{
  "kind": "comparison",
  "compare_name": "compare_gw6_text_grounded_reptok_attention_small_heads_lpips_clip",
  "run_name": "gw6_imagenet_16h_plus96h_small_1gpu_sonar_m1_c16_v2p5_s24",
  "checkpoint": "last.ckpt",
  "input_modality": "bge",
  "eval_samples": 256,
  "display_samples": 12,
  "attention_models": 6,
  "modalities": [
    "dinov2",
    "reptok",
    "bge",
    "sonar",
    "symbolic"
  ],
  "metrics": {
    "lpips": "Decoded RepTok image vs ground-truth ImageNet image; lower is better.",
    "clip_score": "ViT-B-32/openai cosine between decoded image and paired caption; higher is better."
  },
  "equal_fusion_baseline": {
    "method": "equal_fusion",
    "lpips": 0.6398796766297892,
    "clip_score": 0.32596434495644644
  },
  "direct_baseline": {
    "method": "direct_bge_to_reptok",
    "lpips": 0.6381985465995967,
    "clip_score": 0.32466380577534437
  },
  "best_by_lpips": {
    "config_id": "hs16_nh16_lr0p0003",
    "head_size": 16,
    "num_heads": 16,
    "learning_rate": 0.0003,
    "latent_eval_mse": 0.31531702916933285,
    "latent_eval_objective": 3.1531703042699455,
    "lpips": 0.6378988596843556,
    "clip_score": 0.32692726684035733,
    "mean_w_dinov2": 0.4096839427947998,
    "mean_w_reptok": 0.16788743436336517,
    "mean_w_bge": 0.2722996771335602,
    "mean_w_sonar": 0.10792897641658783,
    "mean_w_symbolic": 0.04220002144575119
  },
  "best_by_clip": {
    "config_id": "hs8_nh16_lr0p0003",
    "head_size": 8,
    "num_heads": 16,
    "learning_rate": 0.0003,
    "latent_eval_mse": 0.31695234135613654,
    "latent_eval_objective": 3.169523427671565,
    "lpips": 0.6388121392810717,
    "clip_score": 0.3272198011400178,
    "mean_w_dinov2": 0.40983736515045166,
    "mean_w_reptok": 0.17639705538749695,
    "mean_w_bge": 0.26894161105155945,
    "mean_w_sonar": 0.10239265859127045,
    "mean_w_symbolic": 0.04243128001689911
  },
  "rankings": [
    {
      "config_id": "hs16_nh16_lr0p0003",
      "head_size": 16,
      "num_heads": 16,
      "learning_rate": 0.0003,
      "latent_eval_mse": 0.31531702916933285,
      "latent_eval_objective": 3.1531703042699455,
      "lpips": 0.6378988596843556,
      "clip_score": 0.32692726684035733,
      "mean_w_dinov2": 0.4096839427947998,
      "mean_w_reptok": 0.16788743436336517,
      "mean_w_bge": 0.2722996771335602,
      "mean_w_sonar": 0.10792897641658783,
      "mean_w_symbolic": 0.04220002144575119
    },
    {
      "config_id": "hs16_nh4_lr0p0003",
      "head_size": 16,
      "num_heads": 4,
      "learning_rate": 0.0003,
      "latent_eval_mse": 0.3176862570267262,
      "latent_eval_objective": 3.1768625856200816,
      "lpips": 0.6382354395464063,
      "clip_score": 0.3265026831650175,
      "mean_w_dinov2": 0.4122699499130249,
      "mean_w_reptok": 0.17648185789585114,
      "mean_w_bge": 0.2660730481147766,
      "mean_w_sonar": 0.10238095372915268,
      "mean_w_symbolic": 0.04279417544603348
    },
    {
      "config_id": "hs16_nh8_lr0p0003",
      "head_size": 16,
      "num_heads": 8,
      "learning_rate": 0.0003,
      "latent_eval_mse": 0.3161244233808622,
      "latent_eval_objective": 3.1612442415114415,
      "lpips": 0.6381167529616505,
      "clip_score": 0.32678486389340833,
      "mean_w_dinov2": 0.3848973512649536,
      "mean_w_reptok": 0.1773742288351059,
      "mean_w_bge": 0.269013375043869,
      "mean_w_sonar": 0.10864579677581787,
      "mean_w_symbolic": 0.0600692443549633
    },
    {
      "config_id": "hs8_nh16_lr0p0003",
      "head_size": 8,
      "num_heads": 16,
      "learning_rate": 0.0003,
      "latent_eval_mse": 0.31695234135613654,
      "latent_eval_objective": 3.169523427671565,
      "lpips": 0.6388121392810717,
      "clip_score": 0.3272198011400178,
      "mean_w_dinov2": 0.40983736515045166,
      "mean_w_reptok": 0.17639705538749695,
      "mean_w_bge": 0.26894161105155945,
      "mean_w_sonar": 0.10239265859127045,
      "mean_w_symbolic": 0.04243128001689911
    },
    {
      "config_id": "hs8_nh4_lr0p0003",
      "head_size": 8,
      "num_heads": 4,
      "learning_rate": 0.0003,
      "latent_eval_mse": 0.319063587040807,
      "latent_eval_objective": 3.1906358955613037,
      "lpips": 0.6391531288390979,
      "clip_score": 0.326928015332669,
      "mean_w_dinov2": 0.4013785719871521,
      "mean_w_reptok": 0.17479237914085388,
      "mean_w_bge": 0.2854422926902771,
      "mean_w_sonar": 0.09470194578170776,
      "mean_w_symbolic": 0.043684788048267365
    },
    {
      "config_id": "hs8_nh8_lr0p0003",
      "head_size": 8,
      "num_heads": 8,
      "learning_rate": 0.0003,
      "latent_eval_mse": 0.31774280896357665,
      "latent_eval_objective": 3.177428109069183,
      "lpips": 0.6388172802980989,
      "clip_score": 0.3254909201641567,
      "mean_w_dinov2": 0.39209896326065063,
      "mean_w_reptok": 0.17834575474262238,
      "mean_w_bge": 0.2899625897407532,
      "mean_w_sonar": 0.09961926192045212,
      "mean_w_symbolic": 0.039973415434360504
    }
  ],
  "artifacts": {
    "csv": "all_attention_lpips_clip.csv",
    "ranking_plot": "assets/attention_lpips_clip_rankings.jpg",
    "baseline_comparison": "assets/baseline_comparison.jpg",
    "per_image_scatter": "assets/per_image_clip_lpips_scatter.jpg",
    "per_image_csv": "per_image_clip_lpips.csv",
    "qualitative_grid": "assets/attention_qualitative_grid.jpg"
  },
  "page_title": "GW6 Boss Attention Mini-Grid: Train-MSE, LPIPS, and CLIP"
}