Add a Python script that runs LLM benchmark

[nv-hwoo]

Running benchmark 1: Prefill phase

python profile.py --prompt-size-range 100 1000 200 --max-tokens 1

# Sample output
# [ Benchmark Summary ]
#   Prompt size: 100, Average first-token latency: 0.0421 sec
#   Prompt size: 300, Average first-token latency: 0.0312 sec
#   Prompt size: 500, Average first-token latency: 0.0289 sec
#   Prompt size: 700, Average first-token latency: 0.0358 sec
#   Prompt size: 900, Average first-token latency: 0.0327 sec

Running benchmark 2: Generation phase

python profile.py --prompt-size-range 100 1000 200 --max-tokens 256 --ignore-eos

# Sample output
# [ Benchmark Summary ]
#   Prompt size: 100, Average first-token latency: 0.0419 sec, Average token-token latency: 0.0068 sec
#   Prompt size: 300, Average first-token latency: 0.0513 sec, Average token-token latency: 0.0070 sec
#   Prompt size: 500, Average first-token latency: 0.0325 sec, Average token-token latency: 0.0069 sec
#   Prompt size: 700, Average first-token latency: 0.0325 sec, Average token-token latency: 0.0071 sec
#   Prompt size: 900, Average first-token latency: 0.0368 sec, Average token-token latency: 0.0071 sec