{
 "protocol": "coreai-models llm-benchmark, 512 prompt / 1024 generation tokens, 5 trials, greedy-free synthetic prompt; Mac Studio M4 Max, macOS 27 beta (26A5416b), release build; GPU exclusive (no other GPU job)",
 "note": "run1 of the per-channel bundle was taken with COREAI_CHUNK_THRESHOLD=1 (single-token prefill), which is why its prompt tok/s is ~27; run2 is the same bundle with the default multi-token prefill.",
 "bundles": {
  "int8 per-channel absmax (minicpm5_int8sym.yaml)": {
   "main.mlirb_bytes": 2519709096,
   "run1_chunk_threshold_1": {
    "averages": {
     "generation_tps": 25.6483678133876,
     "prompt_tps": 27.128279417899755
    },
    "generation_tokens": 1024,
    "model": "minicpm5_2b_minicpm5_int8sym_dynamic",
    "num_trials": 5,
    "prompt_tokens": 512,
    "trials": [
     {
      "gen_tps": 25.65167494059603,
      "prompt_tps": 27.148753828933575
     },
     {
      "gen_tps": 25.642704887590583,
      "prompt_tps": 27.14086679101812
     },
     {
      "gen_tps": 25.646659089725173,
      "prompt_tps": 27.127529902868545
     },
     {
      "gen_tps": 25.65063120674598,
      "prompt_tps": 27.115565979982375
     },
     {
      "gen_tps": 25.650168942280242,
      "prompt_tps": 27.108680586696178
     }
    ]
   },
   "run2": {
    "averages": {
     "generation_tps": 25.55785331147675,
     "prompt_tps": 2018.465761950693
    },
    "generation_tokens": 1024,
    "model": "minicpm5_2b_minicpm5_int8sym_dynamic",
    "num_trials": 5,
    "prompt_tokens": 512,
    "trials": [
     {
      "gen_tps": 25.57442311006558,
      "prompt_tps": 2015.9560400625905
     },
     {
      "gen_tps": 25.591411025261085,
      "prompt_tps": 2020.3468420181462
     },
     {
      "gen_tps": 25.5693424633632,
      "prompt_tps": 2020.1585067657497
     },
     {
      "gen_tps": 25.53516184901101,
      "prompt_tps": 2019.5150790879973
     },
     {
      "gen_tps": 25.51892810968287,
      "prompt_tps": 2016.35234181898
     }
    ]
   }
  },
  "fp16 control (--compression none)": {
   "main.mlirb_bytes": 5034079259,
   "run": {
    "averages": {
     "generation_tps": 80.04816155587179,
     "prompt_tps": 2775.137587638996
    },
    "generation_tokens": 1024,
    "model": "minicpm5_2b_dynamic",
    "num_trials": 5,
    "prompt_tokens": 512,
    "trials": [
     {
      "gen_tps": 79.5660229554393,
      "prompt_tps": 2776.444686660322
     },
     {
      "gen_tps": 79.35758752933911,
      "prompt_tps": 2751.2533786143786
     },
     {
      "gen_tps": 80.40986419769982,
      "prompt_tps": 2788.0174191407696
     },
     {
      "gen_tps": 80.39034160128136,
      "prompt_tps": 2788.576734834587
     },
     {
      "gen_tps": 80.51699149559938,
      "prompt_tps": 2771.3957189449225
     }
    ]
   }
  },
  "int8 per-block-32 (minicpm5_int8sym_b32.yaml)": {
   "main.mlirb_bytes": 2674882267,
   "run": {
    "averages": {
     "generation_tps": 127.63468356855915,
     "prompt_tps": 2654.2086244874386
    },
    "generation_tokens": 1024,
    "model": "minicpm5_2b_minicpm5_int8sym_b32_dynamic",
    "num_trials": 5,
    "prompt_tokens": 512,
    "trials": [
     {
      "gen_tps": 127.73938269006808,
      "prompt_tps": 2568.9232705819136
     },
     {
      "gen_tps": 127.35778545081779,
      "prompt_tps": 2645.1461433944883
     },
     {
      "gen_tps": 127.53614046030137,
      "prompt_tps": 2697.5296087066813
     },
     {
      "gen_tps": 127.67577327107588,
      "prompt_tps": 2695.1399215579527
     },
     {
      "gen_tps": 127.8643359705327,
      "prompt_tps": 2664.304178196157
     }
    ]
   }
  }
 }
}