{
  "models": [
    {
      "candidate_plan": {
        "affine_prefill": false,
        "attention": false,
        "dense_mlp": true,
        "graph_fusion": false,
        "quantized_mlp": false,
        "rms_norm": false
      },
      "comparison_mode": "alternating",
      "dense_mlp_implementation": "fused",
      "features": {
        "affine_prefill": false,
        "attention": true,
        "dense_mlp": true,
        "graph_fusion": true,
        "model_autotune": true,
        "quantized_mlp": false,
        "rms_norm": true
      },
      "hardware": {
        "chip": "Apple M5",
        "machine": "arm64",
        "memory": "32 GB",
        "model_name": "Mac17,2",
        "processor": "arm",
        "release": "25.2.0",
        "system": "Darwin"
      },
      "medians": {
        "decode_speedup": 0.9988798574304839,
        "end_to_end_speedup": 0.9993774746551052,
        "metile_decode_tokens_per_second": 38.477651739119544,
        "metile_elapsed_seconds": 1.8411430419655517,
        "metile_prefill_tokens_per_second": 1555.7991060126203,
        "metile_time_to_first_token_seconds": 0.16343829198740423,
        "mlx_decode_tokens_per_second": 38.41381458225577,
        "mlx_elapsed_seconds": 1.8423976249760017,
        "mlx_prefill_tokens_per_second": 1493.9279925654016,
        "mlx_time_to_first_token_seconds": 0.16422520903870463,
        "prefill_speedup": 1.0597539284728874,
        "ttft_speedup": 1.0043642050931378
      },
      "memory": {
        "active_bytes": 4629146640,
        "cache_bytes": 9527086,
        "peak_bytes": 4825525896
      },
      "model": "mlx-community/Qwen2.5-1.5B-Instruct-bf16",
      "model_config": {
        "hidden_size": 1536,
        "model_type": "qwen2",
        "num_attention_heads": 12,
        "num_hidden_layers": 28,
        "num_key_value_heads": 2,
        "vocab_size": 151936
      },
      "plan_confirmation": {
        "accepted": true,
        "decode_speedup_floor": 0.99,
        "medians": {
          "decode_speedup": 1.0013623128071962,
          "end_to_end_speedup": 1.0115213159093093,
          "prefill_speedup": 1.0281625115952058,
          "ttft_speedup": 1.0215181659091284
        },
        "pairs": [
          {
            "decode_speedup": 1.0013623128071962,
            "end_to_end_speedup": 1.005591147433832,
            "prefill_speedup": 1.047072505618833,
            "ttft_speedup": 1.0571168576249692
          },
          {
            "decode_speedup": 1.1463423113752873,
            "end_to_end_speedup": 1.0972727768825206,
            "prefill_speedup": 0.9165644298205716,
            "ttft_speedup": 0.6967486379283915
          },
          {
            "decode_speedup": 0.990893666620941,
            "end_to_end_speedup": 0.9931801311282602,
            "prefill_speedup": 1.1616303077484793,
            "ttft_speedup": 1.0215181659091284
          },
          {
            "decode_speedup": 1.0129542436953187,
            "end_to_end_speedup": 1.0154558842138282,
            "prefill_speedup": 1.0332603892468168,
            "ttft_speedup": 1.0112972921426697
          },
          {
            "decode_speedup": 0.9981774737943376,
            "end_to_end_speedup": 0.9979751286831621,
            "prefill_speedup": 0.983207861509005,
            "ttft_speedup": 0.9976850968699864
          },
          {
            "decode_speedup": 0.9990374695334002,
            "end_to_end_speedup": 1.022765779773112,
            "prefill_speedup": 1.0270908248866966,
            "ttft_speedup": 1.2284686098354336
          },
          {
            "decode_speedup": 1.032720394898527,
            "end_to_end_speedup": 1.0408426649762867,
            "prefill_speedup": 1.0281625115952058,
            "ttft_speedup": 1.1308314086425908
          },
          {
            "decode_speedup": 0.9987979085776578,
            "end_to_end_speedup": 0.9984978171091202,
            "prefill_speedup": 1.017102603085651,
            "ttft_speedup": 0.9930984434839645
          },
          {
            "decode_speedup": 1.00609570121616,
            "end_to_end_speedup": 1.0115213159093093,
            "prefill_speedup": 1.035887904019971,
            "ttft_speedup": 1.04173955337787
          }
        ]
      },
      "precision_comparison": {
        "baseline_weights": "model_native",
        "class": "same_precision",
        "derived_from": "selected_plan",
        "native_weights_preserved": true,
        "optimized_decode_weights": [
          "model_native"
        ],
        "prefill_weights": "model_native",
        "same_weight_representation": true
      },
      "recorded_at": "2026-07-18T10:22:45.985454+00:00",
      "revision": "29d8263c520bfdf9778d0e1906c378b262a5ff63",
      "samples": {
        "MLX": [
          {
            "decode_tokens_per_second": 38.72599640615015,
            "elapsed_seconds": 1.8353860839270055,
            "prefill_tokens_per_second": 1417.7536679168925,
            "time_to_first_token_seconds": 0.1692221669945866
          },
          {
            "decode_tokens_per_second": 38.64883473148064,
            "elapsed_seconds": 1.8232028330676258,
            "prefill_tokens_per_second": 1546.408086888127,
            "time_to_first_token_seconds": 0.15362783300224692
          },
          {
            "decode_tokens_per_second": 38.24307432514238,
            "elapsed_seconds": 1.8524526249384508,
            "prefill_tokens_per_second": 1507.0688054684206,
            "time_to_first_token_seconds": 0.1650485839927569
          },
          {
            "decode_tokens_per_second": 38.30329853413263,
            "elapsed_seconds": 1.8441672089975327,
            "prefill_tokens_per_second": 1541.565049852018,
            "time_to_first_token_seconds": 0.15939150005578995
          },
          {
            "decode_tokens_per_second": 38.52833499567692,
            "elapsed_seconds": 1.8390244579641148,
            "prefill_tokens_per_second": 1545.432554472049,
            "time_to_first_token_seconds": 0.16427924996241927
          },
          {
            "decode_tokens_per_second": 38.41381458225577,
            "elapsed_seconds": 1.8423976249760017,
            "prefill_tokens_per_second": 1493.9279925654016,
            "time_to_first_token_seconds": 0.162326083984226
          },
          {
            "decode_tokens_per_second": 38.84251627037333,
            "elapsed_seconds": 1.8334587079007179,
            "prefill_tokens_per_second": 1308.9127983087565,
            "time_to_first_token_seconds": 0.17205733293667436
          },
          {
            "decode_tokens_per_second": 37.64815690975107,
            "elapsed_seconds": 1.8784270000178367,
            "prefill_tokens_per_second": 1468.0758091215923,
            "time_to_first_token_seconds": 0.16422520903870463
          },
          {
            "decode_tokens_per_second": 35.47395560328252,
            "elapsed_seconds": 1.9828076249687,
            "prefill_tokens_per_second": 1391.67684885874,
            "time_to_first_token_seconds": 0.16351520805619657
          }
        ],
        "MLX + meTile": [
          {
            "decode_tokens_per_second": 38.27599248460705,
            "elapsed_seconds": 1.8558478330960497,
            "prefill_tokens_per_second": 1545.2273129518185,
            "time_to_first_token_seconds": 0.16936537506990135
          },
          {
            "decode_tokens_per_second": 38.247113856019425,
            "elapsed_seconds": 1.8378000420052558,
            "prefill_tokens_per_second": 1551.405444778955,
            "time_to_first_token_seconds": 0.14918733295053244
          },
          {
            "decode_tokens_per_second": 38.20023662960162,
            "elapsed_seconds": 1.8536065419903025,
            "prefill_tokens_per_second": 1531.5974230264817,
            "time_to_first_token_seconds": 0.16454995807725936
          },
          {
            "decode_tokens_per_second": 38.742421997246495,
            "elapsed_seconds": 1.8252665000036359,
            "prefill_tokens_per_second": 1603.2465742103466,
            "time_to_first_token_seconds": 0.15983058302663267
          },
          {
            "decode_tokens_per_second": 38.69857986812752,
            "elapsed_seconds": 1.8321207920089364,
            "prefill_tokens_per_second": 1503.6107154598624,
            "time_to_first_token_seconds": 0.16356541693676263
          },
          {
            "decode_tokens_per_second": 37.86980287720701,
            "elapsed_seconds": 1.8621436670655385,
            "prefill_tokens_per_second": 1586.4580927458526,
            "time_to_first_token_seconds": 0.15850120899267495
          },
          {
            "decode_tokens_per_second": 38.477651739119544,
            "elapsed_seconds": 1.8411430419655517,
            "prefill_tokens_per_second": 1581.0398730731893,
            "time_to_first_token_seconds": 0.1629430839093402
          },
          {
            "decode_tokens_per_second": 38.74125232719633,
            "elapsed_seconds": 1.8287780419923365,
            "prefill_tokens_per_second": 1555.7991060126203,
            "time_to_first_token_seconds": 0.16343829198740423
          },
          {
            "decode_tokens_per_second": 38.572851715196144,
            "elapsed_seconds": 1.855348041979596,
            "prefill_tokens_per_second": 1557.247085842953,
            "time_to_first_token_seconds": 0.1819280000636354
          }
        ]
      },
      "schema_version": 8,
      "selected_dispatches": {
        "add_rms_norm": [
          {
            "algorithm": "mlx",
            "block": 0,
            "dtype": "mlx.core.bfloat16",
            "eps": 1e-06,
            "hidden": 1536,
            "row_bucket": 1
          },
          {
            "algorithm": "mlx",
            "block": 0,
            "dtype": "mlx.core.bfloat16",
            "eps": 1e-06,
            "hidden": 1536,
            "row_bucket": 128
          }
        ],
        "affine_matmul": [],
        "affine_residual_qmv": [],
        "affine_swiglu": [],
        "attention": [
          {
            "algorithm": "mlx",
            "batch": 1,
            "block": 0,
            "dimension": 128,
            "dtype": "mlx.core.bfloat16",
            "key_value_heads": 2,
            "query_heads": 12,
            "token_bucket": 256
          }
        ],
        "block_scaled": [],
        "dense_matmul": [
          {
            "algorithm": "mlx",
            "block_m": 0,
            "block_n": 0,
            "dtype": "mlx.core.bfloat16",
            "input_features": 1536,
            "k_unroll": 1,
            "outer_k": 0,
            "output_features": 8960,
            "rows": 127,
            "schedule": "",
            "wm": 0,
            "wn": 0
          },
          {
            "algorithm": "metile",
            "block_m": 128,
            "block_n": 128,
            "dtype": "mlx.core.bfloat16",
            "input_features": 1536,
            "k_unroll": 1,
            "outer_k": 128,
            "output_features": 8960,
            "rows": 128,
            "schedule": "grouped4",
            "wm": 4,
            "wn": 4
          }
        ],
        "dense_swiglu": [
          {
            "algorithm": "metile",
            "block_m": 64,
            "block_n": 64,
            "dtype": "mlx.core.bfloat16",
            "input_features": 1536,
            "k_unroll": 2,
            "output_features": 8960,
            "rows": 127,
            "schedule": "grouped8"
          },
          {
            "algorithm": "metile",
            "block_m": 64,
            "block_n": 128,
            "dtype": "mlx.core.bfloat16",
            "input_features": 1536,
            "k_unroll": 2,
            "output_features": 8960,
            "rows": 128,
            "schedule": "morton"
          }
        ],
        "rms_norm": [
          {
            "algorithm": "mlx",
            "block": 0,
            "dtype": "mlx.core.bfloat16",
            "eps": 1e-06,
            "hidden": 1536,
            "row_bucket": 1
          },
          {
            "algorithm": "mlx",
            "block": 0,
            "dtype": "mlx.core.bfloat16",
            "eps": 1e-06,
            "hidden": 1536,
            "row_bucket": 128
          }
        ]
      },
      "selected_plan": {
        "affine_prefill": false,
        "attention": false,
        "dense_mlp": true,
        "graph_fusion": false,
        "quantized_mlp": false,
        "rms_norm": false
      },
      "software": {
        "mlx": "0.32.0",
        "mlx_lm": "0.31.3",
        "python": "3.12.12"
      },
      "verification": {
        "actual_next_token": 11,
        "kl_divergence": 0.0,
        "max_logit_error": 0.0,
        "mean_logit_error": 0.0,
        "next_token": 11,
        "reference_dtype": "mlx.core.bfloat16"
      },
      "workload": {
        "confirmation_trials": 9,
        "delay_seconds": 0.1,
        "generation_tokens": 64,
        "plan_decode_steps": 8,
        "plan_trials": 7,
        "prefill_step_size": 2048,
        "prompt_tokens": 128,
        "seed": 0,
        "trials": 9
      }
    }
  ],
  "recorded_at": "2026-07-18T10:50:41Z",
  "schema_version": 3,
  "suite": "bf16-dense"
}
