{
  "prompt_count": 1,
  "generated_tokens": 64,
  "reference": "Stored vLLM greedy reference",
  "matching_tokens": 64,
  "rows": 8,
  "chunk": 512,
  "prefill_mode": "eager",
  "decode_mode": "CUDA Graph"
}