dougvk/Unlimited-OCR-RDNA4

1#!/usr/bin/env python32from __future__ import annotations3 4import argparse5import copy6import hashlib7import json8from pathlib import Path9 10from unlimited_ocr_rdna4.constants import MODEL_REVISION, VERIFIED_HIP_VERSION, VERIFIED_PACKAGE_VERSIONS11 12 13def _load_json(path: Path) -> dict:14    value = json.loads(path.read_text(encoding="utf-8"))15    if not isinstance(value, dict):16        raise AssertionError(f"expected a JSON object: {path}")17    return value18 19 20def _sha256(path: Path) -> str:21    return hashlib.sha256(path.read_bytes()).hexdigest()22 23 24def _check_output(path: Path) -> None:25    text = path.read_text(encoding="utf-8")26    ordered = (27        "Unlimited OCR RDNA4 Smoke Test",28        "Items",29        "Document scan",30        "Table extraction",31        "Total",32        "Verification notes",33        "RDNA4-OCR-PASS",34    )35    positions = [text.index(marker) for marker in ordered]36    assert positions == sorted(positions), f"reading order mismatch in {path}"37    assert "€48.50" in text, f"euro table total missing from {path}"38    assert "mc}^2" in text or "mc^2" in text, f"formula missing from {path}"39    assert "<table" in text or "| Item" in text, f"structured table missing from {path}"40 41 42def _check_run(payload: dict, *, expected_pages: int) -> None:43    assert payload["schema_version"] == 144    assert payload["status"] == "ok"45    assert payload["model_revision"] == MODEL_REVISION46    assert payload["architecture"] == "gfx1201"47    assert payload["hardware_verified"] is True48    assert payload["software_verified"] is True49    assert payload["pages_processed"] == expected_pages50    assert payload["pages_total"] == expected_pages51    assert 0 < payload["peak_allocated_gib"] < 1652 53 54def main() -> None:55    parser = argparse.ArgumentParser(description="Validate machine-readable RDNA 4 smoke evidence")56    parser.add_argument("--doctor", type=Path, required=True)57    parser.add_argument("--image-run", type=Path, action="append", required=True)58    parser.add_argument("--pdf-run", type=Path, required=True)59    parser.add_argument("--image-output", type=Path, action="append", required=True)60    parser.add_argument("--pdf-output", type=Path, required=True)61    parser.add_argument("--expected-image-sha256")62    parser.add_argument("--expected-pdf-sha256")63    parser.add_argument("--network-isolated", action="store_true")64    parser.add_argument("--gpu-processes", type=Path)65    parser.add_argument("--evidence-output", type=Path)66    args = parser.parse_args()67 68    assert len(args.image_run) == 2, "exactly two image-run JSON files are required"69    assert len(args.image_output) == 2, "exactly two image outputs are required"70    doctor = _load_json(args.doctor)71    assert doctor["schema_version"] == 172    assert doctor["status"] == "ok"73    assert doctor["runtime_issues"] == []74    runtime = doctor["runtime"]75    assert runtime["visible_devices"] == 176    assert runtime["accepted_architecture"] is True77    assert runtime["hardware_verified"] is True78    assert runtime["software_verified"] is True79    assert runtime["hip_version"] == VERIFIED_HIP_VERSION80    assert runtime["package_versions"] == VERIFIED_PACKAGE_VERSIONS81    assert doctor["model"]["prepared"] is True82    assert doctor["model"]["revision"] == MODEL_REVISION83 84    image_runs = [_load_json(run_path) for run_path in args.image_run]85    pdf_run = _load_json(args.pdf_run)86    for image_run in image_runs:87        _check_run(image_run, expected_pages=1)88    _check_run(pdf_run, expected_pages=1)89    for output_path in (*args.image_output, args.pdf_output):90        _check_output(output_path)91 92    image_hashes = [_sha256(path) for path in args.image_output]93    assert image_hashes[0] == image_hashes[1], "fresh-process image outputs are not byte-identical"94    pdf_hash = _sha256(args.pdf_output)95    if args.expected_image_sha256:96        assert image_hashes[0] == args.expected_image_sha25697    if args.expected_pdf_sha256:98        assert pdf_hash == args.expected_pdf_sha25699    if args.network_isolated:100        assert args.network_isolated is True101    processes = _load_json_array(args.gpu_processes) if args.gpu_processes else None102    if processes is not None:103        names = []104        for gpu in processes:105            for process in gpu.get("process_list", []):106                process_info = process.get("process_info")107                if isinstance(process_info, dict) and isinstance(process_info.get("name"), str):108                    names.append(process_info["name"])109        assert not any("python" in name.lower() for name in names), f"OCR Python process still owns VRAM: {names}"110 111    summary = {112        "schema_version": 1,113        "validation": "PASS",114        "network_isolated": args.network_isolated,115        "image_sha256": image_hashes[0],116        "pdf_sha256": pdf_hash,117    }118    if args.evidence_output:119        doctor_evidence = copy.deepcopy(doctor)120        doctor_evidence["model"]["model_dir"] = "<prepared-model>"121        image_evidence = copy.deepcopy(image_runs)122        for index, image_run in enumerate(image_evidence, start=1):123            image_run["input"] = "tests/fixtures/rdna4-smoke.png"124            image_run["output"] = f"<temporary-output>/smoke-image-{index}.md"125        pdf_evidence = copy.deepcopy(pdf_run)126        pdf_evidence["input"] = "tests/fixtures/rdna4-smoke.pdf"127        pdf_evidence["output"] = "<temporary-output>/smoke-pdf.md"128        evidence = {129            **summary,130            "fixtures": {131                "image_sha256": "f5099e17be868abfb4213dbdab220deac82a2db93ba87ab22f03219178246972",132                "pdf_sha256": "cdd2b484d0ac90bd98b489dd97565a65eb17246f713359524cddd41d78cc10cb",133            },134            "doctor": doctor_evidence,135            "image_runs": image_evidence,136            "pdf_run": pdf_evidence,137            "gpu_processes_after": processes,138        }139        args.evidence_output.parent.mkdir(parents=True, exist_ok=True)140        args.evidence_output.write_text(json.dumps(evidence, indent=2, sort_keys=True) + "\n", encoding="utf-8")141    print(json.dumps(summary, sort_keys=True))142 143 144def _load_json_array(path: Path) -> list[dict]:145    value = json.loads(path.read_text(encoding="utf-8"))146    if not isinstance(value, list):147        raise AssertionError(f"expected a JSON array: {path}")148    return value149 150 151if __name__ == "__main__":152    main()153