v49_distilled6l: v20 mixed int4/int8 - 116MB, 300k EN 89.1/7.4 oversmash, FR 86.8 DE 77.1 NL 85, secret 35/0, bench EN 76.6

v48_distilled6l: xlm-r-large-teacher distilled 6L (v19) - 300k EN 87.1, FR 81.9, DE 74.7, oversmash 9.7, 130MB, ~3ms

v47_distilled6l: two-teacher distilled 6L (v16) - 300k EN 82.0, FR 72.5, DE 65.3, oversmash 11.1, 130MB, 4.2ms

v46_distilled6l: 6L distilled student (balanced SKU) - 130MB int8, ~5ms CPU, EN 74.3 / FR 69.8 / ES 86.0 / IT 83.3, oversmash 11.4%, secret probe 35/1

Upload v45_phase5_pruned/remap.json with huggingface_hub

Upload v45_phase5_pruned/config.json with huggingface_hub

Upload v45_phase5_pruned/tokenizer.json with huggingface_hub

Upload v45_phase5_pruned/model_quantized.onnx with huggingface_hub

chore: remove retired OpenAI-Privacy-Filter model weights + provenance trail (unused by prod/demo)

chore: remove retired OpenAI-Privacy-Filter model + its provenance trail (served model is screenpipe's own); rewrite NOTICE

docs: rewrite card — screenpipe's own model (not OpenAI Privacy Filter), ScreenLeak numbers, demo link

web demo root layout: onnx/model_quantized.onnx

web demo root layout: special_tokens_map.json

web demo root layout: tokenizer_config.json

web demo root layout: tokenizer.json

web demo root layout: config.json

web demo (transformers.js layout): web/onnx/model_quantized.onnx

web demo (transformers.js layout): web/special_tokens_map.json

web demo (transformers.js layout): web/tokenizer_config.json

web demo (transformers.js layout): web/tokenizer.json

web demo (transformers.js layout): web/config.json

v45 secret-loop champion (iter7 INT8): config.json

v45 secret-loop champion (iter7 INT8): tokenizer.json

v45 secret-loop champion (iter7 INT8): model_quantized.onnx

v45 phase 3 ONNX INT8: config.json

v45 phase 3 ONNX INT8: tokenizer.json

v45 phase 3 ONNX INT8: model_quantized.onnx

contact: correct Google Workspace inbox (louis@screenpi.pe)

contact: correct Google Workspace inbox (louis@screenpi.pe)

contact: correct Google Workspace inbox (louis@screenpi.pe)

public-ready: vague methodology, headline numbers only

upstream-license compliance + rigor disclaimer + correct email

upstream-license compliance + rigor disclaimer + correct email

upstream-license compliance + rigor disclaimer + correct email

upstream-license compliance + rigor disclaimer + correct email

hf: add YAML frontmatter (license, tags, langs)

reframe: PII redactor for accessibility / OCR / computer-use

reframe: PII redactor for accessibility / OCR / computer-use

upload v3 model + card + license

initial commit