.

config_rnn_measure_latency.yaml

@@ -26,9 +26,7 @@ model:
         in_channels: 20
         out_channels: 4
         model_channels: 256
-        attention_resolutions:
-        - 4
-        - 2
+        attention_resolutions: []
         num_res_blocks: 2
         channel_mult:
         - 1
@@ -41,7 +39,7 @@ model:
       target: ldm.modules.encoders.temporal_encoder.TemporalEncoder
       params:
         input_channels: 6
-        hidden_size: 4096
+        hidden_size: 2048
         num_layers: 1
         dropout: 0.1
         output_channels: 16

main.py

@@ -28,7 +28,7 @@ LATENT_DIMS = (4, SCREEN_HEIGHT // 8, SCREEN_WIDTH // 8)
 device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
 # Initialize the model at the start of your application
 #model = initialize_model("config_csllm.yaml", "yuntian-deng/computer-model")
-model = initialize_model("config_rnn.yaml", "yuntian-deng/computer-model")
+model = initialize_model("config_rnn_measure_latency.yaml", "yuntian-deng/computer-model")
 
 device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
 model = model.to(device)

utils.py

@@ -10,7 +10,7 @@ import os
 import time
 DEBUG = False
 
-def load_model_from_config(config_path, model_name, device='cuda', load=True):
+def load_model_from_config(config_path, model_name, device='cuda', load=False):
     # Load the config file
     config = OmegaConf.load(config_path)