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>
2026-06-17 10:20:02 +04:00
commit 1a585693be
147 changed files with 13827 additions and 0 deletions
@@ -0,0 +1,68 @@
+from collections import OrderedDict
+from typing import Mapping
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.onnx import OnnxConfig
+
+from surya.common.s3 import S3DownloaderMixin
+
+ID2LABEL = {
+    0: 'good',
+    1: 'bad'
+}
+
+class DistilBertConfig(S3DownloaderMixin, PretrainedConfig):
+    model_type = "distilbert"
+    attribute_map = {
+        "hidden_size": "dim",
+        "num_attention_heads": "n_heads",
+        "num_hidden_layers": "n_layers",
+    }
+
+    def __init__(
+        self,
+        vocab_size=30522,
+        max_position_embeddings=512,
+        sinusoidal_pos_embds=False,
+        n_layers=6,
+        n_heads=12,
+        dim=768,
+        hidden_dim=4 * 768,
+        dropout=0.1,
+        attention_dropout=0.1,
+        activation="gelu",
+        initializer_range=0.02,
+        qa_dropout=0.1,
+        seq_classif_dropout=0.2,
+        pad_token_id=0,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.sinusoidal_pos_embds = sinusoidal_pos_embds
+        self.n_layers = n_layers
+        self.n_heads = n_heads
+        self.dim = dim
+        self.hidden_dim = hidden_dim
+        self.dropout = dropout
+        self.attention_dropout = attention_dropout
+        self.activation = activation
+        self.initializer_range = initializer_range
+        self.qa_dropout = qa_dropout
+        self.seq_classif_dropout = seq_classif_dropout
+        super().__init__(**kwargs, pad_token_id=pad_token_id)
+
+
+class DistilBertOnnxConfig(OnnxConfig):
+    @property
+    def inputs(self) -> Mapping[str, Mapping[int, str]]:
+        if self.task == "multiple-choice":
+            dynamic_axis = {0: "batch", 1: "choice", 2: "sequence"}
+        else:
+            dynamic_axis = {0: "batch", 1: "sequence"}
+        return OrderedDict(
+            [
+                ("input_ids", dynamic_axis),
+                ("attention_mask", dynamic_axis),
+            ]
+        )