import torch
 
vanilla_res = generate_helper(pipe)
 
print(f"Latency: {vanilla_res['latency']}")
print(f"GPU memory: {torch.cuda.memory_allocated() / 1024**3:.2f} GB")
print(f"Generated Instruction: {vanilla_res['text']}")
 
# Latency: 37.49ms/token
# GPU memory: 12.62 GB
# Generated Instruction: Write a request for PTO letter to my boss