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_plot.py

@@ -0,0 +1,28 @@
+from scripts.quant.aggregate import build_row
+from scripts.quant.plot import pareto_points, render_all
+
+
+def test_pareto_points_skips_failed_and_missing():
+    rows = [
+        build_row(method="bf16", status="ok", mean_latency_s=6.0, mean_cer=0.0, t4_deployable=True),
+        build_row(method="awq", status="ok", mean_latency_s=4.0, mean_cer=0.01, t4_deployable=True),
+        build_row(method="gptq", status="failed", t4_deployable=True),
+        build_row(method="fp8", status="ok", mean_latency_s=3.0, mean_cer=0.02, t4_deployable=False),
+    ]
+    points = pareto_points(rows, x_key="mean_latency_s", y_key="mean_cer")
+    methods = {p["method"] for p in points}
+    assert methods == {"bf16", "awq", "fp8"}  # gptq failed -> skipped
+    awq = next(p for p in points if p["method"] == "awq")
+    assert awq["x"] == 4.0 and awq["y"] == 0.01 and awq["t4_deployable"] is True
+
+
+def test_render_all_writes_png_files(tmp_path):
+    rows = [
+        build_row(method="bf16", status="ok", mean_latency_s=6.0, mean_cer=0.0,
+                  mean_bbox_iou=1.0, t4_deployable=True),
+        build_row(method="awq", status="ok", mean_latency_s=4.0, mean_cer=0.01,
+                  mean_bbox_iou=0.98, t4_deployable=True),
+    ]
+    written = render_all(rows, tmp_path)
+    assert all(p.exists() for p in written)
+    assert any(p.name == "pareto_latency_cer.png" for p in written)