Add tesseract-ocr.py CPU baseline; fix pp-ocrv6 inference_info key (#74)

- tesseract-ocr.py: Tesseract 5 baseline following the collection
  contract (config PRs, inference_info with column_name/model_id,
  same CLI surface). Installs the tesseract binary via apt at startup
  (not in the default Jobs image); CPU-only, --lang/--psm/--oem
  exposed, threaded workers. Smoke-tested locally: full OCR
  round-trip on rendered text + real Britannica pages.
- pp-ocrv6.py: write column_name (the key ocr-bench column discovery
  reads) alongside output_column — previously only the markdown-name
  heuristic made discovery work, and a non-default --output-column
  would have matched the pp_ocr_blocks JSON column instead.
- README: add tesseract row to the models table.

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

Author: davanstrien

ocr/README.md

@@ -49,7 +49,8 @@ _Sorted by model size:_
 
 | Script | Model | Size | Backend | Notes |
 |--------|-------|------|---------|-------|
-| `pp-ocrv6.py` | [PP-OCRv6](https://huggingface.co/collections/PaddlePaddle/pp-ocrv6) | 1.5–34.5M | PaddleOCR (paddle) | **Smallest by far** — classical det+rec pipeline, not a VLM. Three tiers (`--model-tier tiny\|small\|medium`), plain-text output (not markdown). 50 langs. Runs on `t4-small`. Apache 2.0 |
+| `tesseract-ocr.py` | [Tesseract 5](https://github.com/tesseract-ocr/tesseract) | n/a (classical) | pytesseract (CPU) | **The legacy baseline** — no GPU at all, runs on `cpu-upgrade`. Plain-text output, `--lang`/`--psm`/`--oem` exposed, 100+ language packs via apt. Apache 2.0 |
+| `pp-ocrv6.py` | [PP-OCRv6](https://huggingface.co/collections/PaddlePaddle/pp-ocrv6) | 1.5–34.5M | PaddleOCR (paddle) | **Smallest neural** — classical det+rec pipeline, not a VLM. Three tiers (`--model-tier tiny\|small\|medium`), plain-text output (not markdown). 50 langs. Runs on `t4-small`. Apache 2.0 |
 | `falcon-ocr.py` | [Falcon-OCR](https://huggingface.co/tiiuae/Falcon-OCR) | 0.3B | falcon-perception | Smallest VLM in collection. #1 on multi-column docs and tables (olmOCR), Apache 2.0 |
 | `smoldocling-ocr.py` | [SmolDocling](https://huggingface.co/ds4sd/SmolDocling-256M-preview) | 256M | Transformers | DocTags structured output |
 | `surya-ocr.py` | [Surya OCR 2](https://huggingface.co/datalab-to/surya-ocr-2) | 0.65B | vLLM | **Structured** OCR + `--task layout\|table`: per-block HTML with bboxes & reading order in an extra `surya_blocks` column. 91 langs, top-under-3B on olmOCR-Bench. Modified OpenRAIL-M license. Needs the **pinned** `vllm/vllm-openai:v0.20.1` image |

ocr/pp-ocrv6.py

@@ -701,6 +701,9 @@ def build_inference_entry(tier: str, det_model: str, rec_model: str, args_dict:
         "rec_accuracy_pct": TIER_REC.get(tier),
         "languages": TIER_LANGUAGES.get(tier, ""),
         "engine": "paddle_static",
+        # column_name is the key ocr-bench's column discovery reads; keep
+        # output_column too for backward compat with existing outputs.
+        "column_name": args_dict.get("output_column", "markdown"),
         "output_column": args_dict.get("output_column", "markdown"),
         "blocks_column": "pp_ocr_blocks",
         "timestamp": datetime.now(timezone.utc).isoformat(),