{
  "aggregate": {
    "attention": {
      "runtime_seconds": {
        "mean": 65.55628659700353,
        "sample_std": 0.7277790091819678
      },
      "test_loss": {
        "mean": 3.4450017677025278,
        "sample_std": 0.01050198902379009
      },
      "test_perplexity": {
        "mean": 31.344489850437487,
        "sample_std": 0.3286103218921654
      }
    },
    "mlp": {
      "runtime_seconds": {
        "mean": 58.228393291666485,
        "sample_std": 0.13956459771163404
      },
      "test_loss": {
        "mean": 3.94332637286606,
        "sample_std": 0.007706849256245094
      },
      "test_perplexity": {
        "mean": 51.59094350068963,
        "sample_std": 0.39676399285471264
      }
    },
    "multihead": {
      "runtime_seconds": {
        "mean": 69.72688850023162,
        "sample_std": 1.268775745451773
      },
      "test_loss": {
        "mean": 3.359689575099665,
        "sample_std": 0.021187387467114294
      },
      "test_perplexity": {
        "mean": 28.784579416639392,
        "sample_std": 0.613505962050175
      }
    }
  },
  "audit": {
    "document_overlap": {
      "train_test": 0,
      "train_validation": 0,
      "validation_test": 0
    },
    "documents": {
      "test": 592,
      "train": 4822,
      "validation": 586
    },
    "duplicates_removed_before_split": 0,
    "oov": {
      "test": {
        "rate": 0.010776203871420104,
        "tokens": 119801,
        "unknown": 1291
      },
      "train": {
        "rate": 0.007354267901918207,
        "tokens": 964338,
        "unknown": 7092
      },
      "validation": {
        "rate": 0.011861176588595215,
        "tokens": 119634,
        "unknown": 1419
      }
    },
    "split_seed": 31415,
    "story_length_tokens": {
      "test": {
        "max": 768,
        "median": 178.0,
        "min": 78
      },
      "train": {
        "max": 942,
        "median": 176.0,
        "min": 60
      },
      "validation": {
        "max": 947,
        "median": 177.0,
        "min": 83
      }
    },
    "vocab_fit_on": "train only",
    "vocab_size": 4000
  },
  "created_utc": "2026-09-22T05:15:13.701154+00:00",
  "environment": {
    "accelerator": "Apple Metal (MPS)",
    "cuda_available": false,
    "cuda_version": null,
    "mps_available": true,
    "platform": "Darwin arm64",
    "python": "3.11.11",
    "torch": "2.14.0"
  },
  "machine": "arm",
  "protocol": {
    "batch_size": 512,
    "context_length": 64,
    "embedding_width": 64,
    "heads": {
      "attention": 1,
      "mlp": 0,
      "multihead": 4
    },
    "hidden_width": 256,
    "hyperparameters": "Frozen baseline settings; four heads reuses one-head settings. No test tuning.",
    "positional_encoding": "learned absolute position rows, added to token rows",
    "raw_sha256": "ecaf3b99879170311c1fcde645251b9793620f1f5bf57159210fb7ab7180a7b3",
    "seeds": [
      11,
      29,
      47
    ],
    "source_hashes": {
      "multihead.py": "84ee11be62956cd035c0f0c341c4ccd0affb8d36df83ab74bfb68e7f398b67b8",
      "wordlm.py": "eb39f53aa7a89f5002b728d04f298e9f2f05594287f048a2bb227413469e1fae"
    },
    "steps": 6000,
    "train_config_attention": {
      "context_len": 64,
      "d_embed": 64,
      "hidden": 256,
      "learning_rate": 0.003,
      "sweep_axis": "comparison_tune",
      "sweep_value": "lr=0.003,wd=0.001",
      "weight_decay": 0.001
    },
    "train_config_mlp": {
      "context_len": 64,
      "d_embed": 64,
      "hidden": 256,
      "learning_rate": 0.001,
      "sweep_axis": "comparison_tune",
      "sweep_value": "lr=0.001,wd=0.0001",
      "weight_decay": 0.0001
    }
  },
  "runs": {
    "attention": [
      {
        "checkpoint_selection": "lowest validation loss on the fixed selection slice",
        "device": "mps",
        "final_train_batch_loss": 3.0844616889953613,
        "parameter_count": 1321120,
        "runtime_seconds": 65.76938187493943,
        "seed": 11,
        "selected_step": 6000,
        "selected_tokens_seen": 3072000,
        "selection_slice_validation_loss": 3.423395104980469,
        "test_loss": 3.4540876047124054,
        "test_perplexity": 31.629416982072325,
        "tokens_seen": 3072000,
        "torch_version": "2.14.0",
        "trace": [
          {
            "step": 1,
            "tokens_seen": 512,
            "train_batch_loss": 8.294988632202148,
            "validation_loss": 7.92612978515625
          },
          {
            "step": 750,
            "tokens_seen": 384000,
            "train_batch_loss": 3.7250728607177734,
            "validation_loss": 3.871896105957031
          },
          {
            "step": 1500,
            "tokens_seen": 768000,
            "train_batch_loss": 3.5651793479919434,
            "validation_loss": 3.656668359375
          },
          {
            "step": 2250,
            "tokens_seen": 1152000,
            "train_batch_loss": 3.4835047721862793,
            "validation_loss": 3.574521746826172
          },
          {
            "step": 3000,
            "tokens_seen": 1536000,
            "train_batch_loss": 3.272444248199463,
            "validation_loss": 3.5097500732421874
          },
          {
            "step": 3750,
            "tokens_seen": 1920000,
            "train_batch_loss": 2.9541540145874023,
            "validation_loss": 3.4712187438964843
          },
          {
            "step": 4500,
            "tokens_seen": 2304000,
            "train_batch_loss": 2.9143686294555664,
            "validation_loss": 3.4434575073242186
          },
          {
            "step": 5250,
            "tokens_seen": 2688000,
            "train_batch_loss": 3.1337575912475586,
            "validation_loss": 3.4490286010742186
          },
          {
            "step": 6000,
            "tokens_seen": 3072000,
            "train_batch_loss": 3.0844616889953613,
            "validation_loss": 3.423395104980469
          }
        ],
        "train_config": {
          "batch_size": 512,
          "eval_every": 750,
          "eval_examples": 20000,
          "gradient_clip": 1.0,
          "learning_rate": 0.003,
          "seed": 11,
          "steps": 6000,
          "weight_decay": 0.001
        },
        "validation_loss": 3.399205933734118,
        "validation_perplexity": 29.94031601066221
      },
      {
        "checkpoint_selection": "lowest validation loss on the fixed selection slice",
        "device": "mps",
        "final_train_batch_loss": 2.9213719367980957,
        "parameter_count": 1321120,
        "runtime_seconds": 66.1537312080618,
        "seed": 29,
        "selected_step": 5250,
        "selected_tokens_seen": 2688000,
        "selection_slice_validation_loss": 3.4351538696289063,
        "test_loss": 3.4474142390377245,
        "test_perplexity": 31.419045041638512,
        "tokens_seen": 3072000,
        "torch_version": "2.14.0",
        "trace": [
          {
            "step": 1,
            "tokens_seen": 512,
            "train_batch_loss": 8.367996215820312,
            "validation_loss": 7.9838879150390625
          },
          {
            "step": 750,
            "tokens_seen": 384000,
            "train_batch_loss": 3.865884304046631,
            "validation_loss": 3.9115534912109373
          },
          {
            "step": 1500,
            "tokens_seen": 768000,
            "train_batch_loss": 3.3901782035827637,
            "validation_loss": 3.6806303955078126
          },
          {
            "step": 2250,
            "tokens_seen": 1152000,
            "train_batch_loss": 3.460838794708252,
            "validation_loss": 3.57853583984375
          },
          {
            "step": 3000,
            "tokens_seen": 1536000,
            "train_batch_loss": 3.226238250732422,
            "validation_loss": 3.5138580932617187
          },
          {
            "step": 3750,
            "tokens_seen": 1920000,
            "train_batch_loss": 3.313788890838623,
            "validation_loss": 3.4662207763671873
          },
          {
            "step": 4500,
            "tokens_seen": 2304000,
            "train_batch_loss": 3.3939013481140137,
            "validation_loss": 3.4426947448730467
          },
          {
            "step": 5250,
            "tokens_seen": 2688000,
            "train_batch_loss": 3.133535861968994,
            "validation_loss": 3.4351538696289063
          },
          {
            "step": 6000,
            "tokens_seen": 3072000,
            "train_batch_loss": 2.9213719367980957,
            "validation_loss": 3.4418499877929687
          }
        ],
        "train_config": {
          "batch_size": 512,
          "eval_every": 750,
          "eval_examples": 20000,
          "gradient_clip": 1.0,
          "learning_rate": 0.003,
          "seed": 29,
          "steps": 6000,
          "weight_decay": 0.001
        },
        "validation_loss": 3.401393623206088,
        "validation_perplexity": 30.005887824016927
      },
      {
        "checkpoint_selection": "lowest validation loss on the fixed selection slice",
        "device": "mps",
        "final_train_batch_loss": 2.8401310443878174,
        "parameter_count": 1321120,
        "runtime_seconds": 64.74574670800939,
        "seed": 47,
        "selected_step": 5250,
        "selected_tokens_seen": 2688000,
        "selection_slice_validation_loss": 3.4308641845703125,
        "test_loss": 3.4335034593574534,
        "test_perplexity": 30.985007527601624,
        "tokens_seen": 3072000,
        "torch_version": "2.14.0",
        "trace": [
          {
            "step": 1,
            "tokens_seen": 512,
            "train_batch_loss": 8.337409973144531,
            "validation_loss": 7.974603271484375
          },
          {
            "step": 750,
            "tokens_seen": 384000,
            "train_batch_loss": 3.8104400634765625,
            "validation_loss": 3.8579754638671875
          },
          {
            "step": 1500,
            "tokens_seen": 768000,
            "train_batch_loss": 3.288633346557617,
            "validation_loss": 3.6484563354492185
          },
          {
            "step": 2250,
            "tokens_seen": 1152000,
            "train_batch_loss": 3.4033164978027344,
            "validation_loss": 3.549428729248047
          },
          {
            "step": 3000,
            "tokens_seen": 1536000,
            "train_batch_loss": 3.191244602203369,
            "validation_loss": 3.4903079467773437
          },
          {
            "step": 3750,
            "tokens_seen": 1920000,
            "train_batch_loss": 3.053774356842041,
            "validation_loss": 3.4630304809570314
          },
          {
            "step": 4500,
            "tokens_seen": 2304000,
            "train_batch_loss": 3.1757707595825195,
            "validation_loss": 3.4355752746582033
          },
          {
            "step": 5250,
            "tokens_seen": 2688000,
            "train_batch_loss": 2.9921493530273438,
            "validation_loss": 3.4308641845703125
          },
          {
            "step": 6000,
            "tokens_seen": 3072000,
            "train_batch_loss": 2.8401310443878174,
            "validation_loss": 3.444860314941406
          }
        ],
        "train_config": {
          "batch_size": 512,
          "eval_every": 750,
          "eval_examples": 20000,
          "gradient_clip": 1.0,
          "learning_rate": 0.003,
          "seed": 47,
          "steps": 6000,
          "weight_decay": 0.001
        },
        "validation_loss": 3.390533275664537,
        "validation_perplexity": 29.681776619775164
      }
    ],
    "mlp": [
      {
        "checkpoint_selection": "lowest validation loss on the fixed selection slice",
        "device": "mps",
        "final_train_batch_loss": 2.632467746734619,
        "parameter_count": 2332832,
        "runtime_seconds": 58.35629449994303,
        "seed": 11,
        "selected_step": 3000,
        "selected_tokens_seen": 1536000,
        "selection_slice_validation_loss": 3.9447265869140624,
        "test_loss": 3.9486035934138615,
        "test_perplexity": 51.86289455868806,
        "tokens_seen": 3072000,
        "torch_version": "2.14.0",
        "trace": [
          {
            "step": 1,
            "tokens_seen": 512,
            "train_batch_loss": 8.35482406616211,
            "validation_loss": 8.233119262695313
          },
          {
            "step": 750,
            "tokens_seen": 384000,
            "train_batch_loss": 3.884162425994873,
            "validation_loss": 4.189762426757812
          },
          {
            "step": 1500,
            "tokens_seen": 768000,
            "train_batch_loss": 3.6303305625915527,
            "validation_loss": 3.9842804443359374
          },
          {
            "step": 2250,
            "tokens_seen": 1152000,
            "train_batch_loss": 3.426093101501465,
            "validation_loss": 3.9448476318359376
          },
          {
            "step": 3000,
            "tokens_seen": 1536000,
            "train_batch_loss": 3.0746166706085205,
            "validation_loss": 3.9447265869140624
          },
          {
            "step": 3750,
            "tokens_seen": 1920000,
            "train_batch_loss": 2.782228946685791,
            "validation_loss": 3.997566418457031
          },
          {
            "step": 4500,
            "tokens_seen": 2304000,
            "train_batch_loss": 2.7576375007629395,
            "validation_loss": 4.028945886230469
          },
          {
            "step": 5250,
            "tokens_seen": 2688000,
            "train_batch_loss": 2.854602336883545,
            "validation_loss": 4.08432958984375
          },
          {
            "step": 6000,
            "tokens_seen": 3072000,
            "train_batch_loss": 2.632467746734619,
            "validation_loss": 4.163871276855469
          }
        ],
        "train_config": {
          "batch_size": 512,
          "eval_every": 750,
          "eval_examples": 20000,
          "gradient_clip": 1.0,
          "learning_rate": 0.001,
          "seed": 11,
          "steps": 6000,
          "weight_decay": 0.0001
        },
        "validation_loss": 3.900230741524653,
        "validation_perplexity": 49.41384961719351
      },
      {
        "checkpoint_selection": "lowest validation loss on the fixed selection slice",
        "device": "mps",
        "final_train_batch_loss": 2.561657428741455,
        "parameter_count": 2332832,
        "runtime_seconds": 58.0795354580041,
        "seed": 29,
        "selected_step": 2250,
        "selected_tokens_seen": 1152000,
        "selection_slice_validation_loss": 3.941741552734375,
        "test_loss": 3.946893303094475,
        "test_perplexity": 51.77426976085886,
        "tokens_seen": 3072000,
        "torch_version": "2.14.0",
        "trace": [
          {
            "step": 1,
            "tokens_seen": 512,
            "train_batch_loss": 8.309721946716309,
            "validation_loss": 8.20043037109375
          },
          {
            "step": 750,
            "tokens_seen": 384000,
            "train_batch_loss": 4.011899948120117,
            "validation_loss": 4.1979369140625
          },
          {
            "step": 1500,
            "tokens_seen": 768000,
            "train_batch_loss": 3.4435975551605225,
            "validation_loss": 3.9914958862304686
          },
          {
            "step": 2250,
            "tokens_seen": 1152000,
            "train_batch_loss": 3.25858736038208,
            "validation_loss": 3.941741552734375
          },
          {
            "step": 3000,
            "tokens_seen": 1536000,
            "train_batch_loss": 3.1132264137268066,
            "validation_loss": 3.946730895996094
          },
          {
            "step": 3750,
            "tokens_seen": 1920000,
            "train_batch_loss": 2.999875068664551,
            "validation_loss": 3.9701077392578124
          },
          {
            "step": 4500,
            "tokens_seen": 2304000,
            "train_batch_loss": 3.163775682449341,
            "validation_loss": 4.023033361816406
          },
          {
            "step": 5250,
            "tokens_seen": 2688000,
            "train_batch_loss": 2.732346773147583,
            "validation_loss": 4.065170947265625
          },
          {
            "step": 6000,
            "tokens_seen": 3072000,
            "train_batch_loss": 2.561657428741455,
            "validation_loss": 4.13652763671875
          }
        ],
        "train_config": {
          "batch_size": 512,
          "eval_every": 750,
          "eval_examples": 20000,
          "gradient_clip": 1.0,
          "learning_rate": 0.001,
          "seed": 29,
          "steps": 6000,
          "weight_decay": 0.0001
        },
        "validation_loss": 3.912883281691898,
        "validation_perplexity": 50.04303232037557
      },
      {
        "checkpoint_selection": "lowest validation loss on the fixed selection slice",
        "device": "mps",
        "final_train_batch_loss": 2.496955633163452,
        "parameter_count": 2332832,
        "runtime_seconds": 58.249349917052314,
        "seed": 47,
        "selected_step": 2250,
        "selected_tokens_seen": 1152000,
        "selection_slice_validation_loss": 3.9290302490234374,
        "test_loss": 3.9344822220898434,
        "test_perplexity": 51.13566618252198,
        "tokens_seen": 3072000,
        "torch_version": "2.14.0",
        "trace": [
          {
            "step": 1,
            "tokens_seen": 512,
            "train_batch_loss": 8.293191909790039,
            "validation_loss": 8.19183671875
          },
          {
            "step": 750,
            "tokens_seen": 384000,
            "train_batch_loss": 4.015819072723389,
            "validation_loss": 4.2038025390625
          },
          {
            "step": 1500,
            "tokens_seen": 768000,
            "train_batch_loss": 3.291044235229492,
            "validation_loss": 3.9982716918945314
          },
          {
            "step": 2250,
            "tokens_seen": 1152000,
            "train_batch_loss": 3.335709571838379,
            "validation_loss": 3.9290302490234374
          },
          {
            "step": 3000,
            "tokens_seen": 1536000,
            "train_batch_loss": 3.046940803527832,
            "validation_loss": 3.9549423095703125
          },
          {
            "step": 3750,
            "tokens_seen": 1920000,
            "train_batch_loss": 2.7661890983581543,
            "validation_loss": 3.9901332275390624
          },
          {
            "step": 4500,
            "tokens_seen": 2304000,
            "train_batch_loss": 2.753131866455078,
            "validation_loss": 4.040422839355469
          },
          {
            "step": 5250,
            "tokens_seen": 2688000,
            "train_batch_loss": 2.624964714050293,
            "validation_loss": 4.077186401367188
          },
          {
            "step": 6000,
            "tokens_seen": 3072000,
            "train_batch_loss": 2.496955633163452,
            "validation_loss": 4.149501623535156
          }
        ],
        "train_config": {
          "batch_size": 512,
          "eval_every": 750,
          "eval_examples": 20000,
          "gradient_clip": 1.0,
          "learning_rate": 0.001,
          "seed": 47,
          "steps": 6000,
          "weight_decay": 0.0001
        },
        "validation_loss": 3.901860838137893,
        "validation_perplexity": 49.49446465339945
      }
    ],
    "multihead": [
      {
        "checkpoint_selection": "lowest validation loss on the fixed selection slice",
        "device": "mps",
        "final_train_batch_loss": 2.971235513687134,
        "parameter_count": 1321120,
        "runtime_seconds": 69.58758166688494,
        "seed": 11,
        "selected_step": 6000,
        "selected_tokens_seen": 3072000,
        "selection_slice_validation_loss": 3.337324810791016,
        "test_loss": 3.3463387954578963,
        "test_perplexity": 28.39857010425462,
        "tokens_seen": 3072000,
        "torch_version": "2.14.0",
        "trace": [
          {
            "step": 1,
            "tokens_seen": 512,
            "train_batch_loss": 8.29377555847168,
            "validation_loss": 7.926408984375
          },
          {
            "step": 750,
            "tokens_seen": 384000,
            "train_batch_loss": 3.626685619354248,
            "validation_loss": 3.7889643310546877
          },
          {
            "step": 1500,
            "tokens_seen": 768000,
            "train_batch_loss": 3.4321608543395996,
            "validation_loss": 3.5628756958007815
          },
          {
            "step": 2250,
            "tokens_seen": 1152000,
            "train_batch_loss": 3.3293113708496094,
            "validation_loss": 3.4884724304199217
          },
          {
            "step": 3000,
            "tokens_seen": 1536000,
            "train_batch_loss": 3.1293325424194336,
            "validation_loss": 3.42502177734375
          },
          {
            "step": 3750,
            "tokens_seen": 1920000,
            "train_batch_loss": 2.8762521743774414,
            "validation_loss": 3.382447381591797
          },
          {
            "step": 4500,
            "tokens_seen": 2304000,
            "train_batch_loss": 2.7776896953582764,
            "validation_loss": 3.35640068359375
          },
          {
            "step": 5250,
            "tokens_seen": 2688000,
            "train_batch_loss": 2.9949440956115723,
            "validation_loss": 3.3576778198242185
          },
          {
            "step": 6000,
            "tokens_seen": 3072000,
            "train_batch_loss": 2.971235513687134,
            "validation_loss": 3.337324810791016
          }
        ],
        "train_config": {
          "batch_size": 512,
          "eval_every": 750,
          "eval_examples": 20000,
          "gradient_clip": 1.0,
          "learning_rate": 0.003,
          "seed": 11,
          "steps": 6000,
          "weight_decay": 0.001
        },
        "validation_loss": 3.2916671648852103,
        "validation_perplexity": 26.887652460391237
      },
      {
        "checkpoint_selection": "lowest validation loss on the fixed selection slice",
        "device": "mps",
        "final_train_batch_loss": 2.8964927196502686,
        "parameter_count": 1321120,
        "runtime_seconds": 71.0595688750036,
        "seed": 29,
        "selected_step": 5250,
        "selected_tokens_seen": 2688000,
        "selection_slice_validation_loss": 3.355985888671875,
        "test_loss": 3.384119482652646,
        "test_perplexity": 29.49201305095732,
        "tokens_seen": 3072000,
        "torch_version": "2.14.0",
        "trace": [
          {
            "step": 1,
            "tokens_seen": 512,
            "train_batch_loss": 8.369367599487305,
            "validation_loss": 7.982183837890625
          },
          {
            "step": 750,
            "tokens_seen": 384000,
            "train_batch_loss": 3.8310956954956055,
            "validation_loss": 3.833468127441406
          },
          {
            "step": 1500,
            "tokens_seen": 768000,
            "train_batch_loss": 3.395033359527588,
            "validation_loss": 3.604370343017578
          },
          {
            "step": 2250,
            "tokens_seen": 1152000,
            "train_batch_loss": 3.402129888534546,
            "validation_loss": 3.5078741760253904
          },
          {
            "step": 3000,
            "tokens_seen": 1536000,
            "train_batch_loss": 3.1808314323425293,
            "validation_loss": 3.4411451416015626
          },
          {
            "step": 3750,
            "tokens_seen": 1920000,
            "train_batch_loss": 3.249908447265625,
            "validation_loss": 3.399086804199219
          },
          {
            "step": 4500,
            "tokens_seen": 2304000,
            "train_batch_loss": 3.2923173904418945,
            "validation_loss": 3.3724864074707033
          },
          {
            "step": 5250,
            "tokens_seen": 2688000,
            "train_batch_loss": 3.0055112838745117,
            "validation_loss": 3.355985888671875
          },
          {
            "step": 6000,
            "tokens_seen": 3072000,
            "train_batch_loss": 2.8964927196502686,
            "validation_loss": 3.359120422363281
          }
        ],
        "train_config": {
          "batch_size": 512,
          "eval_every": 750,
          "eval_examples": 20000,
          "gradient_clip": 1.0,
          "learning_rate": 0.003,
          "seed": 29,
          "steps": 6000,
          "weight_decay": 0.001
        },
        "validation_loss": 3.3284040741388186,
        "validation_perplexity": 27.89378974146384
      },
      {
        "checkpoint_selection": "lowest validation loss on the fixed selection slice",
        "device": "mps",
        "final_train_batch_loss": 2.7910141944885254,
        "parameter_count": 1321120,
        "runtime_seconds": 68.53351495880634,
        "seed": 47,
        "selected_step": 6000,
        "selected_tokens_seen": 3072000,
        "selection_slice_validation_loss": 3.3381174682617187,
        "test_loss": 3.348610447188453,
        "test_perplexity": 28.463155094706238,
        "tokens_seen": 3072000,
        "torch_version": "2.14.0",
        "trace": [
          {
            "step": 1,
            "tokens_seen": 512,
            "train_batch_loss": 8.337321281433105,
            "validation_loss": 7.977139208984375
          },
          {
            "step": 750,
            "tokens_seen": 384000,
            "train_batch_loss": 3.7893385887145996,
            "validation_loss": 3.7995723876953127
          },
          {
            "step": 1500,
            "tokens_seen": 768000,
            "train_batch_loss": 3.240550994873047,
            "validation_loss": 3.57667685546875
          },
          {
            "step": 2250,
            "tokens_seen": 1152000,
            "train_batch_loss": 3.3477418422698975,
            "validation_loss": 3.4717271118164064
          },
          {
            "step": 3000,
            "tokens_seen": 1536000,
            "train_batch_loss": 3.093116283416748,
            "validation_loss": 3.4135029541015625
          },
          {
            "step": 3750,
            "tokens_seen": 1920000,
            "train_batch_loss": 2.9752979278564453,
            "validation_loss": 3.386339001464844
          },
          {
            "step": 4500,
            "tokens_seen": 2304000,
            "train_batch_loss": 3.013843536376953,
            "validation_loss": 3.3542461242675783
          },
          {
            "step": 5250,
            "tokens_seen": 2688000,
            "train_batch_loss": 2.8977322578430176,
            "validation_loss": 3.346482122802734
          },
          {
            "step": 6000,
            "tokens_seen": 3072000,
            "train_batch_loss": 2.7910141944885254,
            "validation_loss": 3.3381174682617187
          }
        ],
        "train_config": {
          "batch_size": 512,
          "eval_every": 750,
          "eval_examples": 20000,
          "gradient_clip": 1.0,
          "learning_rate": 0.003,
          "seed": 47,
          "steps": 6000,
          "weight_decay": 0.001
        },
        "validation_loss": 3.299205748932946,
        "validation_perplexity": 27.091113227798363
      }
    ]
  },
  "source": {
    "api": "https://datasets-server.huggingface.co/rows",
    "dataset": "roneneldan/TinyStories",
    "dataset_revision": "f54c09fd23315a6f9c86f9dc80f725de7d8f9c64",
    "downloaded_stories": 6000,
    "license": "CDLA-Sharing-1.0",
    "offset_seed": 1701,
    "page_offsets": [
      12400,
      25400,
      190300,
      199700,
      345900,
      374600,
      525300,
      530900,
      543900,
      552600,
      569600,
      585700,
      588300,
      596200,
      602900,
      650600,
      683700,
      694900,
      734600,
      769200,
      778700,
      796300,
      801100,
      808100,
      808800,
      855200,
      901400,
      950200,
      961000,
      962400,
      977700,
      1120800,
      1182200,
      1233500,
      1252500,
      1322500,
      1525300,
      1555200,
      1561400,
      1562400,
      1570100,
      1581500,
      1611900,
      1616700,
      1619000,
      1632700,
      1676000,
      1678000,
      1810300,
      1866600,
      1941700,
      1956900,
      1979700,
      1985700,
      1992300,
      1992400,
      1996000,
      2071100,
      2102200,
      2113000
    ],
    "profile": "dgx",
    "raw_sha256": "ecaf3b99879170311c1fcde645251b9793620f1f5bf57159210fb7ab7180a7b3",
    "requested_stories": 6000,
    "retrieved_utc": "2026-09-16T04:24:08Z",
    "selection": "deterministic non-overlapping 100-row pages, then truncate",
    "source_split": "train"
  }
}