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>

surya/ocr_error/loader.py

@@ -0,0 +1,48 @@
+from typing import Optional
+
+
+from surya.common.load import ModelLoader
+from surya.logging import get_logger
+from surya.ocr_error.model.config import DistilBertConfig
+from surya.ocr_error.model.encoder import DistilBertForSequenceClassification
+from surya.ocr_error.tokenizer import DistilBertTokenizer
+from surya.settings import settings
+
+logger = get_logger()
+
+
+class OCRErrorModelLoader(ModelLoader):
+    def __init__(self, checkpoint: Optional[str] = None):
+        super().__init__(checkpoint)
+
+        if self.checkpoint is None:
+            self.checkpoint = settings.OCR_ERROR_MODEL_CHECKPOINT
+
+    def model(
+        self,
+        device=settings.TORCH_DEVICE_MODEL,
+        dtype=settings.MODEL_DTYPE,
+        attention_implementation: Optional[str] = None,
+    ) -> DistilBertForSequenceClassification:
+        if device is None:
+            device = settings.TORCH_DEVICE_MODEL
+        if dtype is None:
+            dtype = settings.MODEL_DTYPE
+
+        config = DistilBertConfig.from_pretrained(self.checkpoint)
+        model = (
+            DistilBertForSequenceClassification.from_pretrained(
+                self.checkpoint,
+                dtype=dtype,
+                config=config,
+            )
+            .to(device)
+            .eval()
+        )
+
+        return model
+
+    def processor(
+        self, device=settings.TORCH_DEVICE_MODEL, dtype=settings.MODEL_DTYPE
+    ) -> DistilBertTokenizer:
+        return DistilBertTokenizer.from_pretrained(self.checkpoint)