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/common/pretrained.py

@@ -0,0 +1,22 @@
+from typing import Optional
+
+from transformers import PreTrainedModel
+from transformers.utils import is_flash_attn_2_available
+
+
+class SuryaPreTrainedModel(PreTrainedModel):
+    # No-op if we pass attention, so we can set attention however we want in the config
+    def _check_and_adjust_attn_implementation(
+        self, attn_implementation: Optional[str], **kwargs
+    ):
+        if attn_implementation is None:
+            try:
+                self._sdpa_can_dispatch(True)
+                attn_implementation = "sdpa"
+            except (ValueError, ImportError):
+                attn_implementation = "eager"
+
+            if self._supports_flash_attn and is_flash_attn_2_available():
+                attn_implementation = "flash_attention_2"
+
+        return attn_implementation