jedec-skill

Source PDF notice

The original JEDEC source PDFs are not included in this public repository. Files named JESD209-4E.pdf, JESD209-5C.pdf, and JESD209-6.pdf are dummy placeholders only, included so the expected local filenames are visible.

Tools and generated reference artifacts for converting JEDEC JESD209 LPDDR PDFs into markdown and structured JSONL that can be used by local search/reference skills.

What’s in this repo

• jedec_convert/ — Python package and CLI for conversion/validation
• docs/ — converted LPDDR4/5/6 artifacts
• reports/ — conversion quality reports
• _probe_* — probe/intermediate outputs used during conversion experiments
• sample_docs/ — synthetic sample data for validation/smoke tests
• pdf_to_md.sh, run_all_jedec_convert.sh — helper scripts

Capabilities

• Convert JEDEC JESD209 PDFs to markdown via the Marker backend
• Generate structured outputs such as sections.jsonl, tables.jsonl, figures.jsonl, and manifest.json
• Validate conversion output and emit markdown/json quality reports
• Run smoke tests against extracted terms and metadata

Install

cd /home/eon/work/jesd_skill
python -m pip install -e .
python -m pip install marker-pdf

marker_single must be available on PATH for the convert command.

CLI examples

Convert

python -m jedec_convert convert \
  --input /home/eon/work/jesd_skill/JESD209-5C.pdf \
  --doc JESD209-5C \
  --generation LPDDR5 \
  --revision C \
  --backend marker \
  --output /home/eon/work/jesd_skill/docs/lpddr5

Validate

python -m jedec_convert validate \
  --input /home/eon/work/jesd_skill/docs/lpddr5 \
  --report /home/eon/work/jesd_skill/reports/lpddr5_conversion_report.md

Smoke test

python -m jedec_convert smoke-test \
  --input /home/eon/work/jesd_skill/docs/lpddr5 \
  --terms tRFC,tRCD,tRP,MR13,MR14,MRW,MPC,WCK,ODT,DBI

Output structure

docs/lpddr5/
  clean.md
  sections.jsonl
  tables.jsonl
  figures.jsonl
  manifest.json
  images/

Validation reports land in reports/.

Notes

• The intended runtime path is to use converted artifacts, not parse PDFs at query time.
• Symbol names are preserved rather than normalized away during post-processing.
• Images are referenced by relative path rather than embedded as base64.
• The current repo includes both converter code and generated artifacts for convenience.

Current status

• Package import/CLI structure verified
• Validation and smoke-test flow verified on synthetic sample data
• Real conversion outputs for LPDDR4/5/6 are included in this repo
• Full PDF conversion requires marker-pdf in the local environment