HyperlinksSpace/TinyModel1

1{2  "reproducibility": {3    "seed": 42,4    "dataset": "fancyzhx/ag_news",5    "dataset_config": null,6    "train_split": "train",7    "eval_split": "test",8    "text_column": "text",9    "label_column": "label",10    "max_train_samples": 3000,11    "max_eval_samples": 600,12    "note": "Train and eval rows are the first N after shuffle(seed) of each split; see texts/eval-reproducibility.md."13  },14  "metrics": {15    "accuracy": 0.538333,16    "macro_f1": 0.455368,17    "weighted_f1": 0.452694,18    "per_class_f1": {19      "World": 0.536585,20      "Sports": 0.730964,21      "Business": 0.0,22      "Sci/Tech": 0.55392223    },24    "confusion_matrix": [25      [26        66,27        67,28        0,29        1530      ],31      [32        1,33        144,34        0,35        236      ],37      [38        12,39        15,40        0,41        12542      ],43      [44        19,45        21,46        0,47        11348      ]49    ],50    "confusion_matrix_axis": "rows=true class, columns=predicted class",51    "label_order": [52      "World",53      "Sports",54      "Business",55      "Sci/Tech"56    ]57  },58  "dataset_quality": {59    "class_distribution": {60      "train": {61        "counts_by_label": {62          "World": 771,63          "Sports": 742,64          "Business": 691,65          "Sci/Tech": 79666        },67        "proportions_by_label": {68          "World": 0.257,69          "Sports": 0.247333,70          "Business": 0.230333,71          "Sci/Tech": 0.26533372        },73        "total": 300074      },75      "eval": {76        "counts_by_label": {77          "World": 148,78          "Sports": 147,79          "Business": 152,80          "Sci/Tech": 15381        },82        "proportions_by_label": {83          "World": 0.246667,84          "Sports": 0.245,85          "Business": 0.253333,86          "Sci/Tech": 0.25587        },88        "total": 60089      }90    }91  },92  "error_analysis": {93    "top_confusions": [94      {95        "true_label": "Business",96        "predicted_label": "Sci/Tech",97        "count": 12598      },99      {100        "true_label": "World",101        "predicted_label": "Sports",102        "count": 67103      },104      {105        "true_label": "Sci/Tech",106        "predicted_label": "Sports",107        "count": 21108      },109      {110        "true_label": "Sci/Tech",111        "predicted_label": "World",112        "count": 19113      },114      {115        "true_label": "World",116        "predicted_label": "Sci/Tech",117        "count": 15118      },119      {120        "true_label": "Business",121        "predicted_label": "Sports",122        "count": 15123      },124      {125        "true_label": "Business",126        "predicted_label": "World",127        "count": 12128      },129      {130        "true_label": "Sports",131        "predicted_label": "Sci/Tech",132        "count": 2133      },134      {135        "true_label": "Sports",136        "predicted_label": "World",137        "count": 1138      }139    ]140  },141  "calibration": {142    "max_prob_histogram": {143      "num_bins": 10,144      "bins": [145        {146          "bin_low": 0.0,147          "bin_high": 0.1,148          "count": 0149        },150        {151          "bin_low": 0.1,152          "bin_high": 0.2,153          "count": 0154        },155        {156          "bin_low": 0.2,157          "bin_high": 0.3,158          "count": 1159        },160        {161          "bin_low": 0.3,162          "bin_high": 0.4,163          "count": 27164        },165        {166          "bin_low": 0.4,167          "bin_high": 0.5,168          "count": 156169        },170        {171          "bin_low": 0.5,172          "bin_high": 0.6,173          "count": 237174        },175        {176          "bin_low": 0.6,177          "bin_high": 0.7,178          "count": 171179        },180        {181          "bin_low": 0.7,182          "bin_high": 0.8,183          "count": 8184        },185        {186          "bin_low": 0.8,187          "bin_high": 0.9,188          "count": 0189        },190        {191          "bin_low": 0.9,192          "bin_high": 1.0,193          "count": 0194        }195      ],196      "note": "Each eval example contributes one max softmax probability (winner class)."197    }198  },199  "routing": {200    "fallback_behavior": "At inference, if the maximum softmax probability is below `min_confidence`, treat the prediction as low-confidence: route to human review, a secondary model, or a safe default class\u2014choose per product.",201    "min_confidence": null,202    "comment": "`min_confidence` is not set by training; typical starting range is 0.5\u20130.7 for routing. Tune on a validation set using `max_prob` histogram and error analysis."203  }204}205