build(logs): Print GPU memory information

modules/inference.py

@@ -8,6 +8,10 @@ from io import BytesIO
 import base64
 
 print(f"GPUs available: {torch.cuda.device_count()}")
+print(f"GPU[0] memory: {int(torch.cuda.get_device_properties(0).total_memory / 1048576)}Mib")
+print(f"GPU[0] memory reserved: {int(torch.cuda.memory_reserved(0) / 1048576)}Mib")
+print(f"GPU[0] memory allocated: {int(torch.cuda.memory_allocated(0) / 1048576)}Mib")
+
 device = "cuda" if torch.cuda.is_available() else "cpu"
 fp16 = torch.cuda.is_available()