validation.json

{
  "model": "meta-encoder-mlx-8bit",
  "load_seconds": 8.406539290997898,
  "embedding_dim": 6656,
  "checks": [
    {
      "name": "text_animal",
      "seconds": 4.789314416004345,
      "scores": [
        0.3625825345516205,
        0.2203747034072876,
        0.16552267968654633
      ],
      "candidates": [
        "cat",
        "dog",
        "horse"
      ],
      "predicted": "cat",
      "expected": "cat",
      "correct": true,
      "norms": [
        1.0,
        1.0,
        1.0,
        1.0
      ]
    },
    {
      "name": "text_action",
      "seconds": 1.82342183298897,
      "scores": [
        0.6134610772132874,
        0.16023504734039307,
        0.20238327980041504
      ],
      "candidates": [
        "Click Cancel order",
        "Open profile settings",
        "Search products"
      ],
      "predicted": "Click Cancel order",
      "expected": "Click Cancel order",
      "correct": true,
      "norms": [
        1.0,
        1.0,
        1.0,
        1.0
      ]
    },
    {
      "name": "image_color",
      "seconds": 1.7876198329904582,
      "scores": [
        0.4851677715778351,
        0.3475082814693451,
        0.3573587238788605
      ],
      "candidates": [
        "red",
        "blue",
        "green"
      ],
      "predicted": "red",
      "expected": "red",
      "correct": true,
      "norms": [
        1.0,
        0.9999999403953552,
        1.0,
        1.0
      ]
    },
    {
      "name": "video_color",
      "seconds": 1.3898598329978995,
      "scores": [
        0.4881054162979126,
        0.384658545255661,
        0.3898668885231018
      ],
      "candidates": [
        "red",
        "blue",
        "green"
      ],
      "predicted": "red",
      "expected": "red",
      "correct": true,
      "norms": [
        1.0,
        0.9999999403953552,
        1.0,
        1.0
      ]
    }
  ],
  "total_seconds": 18.199415957991732,
  "peak_mlx_memory_gb": 33.596695948,
  "bf16_mlx_reference_task_cosine": 0.9996585845947266,
  "bf16_mlx_reference_video_task_cosine": 0.9993553161621094,
  "reference_scope": "One text task and one two-frame video task compared with streamed BF16 MLX inference from the original checkpoint. Full-model Transformers parity and benchmark accuracy were not measured."
}