{
  "measured_at": "2026-09-20T08:10:40.233Z",
  "method": "One fixed input per task type, one warm-up run then one timed run, on the hub's own machine. The harness is scripts/measure.py; the input shapes are defined there and named in each figure.",
  "results": {
    "smollm2-135m-instruct": {
      "value": 110.2,
      "metric": "one 64-token reply to a fixed prompt",
      "short": "64-token reply · mlx-lm",
      "note": "greedy decoding, mlx-lm on the GPU, publisher bf16 weights",
      "slug": "smollm2-135m-instruct",
      "unit": "ms",
      "hardware": "Apple M4 Pro, 48 GB unified memory, macOS 15.6",
      "rss_mb": 658
    },
    "qwen2-5-0-5b-instruct": {
      "value": 266.8,
      "metric": "one 64-token reply to a fixed prompt",
      "short": "64-token reply · mlx-lm",
      "note": "greedy decoding, mlx-lm on the GPU, publisher bf16 weights",
      "slug": "qwen2-5-0-5b-instruct",
      "unit": "ms",
      "hardware": "Apple M4 Pro, 48 GB unified memory, macOS 15.6",
      "rss_mb": 1451
    },
    "qwen3-0-6b": {
      "value": 469.6,
      "metric": "one 64-token reply to a fixed prompt",
      "short": "64-token reply · mlx-lm",
      "note": "greedy decoding, mlx-lm on the GPU, publisher bf16 weights",
      "slug": "qwen3-0-6b",
      "unit": "ms",
      "hardware": "Apple M4 Pro, 48 GB unified memory, macOS 15.6",
      "rss_mb": 1697
    },
    "qwen2-5-0-5b-instruct-4bit-mlx": {
      "value": 241.4,
      "metric": "one 64-token reply to a fixed prompt",
      "short": "64-token reply · mlx-lm",
      "note": "greedy decoding, mlx-lm on the GPU, publisher bf16 weights",
      "slug": "qwen2-5-0-5b-instruct-4bit-mlx",
      "unit": "ms",
      "hardware": "Apple M4 Pro, 48 GB unified memory, macOS 15.6",
      "rss_mb": 786
    },
    "all-minilm-l6-v2": {
      "value": 5.3,
      "metric": "one fixed 24-word sentence",
      "short": "1 sentence · transformers",
      "note": "one sentence embedded (one forward pass), transformers, float32",
      "slug": "all-minilm-l6-v2",
      "unit": "ms",
      "hardware": "Apple M4 Pro, 48 GB unified memory, macOS 15.6",
      "rss_mb": 395
    },
    "distilbert-sst-2-english": {
      "value": 13.1,
      "metric": "one fixed 24-word sentence",
      "short": "1 sentence · transformers",
      "note": "one sentence classified (one forward pass), transformers, float32",
      "slug": "distilbert-sst-2-english",
      "unit": "ms",
      "hardware": "Apple M4 Pro, 48 GB unified memory, macOS 15.6",
      "rss_mb": 525
    },
    "resnet-50": {
      "value": 21.7,
      "metric": "one 640x480 photo (coco-val2017-000000039769.jpg)",
      "short": "1 photo (640x480) · transformers",
      "note": "one forward pass, transformers, float32",
      "slug": "resnet-50",
      "unit": "ms",
      "hardware": "Apple M4 Pro, 48 GB unified memory, macOS 15.6",
      "rss_mb": 505
    },
    "detr-resnet-50": {
      "value": 251.1,
      "metric": "one 640x480 photo (coco-val2017-000000039769.jpg)",
      "short": "1 photo (640x480) · transformers",
      "note": "one forward pass, transformers, float32",
      "slug": "detr-resnet-50",
      "unit": "ms",
      "hardware": "Apple M4 Pro, 48 GB unified memory, macOS 15.6",
      "rss_mb": 1373
    },
    "videomae-base-finetuned-kinetics": {
      "value": 214.2,
      "metric": "one clip of 16 frames at 224x224",
      "short": "1 clip (16 frames) · transformers",
      "note": "one forward pass, transformers, float32, synthetic clip",
      "slug": "videomae-base-finetuned-kinetics",
      "unit": "ms",
      "hardware": "Apple M4 Pro, 48 GB unified memory, macOS 15.6",
      "rss_mb": 976
    },
    "smolvlm2-256m-video-instruct": {
      "value": 2841.4,
      "metric": "one clip of 16 frames at 224x224 + one question",
      "short": "1 clip + question · transformers",
      "note": "greedy decoding, transformers, float32, synthetic clip",
      "slug": "smolvlm2-256m-video-instruct",
      "unit": "ms",
      "hardware": "Apple M4 Pro, 48 GB unified memory, macOS 15.6",
      "rss_mb": 3171
    },
    "florence-2-base-ft": {
      "value": 2124.4,
      "metric": "one 640x480 photo (coco-val2017-000000039769.jpg) + '<OD>'",
      "short": "1 photo + prompt · transformers",
      "note": "greedy decoding, transformers, float32",
      "slug": "florence-2-base-ft",
      "unit": "ms",
      "hardware": "Apple M4 Pro, 48 GB unified memory, macOS 15.6",
      "rss_mb": 2109
    },
    "qwen2-5-0-5b-instruct-gguf-q4-k-m": {
      "value": 255.4,
      "metric": "one 64-token reply to a fixed prompt",
      "short": "64-token reply · llama.cpp",
      "note": "greedy decoding, llama.cpp",
      "slug": "qwen2-5-0-5b-instruct-gguf-q4-k-m",
      "unit": "ms",
      "hardware": "Apple M4 Pro, 48 GB unified memory, macOS 15.6",
      "rss_mb": 795
    }
  }
}