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_run_all.py

@@ -0,0 +1,29 @@
+import scripts.quant.run_all as ra
+from scripts.quant.aggregate import SUMMARY_FIELDS
+
+
+def test_run_method_failure_becomes_failed_row(monkeypatch):
+    def boom(*a, **k):
+        raise RuntimeError("OOM at load")
+
+    monkeypatch.setattr(ra, "_measure_method", boom)
+    row = ra.run_method("gptq", base_model="b", work_dir=ra.Path("/tmp/x"),
+                        eval_images=[], reference_dir=ra.Path("/tmp/ref"))
+    assert row["method"] == "gptq"
+    assert row["status"] == "failed"
+    assert "OOM at load" in row["error"]
+    assert set(row.keys()) == set(SUMMARY_FIELDS)
+
+
+def test_run_method_success_passes_through_metrics(monkeypatch):
+    def fake_measure(method, base_model, work_dir, eval_images, reference_dir):
+        return {"mean_cer": 0.01, "mean_bbox_iou": 0.98, "mean_latency_s": 4.2,
+                "model_size_mb": 512.0}
+
+    monkeypatch.setattr(ra, "_measure_method", fake_measure)
+    row = ra.run_method("awq", base_model="b", work_dir=ra.Path("/tmp/x"),
+                        eval_images=[], reference_dir=ra.Path("/tmp/ref"))
+    assert row["status"] == "ok"
+    assert row["mean_cer"] == 0.01
+    assert row["t4_deployable"] is True
+    assert row["mean_latency_s"] == 4.2