motion
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- _model_is_loadable: narrow bare except to (RuntimeError, OSError, ValueError)
with DEBUG-level fallback for unexpected exceptions — prevents silent masking
of missing kraken install or AttributeError on vgsl
- _run_segtrain: replace bare except:pass with log.warning so height-check
fallback is visible in container logs
- New test_ensure_blla_model.py: covers model-OK early return, incompatible
model rename+replace, and missing model download paths
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Both training panels (OCR and segmentation) share TrainingHistory.
Show only the 3 most recent runs by default; render a Mehr/Weniger
anzeigen button when there are more.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
setCer() was called for recognition training but not for segmentation.
The OCR service now returns cer = 1 - accuracy for segtrain; persist it
so the admin panel can display Fehlerrate for both training types.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Pass OCR_TRAINING_TOKEN through to the backend container as
APP_OCR_TRAINING_TOKEN so RestClientOcrClient sends the X-Training-Token
header when calling /train and /segtrain.
- Raise mem_limit/memswap_limit from 8g to 12g to give segtrain headroom
on hosts with more available RAM.
- Uncomment OCR_TRAINING_TOKEN in .env.example — it is now required.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Three issues fixed:
1. --resize both was removed in ketos 7; replaced with --resize union
which extends the model's class mapping to include training data classes.
2. ketos ignores -s when -i is present, so the 1800px blla model caused
7+ GB peak RAM and OOM-killed the host (no swap, 5 GB free).
Now checks the loaded model's input height: only uses the base model
when it was already fine-tuned at 800px; otherwise trains from scratch
at 800px (~200 MB peak). After the first run the trained 800px model
becomes the base for all subsequent fine-tuning runs.
3. segtrain now computes and returns cer = 1 - accuracy, matching the
recognition training path.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds ensure_blla_model.py which loads the blla segmentation model with
ketos on every container start. If the model is missing or in the legacy
PyTorch ZIP format (incompatible with ketos 7), it re-downloads the
correct CoreML protobuf model from Zenodo (DOI 10.5281/zenodo.14602569).
The Dockerfile now uses entrypoint.sh which runs this check before
starting uvicorn.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
