{
  "argv": [
    "--model",
    "/workspace/wave0-model",
    "--dtype",
    "bfloat16",
    "--tensor-parallel-size",
    "1",
    "--max-model-len",
    "1024",
    "--gpu-memory-utilization",
    "0.5",
    "--enforce-eager",
    "--no-enable-prefix-caching",
    "--dataset-name",
    "random",
    "--random-input-len",
    "128",
    "--random-output-len",
    "32",
    "--num-prompts",
    "32",
    "--num-warmups",
    "8",
    "--seed",
    "0",
    "--output-json",
    "/results/ARM.json",
    "--tokenizer",
    "/workspace/wave0-model",
    "--prefix-len",
    "0",
    "--random-prefix-len",
    "0",
    "--random-range-ratio",
    "0",
    "--backend",
    "vllm",
    "--n",
    "1"
  ],
  "expected_values": {
    "model": "/workspace/wave0-model",
    "tokenizer": "/workspace/wave0-model",
    "dtype": "bfloat16",
    "tensor_parallel_size": 1,
    "max_model_len": 1024,
    "gpu_memory_utilization": 0.5,
    "enforce_eager": true,
    "enable_prefix_caching": false,
    "dataset_name": "random",
    "random_input_len": 128,
    "random_output_len": 32,
    "num_prompts": 32,
    "num_warmups": 8,
    "seed": 0,
    "prefix_len": 0,
    "random_prefix_len": 0,
    "random_range_ratio": "0",
    "backend": "vllm",
    "n": 1,
    "trust_remote_code": false,
    "quantization": null
  },
  "actual_values": {
    "model": "/workspace/wave0-model",
    "tokenizer": "/workspace/wave0-model",
    "dtype": "bfloat16",
    "tensor_parallel_size": 1,
    "max_model_len": 1024,
    "gpu_memory_utilization": 0.5,
    "enforce_eager": true,
    "enable_prefix_caching": false,
    "dataset_name": "random",
    "random_input_len": 128,
    "random_output_len": 32,
    "num_prompts": 32,
    "num_warmups": 8,
    "seed": 0,
    "prefix_len": 0,
    "random_prefix_len": 0,
    "random_range_ratio": "0",
    "backend": "vllm",
    "n": 1,
    "trust_remote_code": false,
    "quantization": null
  }
}