An open-source document parsing model with SOTA accuracy and speed, using an RSI loop during post-training.
import numpy as np
from fastapi.testclient import TestClient
from dfine_la.api import create_app
class FakeRuntime:
def __init__(self, settings):
self.settings = settings
def prepare(self, payloads):
return payloads, 1.0, 2.0
def infer(self, ready):
count = len(ready)
outputs = (
np.tile(np.array([[0, 1, 1, 2]], dtype=np.int64), (count, 1)),
np.tile(
np.array(
[[
[0.0, 0.0, 1.0, 1.0],
[1.0, 1.0, 2.0, 2.0],
[2.0, 2.0, 3.0, 3.0],
[3.0, 3.0, 4.0, 4.0],
]],
dtype=np.float32,
),
(count, 1, 1),
),
np.tile(
np.array([[0.4, 0.6, 0.8, 0.9]], dtype=np.float32),
(count, 1),
),
)
return outputs, {"queue": 0.0, "inference": 3.0}
def health(self):
return {"status": "healthy", "cuda_graph_ready": True}
def info(self):
return {"endpoint": "/inference"}
def test_api_contract_and_no_versioned_route():
app = create_app(runtime_factory=FakeRuntime)
with TestClient(app) as client:
response = client.post(
"/inference",
files=[("images", ("image.png", b"image", "image/png"))],
)
assert response.status_code == 200
assert response.json() == {
"status": "success",
"params": {},
"data": {
"labels": [[0, 1, 1, 2]],
"boxes": [[
[0.0, 0.0, 1.0, 1.0],
[1.0, 1.0, 2.0, 2.0],
[2.0, 2.0, 3.0, 3.0],
[3.0, 3.0, 4.0, 4.0],
]],
"scores": [[
0.4000000059604645,
0.6000000238418579,
0.800000011920929,
0.8999999761581421,
]],
},
}
assert "request_read" in response.headers["server-timing"]
assert client.get("/health").json()["status"] == "healthy"
assert client.get("/model/info").json()["endpoint"] == "/inference"
assert client.post("/legacy-inference").status_code == 404
assert client.get("/docs").status_code == 404
assert client.get("/redoc").status_code == 404
assert client.get("/openapi.json").status_code == 404
def test_missing_images_returns_422():
app = create_app(runtime_factory=FakeRuntime)
with TestClient(app) as client:
assert client.post("/inference").status_code == 422
def test_global_confidence_filters_all_classes_and_keeps_fields_aligned():
app = create_app(runtime_factory=FakeRuntime)
with TestClient(app) as client:
response = client.post(
"/inference",
files=[("images", ("image.png", b"image", "image/png"))],
data={"confidence": "0.7"},
)
assert response.status_code == 200
body = response.json()
assert body["params"] == {"confidence": 0.7, "class_confidences": {}}
assert body["data"]["labels"] == [[1, 2]]
assert body["data"]["boxes"] == [[
[2.0, 2.0, 3.0, 3.0],
[3.0, 3.0, 4.0, 4.0],
]]
assert len(body["data"]["scores"][0]) == 2
def test_class_confidences_override_global_confidence():
app = create_app(runtime_factory=FakeRuntime)
with TestClient(app) as client:
response = client.post(
"/inference",
files=[("images", ("image.png", b"image", "image/png"))],
data={
"confidence": "0.7",
"class_confidences": '{"1": 0.5, "2": 0.95}',
},
)
assert response.status_code == 200
body = response.json()
assert body["params"] == {
"confidence": 0.7,
"class_confidences": {"1": 0.5, "2": 0.95},
}
assert body["data"]["labels"] == [[1, 1]]
assert body["data"]["boxes"] == [[
[1.0, 1.0, 2.0, 2.0],
[2.0, 2.0, 3.0, 3.0],
]]
assert len(body["data"]["scores"][0]) == 2
def test_invalid_confidence_parameters_return_422_before_inference():
app = create_app(runtime_factory=FakeRuntime)
invalid_data = [
{"confidence": "-0.1"},
{"confidence": "nan"},
{"class_confidences": "[]"},
{"class_confidences": '{"12": 0.5}'},
{"class_confidences": '{"1": 1.1}'},
]
with TestClient(app) as client:
for data in invalid_data:
response = client.post(
"/inference",
files=[("images", ("image.png", b"image", "image/png"))],
data=data,
)
assert response.status_code == 422, data