An open-source document parsing model with SOTA accuracy and speed, using an RSI loop during post-training.
from pathlib import Path
import torch
from dfine_la.config import Settings
import dfine_la.runtime as runtime_module
class FakeModel:
def __call__(self, images, sizes):
batch_size = images.shape[0]
return (
torch.zeros((batch_size, 1), dtype=torch.int64, device=images.device),
torch.zeros((batch_size, 1, 4), device=images.device),
torch.ones((batch_size, 1), device=images.device),
)
def test_cpu_runtime_uses_fp32_eager_inference(monkeypatch):
monkeypatch.setattr(Settings, "validate_checkpoint", lambda self: self)
monkeypatch.setattr(runtime_module, "load_model", lambda *args, **kwargs: FakeModel())
settings = Settings(
model_path=Path("unused.pth"),
device="cpu",
cuda_graph_enabled=True,
)
runtime = runtime_module.ModelRuntime(settings)
outputs, _ = runtime.infer(
(torch.zeros((1, 3, 8, 8)), torch.tensor([[8, 8]]))
)
assert runtime.device == torch.device("cpu")
assert runtime.cuda_graph_enabled is False
assert runtime.health()["inference_precision"] == "fp32"
assert "model_path" not in runtime.health()
assert "model_path" not in runtime.info()["model_config"]
assert "metadata_path" not in runtime.info()["model_config"]
assert runtime.info()["devices"] == ["cpu"]
assert runtime.info()["model_config"]["device"] == "cpu"
assert runtime.info()["model_config"]["cuda_graph_enabled"] is False
assert outputs[0].shape == (1, 1)
def test_cuda_mode_still_requires_cuda(monkeypatch):
monkeypatch.setattr(Settings, "validate_checkpoint", lambda self: self)
monkeypatch.setattr(torch.cuda, "is_available", lambda: False)
settings = Settings(model_path=Path("unused.pth"), device="cuda")
try:
runtime_module.ModelRuntime(settings)
except RuntimeError as exc:
assert "NVIDIA GPU" in str(exc)
else:
raise AssertionError("CUDA mode must fail when CUDA is unavailable")