-
Wording: faster, not fastest
-
Move the try-it-live links to the top of the card
-
Link the medai.eka.care live demo and launch blog
-
Create config.json
-
Remove gated-auth instructions (repo un-gated)
-
Gated-weights authentication instructions (hf auth login / HF_TOKEN)
-
Intro: the token classifier is the main contribution; OCR is pluggable
-
Remove text_score_threshold bullet (feature reverted in 0.10.1)
-
Document text_score_threshold (default 0.75)
-
Link the Nemotron BYO-OCR and quickstart notebooks
-
Link Nemotron OCR to nvidia/nemotron-ocr-v2
-
Intro: Tesseract suits PDFs/good scans; point difficult images to BYO OCR (e.g. Nemotron OCR)
-
Install: drop the [server] extra line (library usage only here)
-
Install: list the [visual] extra first — the default examples need it
-
Install: define visual entities (image-region PII) and enumerate all six types
-
Install: list all extras; note [visual] is needed for default image detection
-
Intro: name Bring-your-own OCR as a highlighted term
-
Docs: per-transform usage with real outputs, BYO-OCR in main snippet, notebook pointer
-
Docs: symmetric variable names image_redactor / text_redactor
-
Document hash pseudonym strategy (0.9.0)
-
Add positioning intro: image redaction, CPU-light, Indian context, pluggable OCR
-
Rename project to document-pii-redactor; update all links
-
Document bring-your-own-OCR support (0.7.0)
-
Correct licensing: per-model licenses following base-model lineage
-
Fix stale pre-0.4.0 example in deid/anon section
-
License weights under CC-BY-4.0 with attribution and citation section
-
detect() result is now a required transform argument (0.4.0)
-
Restructure usage around detect as the core primitive (0.3.0)
-
Document de-identify and anonymize modes (0.2.0)
-
Add Tesseract install commands, visual-entity examples
-
De-emphasize specific architecture names in prose
-
Add plain-text GitHub repo URL
-
Reference Eka-PII-redactors repo by name, drop hyperlink
-
Update README: fix GitHub link, add text usage, link demo Space
-
Add MiniLM-L12-H384 text PII token classifier (text_PII_redaction v1, 97 BIO labels, entity F1 0.739)
-
Keep YOLO11m as visual detector (better on populated classes incl. seal_stamp)
-
Swap visual detector to YOLO11s (tied accuracy, 2.3x faster, half size)
-
Update YOLO11m visual detector to final post-training weights
-
Upload folder using huggingface_hub
-
initial commit
