evaluation/swift15-baseline/speed-c1.json

{
  "concurrency": 1,
  "requests": 8,
  "output_tokens_per_request": 512,
  "wall_seconds": 41.13470574293751,
  "aggregate_output_tps": 99.57528383932221,
  "median_decode_tps": 103.95189430818986,
  "median_ttft_seconds": 0.0777347715338692,
  "sampling": "greedy; ignore_eos for this throughput test only",
  "speculation_counter_deltas": {
    "vllm:spec_decode_num_drafts_total": 1362.0,
    "vllm:spec_decode_num_draft_tokens_total": 4086.0,
    "vllm:spec_decode_num_accepted_tokens_total": 2740.0
  },
  "decode_tps_note": "Client stream timing estimate; speculative decoding may deliver several tokens in one event."
}