test started failing

ivy-llc · Sep 12, 2023 · 3da8c14 · 3da8c14
1 parent dacca59
commit 3da8c14
Show file tree

Hide file tree

Showing 4 changed files with 21 additions and 26 deletions.
diff --git a/ivy_models/regnet/layers.py b/ivy_models/regnet/layers.py
@@ -134,6 +134,7 @@ def __init__(
             kernel_size,
             stride,
             padding,
+            groups,
             norm_layer,
             activation_layer,
             dilations,
@@ -183,9 +184,9 @@ def __init__(
         scale_activation: Callable[..., ivy.Module] = ivy.Sigmoid,
     ) -> None:
         super().__init__()
-        self.avgpool = ivy.AdaptiveAvgPool2D(1)
-        self.fc1 = ivy.Conv2D(input_channels, squeeze_channels, [1, 1])
-        self.fc2 = ivy.Conv2D(squeeze_channels, input_channels, [1, 1])
+        self.avgpool = ivy.AdaptiveAvgPool2d(1)
+        self.fc1 = ivy.Conv2D(input_channels, squeeze_channels, [1, 1], 1, 0)
+        self.fc2 = ivy.Conv2D(squeeze_channels, input_channels, [1, 1], 1, 0)
         self.activation = activation()
         self.scale_activation = scale_activation()
 
@@ -296,7 +297,7 @@ def __init__(
             bottleneck_multiplier,
             se_ratio,
         )
-        self.activation = activation_layer(inplace=True)
+        self.activation = activation_layer()  # inplace=True
 
     def _forward(self, x: ivy.Array) -> ivy.Array:
         if self.proj is not None:
@@ -336,7 +337,7 @@ def __init__(
                 bottleneck_multiplier,
                 se_ratio,
             )
-            self._submodules = list(block)
+            self._submodules = block
 
 
 def _make_divisible(v: float, divisor: int, min_value: Optional[int] = None) -> int:

diff --git a/ivy_models/regnet/regnet.py b/ivy_models/regnet/regnet.py
@@ -1,5 +1,4 @@
 # global
-import math
 from collections import OrderedDict
 from typing import Callable, Optional, Type
 import builtins
@@ -138,23 +137,8 @@ def _build(self, *args, **kwargs):
 
         self.trunk_output = ivy.Sequential(OrderedDict(blocks))
 
-        self.avgpool = ivy.AdaptiveAvgPool2D((1, 1))
-        self.fc = ivy.Linear(
-            in_featuReg=current_width, out_featuReg=self.spec.num_classes
-        )
-
-        # Performs RegNet-style weight initialization
-        for m in self.modules():
-            if isinstance(m, ivy.Conv2D):
-                # Note that there is no bias due to BN
-                fan_out = m._filter_shape[0] * m._filter_shape[1] * m._output_channels
-                ivy.init.normal_(m.weight, mean=0.0, std=math.sqrt(2.0 / fan_out))
-            elif isinstance(m, ivy.BatchNorm2D):
-                ivy.init.ones_(m.weight)
-                ivy.init.zeros_(m.bias)
-            elif isinstance(m, ivy.Linear):
-                ivy.init.normal_(m.weight, mean=0.0, std=0.01)
-                ivy.init.zeros_(m.bias)
+        self.avgpool = ivy.AdaptiveAvgPool2d((1, 1))
+        self.fc = ivy.Linear(current_width, self.spec.num_classes)
 
     def _forward(self, x):
         x = self.stem(x)

diff --git a/ivy_models/regnet/regnet_variant_helper.py b/ivy_models/regnet/regnet_variant_helper.py
@@ -23,7 +23,7 @@
 torch_img = torch.unsqueeze(torch_img, 0)
 
 
-from torchvision.models import *
+from torchvision.models import RegNet_X_16GF_Weights, RegNet_X_1_6GF_Weights, RegNet_X_32GF_Weights, RegNet_X_3_2GF_Weights, RegNet_X_400MF_Weights, RegNet_X_800MF_Weights, RegNet_X_8GF_Weights, RegNet_Y_16GF_Weights, RegNet_Y_1_6GF_Weights, RegNet_Y_32GF_Weights, RegNet_Y_3_2GF_Weights, RegNet_Y_400MF_Weights, RegNet_Y_800MF_Weights, RegNet_Y_8GF_Weights, regnet_x_16gf, regnet_x_1_6gf, regnet_x_32gf, regnet_x_3_2gf, regnet_x_400mf, regnet_x_800mf, regnet_x_8gf, regnet_y_16gf, regnet_y_1_6gf, regnet_y_32gf, regnet_y_3_2gf, regnet_y_400mf, regnet_y_800mf, regnet_y_8gf
 
 VARIANTS = {
     RegNet_Y_400MF_Weights: regnet_y_400mf,
@@ -43,7 +43,6 @@
     RegNet_X_32GF_Weights: regnet_x_32gf,
 }
 
-import numpy as np
 
 variant_code_list = [
     "y_400mf",
@@ -74,7 +73,7 @@
     torch_logits = torch.take(torch_output[0], torch_classes)
 
     with open("output_4.txt", "a") as file:
-        file.write(f" {variant_code_list[variant_count]} : np.array({torch_logits}\n")
+        file.write(f" {variant_code_list[variant_count]} : np.array({torch_classes}\n")
         variant_count += 1
 
 # print("Indices of the top 3 classes are:", torch_classes)

diff --git a/output_4.txt b/output_4.txt
@@ -12,3 +12,14 @@ x_3_2gf400 : np.array([0.8212, 0.1721, 0.0055])
 x_8gf : np.array([0.8271, 0.1659, 0.0063])
 x_16gf : np.array([0.7613, 0.2340, 0.0043])
 x_32gf : np.array([0.7828, 0.2146, 0.0022])
+ y_400mf : np.array(tensor([282, 281, 285], device='cuda:0')
+ y_800mf : np.array(tensor([282, 281, 285], device='cuda:0')
+ y_1_6gf : np.array(tensor([282, 281, 285], device='cuda:0')
+ y_3_2gf : np.array(tensor([282, 281, 285], device='cuda:0')
+ y_8gf : np.array(tensor([282, 281, 285], device='cuda:0')
+ y_16gf : np.array(tensor([282, 281, 285], device='cuda:0')
+ y_32gf : np.array(tensor([282, 281, 285], device='cuda:0')
+ x_400mf : np.array(tensor([282, 281, 285], device='cuda:0')
+ x_800mf : np.array(tensor([282, 281, 285], device='cuda:0')
+ x_1_6gf : np.array(tensor([282, 281, 285], device='cuda:0')
+ x_3_2gf400 : np.array(tensor([282, 281, 285], device='cuda:0')