validation.json

{
  "model": "meta-encoder-mlx-4bit",
  "load_seconds": 4.304674542014254,
  "embedding_dim": 6656,
  "checks": [
    {
      "name": "text_animal",
      "seconds": 3.765016416000435,
      "scores": [
        0.3861164450645447,
        0.2507832944393158,
        0.19388027489185333
      ],
      "candidates": [
        "cat",
        "dog",
        "horse"
      ],
      "predicted": "cat",
      "expected": "cat",
      "correct": true,
      "norms": [
        1.0,
        1.0,
        1.0,
        0.9999999403953552
      ]
    },
    {
      "name": "text_action",
      "seconds": 1.8200458750070538,
      "scores": [
        0.6188708543777466,
        0.17132438719272614,
        0.19522921741008759
      ],
      "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": 1.9545344169891905,
      "scores": [
        0.4928099513053894,
        0.4084464907646179,
        0.40224990248680115
      ],
      "candidates": [
        "red",
        "blue",
        "green"
      ],
      "predicted": "red",
      "expected": "red",
      "correct": true,
      "norms": [
        1.0,
        1.0,
        1.0,
        1.0
      ]
    },
    {
      "name": "video_color",
      "seconds": 1.4857492909941357,
      "scores": [
        0.5262743830680847,
        0.4340401291847229,
        0.418895423412323
      ],
      "candidates": [
        "red",
        "blue",
        "green"
      ],
      "predicted": "red",
      "expected": "red",
      "correct": true,
      "norms": [
        1.0,
        1.0,
        1.0,
        1.0
      ]
    }
  ],
  "total_seconds": 13.330337374994997,
  "peak_mlx_memory_gb": 19.696706988,
  "bf16_mlx_reference_task_cosine": 0.97298264503479,
  "bf16_mlx_reference_video_task_cosine": 0.9559712409973145,
  "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."
}