adjusted level 3 sizes

KernelBench/level3/18_SqueezeNet.py

@@ -72,7 +72,7 @@ def forward(self, x):
         return torch.flatten(x, 1)
 
 # Test code
-batch_size = 64
+batch_size = 32
 input_channels = 3
 height = 512
 width = 512

KernelBench/level3/2_ShallowWideMLP.py

@@ -31,7 +31,7 @@ def forward(self, x):
         return self.network(x)
 
 # Test code
-batch_size = 128
+batch_size = 64
 input_size = 16384
 hidden_layer_sizes = [32768, 32768]
 output_size = 16384

KernelBench/level3/35_LSTM.py

@@ -41,10 +41,10 @@ def forward(self, x, h0=None, c0=None):
         return out
 
 # === Test configuration ===
-batch_size = 10
+batch_size = 3
 sequence_length = 512
-input_size = 128
-hidden_size = 256
+input_size = 64
+hidden_size = 128
 num_layers = 6
 output_size = 10
 dropout = 0.0

KernelBench/level3/36_LSTMHn.py

@@ -34,10 +34,10 @@ def forward(self, x,h0,c0):
         return state[0]
 
 # Test code
-batch_size = 10
+batch_size = 3
 sequence_length = 512
-input_size = 128
-hidden_size = 256
+input_size = 64
+hidden_size = 128
 num_layers = 6
 output_size = 10
 dropout = 0.0

KernelBench/level3/37_LSTMCn.py

@@ -34,10 +34,10 @@ def forward(self, x, h0, c0):
         return state[1]
 
 # Test code
-batch_size = 10
+batch_size = 3
 sequence_length = 512
-input_size = 128
-hidden_size = 256
+input_size = 64
+hidden_size = 128
 num_layers = 6
 output_size = 10
 dropout = 0.0

KernelBench/level3/38_LSTMBidirectional.py

@@ -33,10 +33,10 @@ def forward(self, x,h0,c0):
         return out
 
 # Test code
-batch_size = 10
+batch_size = 3
 sequence_length = 512
-input_size = 128
-hidden_size = 256
+input_size = 64
+hidden_size = 128
 num_layers = 6
 output_size = 10
 dropout = 0.0

KernelBench/level3/39_GRU.py

@@ -27,10 +27,10 @@ def forward(self, x,h0):
         return output
 
 # Test code
-batch_size = 10
+batch_size = 3
 seq_len = 512
-input_size = 128
-hidden_size = 256
+input_size = 64
+hidden_size = 128
 num_layers = 6
 
 def get_inputs():

KernelBench/level3/40_GRUHidden.py

@@ -27,10 +27,10 @@ def forward(self, x,h0):
         return h_n
 
 # Test code
-batch_size = 10
+batch_size = 3
 seq_len = 512
-input_size = 128
-hidden_size = 256
+input_size = 64
+hidden_size = 128
 num_layers = 6
 
 def get_inputs():

KernelBench/level3/41_GRUBidirectional.py

@@ -28,10 +28,10 @@ def forward(self, x,h0):
         return output
 
 # Test code
-batch_size = 10
+batch_size = 3
 seq_len = 512
-input_size = 128
-hidden_size = 256
+input_size = 64
+hidden_size = 128
 num_layers = 6
 
 def get_inputs():

KernelBench/level3/42_GRUBidirectionalHidden.py

@@ -27,10 +27,10 @@ def forward(self, x,h0):
         return h_n
 
 # Test code
-batch_size = 10
+batch_size = 3
 seq_len = 512
-input_size = 128
-hidden_size = 256
+input_size = 64
+hidden_size = 128
 num_layers = 6
 
 def get_inputs():

KernelBench/level3/44_MiniGPTBlock.py

@@ -82,7 +82,7 @@ def forward(self, x):
         x = x + self.mlpf(self.ln_2(x))
         return x
 
-batch_size = 128
+batch_size = 32
 max_seqlen = 1024
 seq_len = 512
 n_embd = 768

KernelBench/level3/5_AlexNet.py

@@ -81,7 +81,7 @@ def forward(self, x):
         return x
 
 # Test code
-batch_size = 1024
+batch_size = 512
 num_classes = 1000
 
 def get_inputs():