Suya OCR API — vLLM-backed, OpenAI-compatible OCR service

FastAPI service wrapping the Surya-OCR-2 model (datalab-to) served through vLLM:
legacy /v1/api/ai/* endpoints, an OpenAI-compatible /v1/chat/completions endpoint,
a coalescing request batcher, a local OCR CLI, Docker packaging, multilingual
example outputs, and quantization/concurrency benchmarks.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tests/test_quant_aggregate.py

@@ -0,0 +1,34 @@
+from scripts.quant.aggregate import SUMMARY_FIELDS, build_row, rows_to_csv, rows_to_markdown
+
+
+def test_build_row_defaults_unset_fields_to_none():
+    row = build_row(method="awq", status="ok", t4_deployable=True, mean_cer=0.01)
+    assert row["method"] == "awq"
+    assert row["status"] == "ok"
+    assert row["t4_deployable"] is True
+    assert row["mean_cer"] == 0.01
+    assert row["throughput_rps"] is None
+    assert set(row.keys()) == set(SUMMARY_FIELDS)
+
+
+def test_build_row_rejects_unknown_field():
+    try:
+        build_row(method="awq", bogus=1)
+        assert False, "expected KeyError"
+    except KeyError as exc:
+        assert "bogus" in str(exc)
+
+
+def test_rows_to_csv_has_header_and_order():
+    rows = [build_row(method="bf16", status="ok")]
+    csv_text = rows_to_csv(rows)
+    assert csv_text.splitlines()[0] == ",".join(SUMMARY_FIELDS)
+    assert "bf16" in csv_text.splitlines()[1]
+
+
+def test_rows_to_markdown_renders_failed_row():
+    rows = [build_row(method="gptq", status="failed", error="OOM at load")]
+    md = rows_to_markdown(rows)
+    assert "| gptq |" in md
+    assert "failed" in md
+    assert "OOM at load" in md