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_build_model.py

@@ -0,0 +1,39 @@
+from pathlib import Path
+
+import scripts.quant.build_model as bm
+
+
+def test_baseline_and_online_need_no_build(tmp_path):
+    assert bm.needs_build("bf16") is False
+    assert bm.needs_build("fp8") is False
+    assert bm.needs_build("awq") is True
+    assert bm.needs_build("bnb4") is True
+
+
+def test_build_is_idempotent_when_output_exists(tmp_path, monkeypatch):
+    out = tmp_path / "awq"
+    out.mkdir()
+    (out / "config.json").write_text("{}", encoding="utf-8")
+    called = {"n": 0}
+
+    def fake_compressor(*a, **k):
+        called["n"] += 1
+
+    monkeypatch.setattr(bm, "_build_compressor", fake_compressor)
+    result = bm.build_model("awq", base_model="datalab-to/surya-ocr-2", out_dir=out, calib_images=[])
+    assert result == out
+    assert called["n"] == 0  # skipped because config.json already present
+
+
+def test_build_dispatches_to_compressor(tmp_path, monkeypatch):
+    out = tmp_path / "gptq"
+    seen = {}
+
+    def fake_compressor(method, base_model, out_dir, calib_images):
+        seen["method"] = method
+        out_dir.mkdir(parents=True, exist_ok=True)
+        (out_dir / "config.json").write_text("{}", encoding="utf-8")
+
+    monkeypatch.setattr(bm, "_build_compressor", fake_compressor)
+    bm.build_model("gptq", base_model="b", out_dir=out, calib_images=[])
+    assert seen["method"] == "gptq"