refactor(silu): add test for the fixed point silu

elasticai/creator/nn/fixed_point/precomputed/adaptable_silu_test.py

@@ -0,0 +1,40 @@
+import pytest
+import torch
+from torch.nn.functional import silu as torch_silu
+
+from elasticai.creator.nn.fixed_point import quantize as fxp_quantize
+from tests.tensor_test_case import assertTensorEqual
+
+from .adaptable_silu import AdaptableSiLU
+
+
+def fxp_args() -> dict[str, int]:
+    return dict(total_bits=16, frac_bits=8)
+
+
+def quantize(x: torch.Tensor) -> torch.Tensor:
+    return fxp_quantize(x=x, **fxp_args())
+
+
+def forge_silu(scale: float, beta: float) -> AdaptableSiLU:
+    silu = AdaptableSiLU(num_steps=5, sampling_intervall=(-10, 10), **fxp_args())
+    silu.load_state_dict(
+        {
+            "_base_module.scale": torch.tensor([scale]),
+            "_base_module.beta": torch.tensor([beta]),
+        }
+    )
+    return silu
+
+
+@pytest.mark.parametrize("scale,beta", [(1, 0), (-1, 0), (1, 1), (1, -1), (2, 3)])
+def test_silu_with_several_scale_and_beta(scale: float, beta):
+    silu = forge_silu(scale, beta)
+
+    inputs = torch.tensor([-15.0, -10.0, -5.0, 0.0, 5.0, 10.0, 15.0])
+    actual_outputs = silu(inputs)
+
+    target_inputs = torch.tensor([-10.0, -10.0, -5.0, 0.0, 5.0, 10.0, 10.0])
+    target_outputs = quantize(scale * torch_silu(target_inputs) + beta)
+
+    assertTensorEqual(target_outputs, actual_outputs)