{
  "config": {
    "input_seed": 42,
    "timed_runs": 10,
    "tolerance_abs": 0.00001,
    "tolerance_rel": 0.001,
    "warmup_runs": 3
  },
  "input_spec": "1x64x768:float32;1x64:float32",
  "license": "apache-2.0",
  "model_id": "vela-1.0-encoder-307m-embedding__vela_l64_embeds_fp16",
  "notes": "ModernBERT sentence embedder (22 layers, 768-d; mean pooling + L2 normalization in the graph), litert-community/Vela-1.0-Encoder-307M-Embedding-LiteRT main f92189a9 (uploaded 2026-10-18; sha256 d2c473a7… = the Hub LFS sha256 = the staged file, 220,882,048 B): host-lookup encoder (inputs_embeds form; fp16 FULLY_CONNECTED weights read through 88 DEQUANTIZE; the host feeds the fp16 token table tok_embeddings_fp16.bin), window 64; 1,689 operators; a random fill of inputs_embeds / attention_mask is not a meaningful input (loads / runs / delegation are the signal); on the phones the GPU runs at FP32 precision (float16 arithmetic misses the per-row cosine floor, see the device records); the one-file vela_L*_fp16.tflite graphs (614-616 MB) are not catalogued",
  "results": [
    {
      "backend": "wasm_xnnpack",
      "date": "2026-10-18",
      "delegation_evidence": [
        "error: WARNING: [npu_registry.cc:34] NPU accelerator could not be loaded and registered: kLiteRtStatusErrorInvalidArgument.",
        "error: WARNING: [npu_registry.cc:34] NPU accelerator could not be loaded and registered: kLiteRtStatusErrorInvalidArgument."
      ],
      "env": {
        "browser": "chromium",
        "browser_version": "151.0.7922.34",
        "headless": true,
        "jspi": true,
        "litertjs_core_version": "2.5.3",
        "machine_label": "mac-studio-m4-max",
        "os": "macOS",
        "os_version": "27.0.0",
        "webgpu_adapter": {
          "architecture": "metal-3",
          "description": "",
          "device": "",
          "vendor": "apple"
        }
      },
      "error": null,
      "failure_class": null,
      "full_delegation": null,
      "latency_p50_ms": 196.29,
      "loads": true,
      "max_abs_diff": null,
      "max_rel_diff": null,
      "output_match": null,
      "provenance": "measured",
      "runs": true
    },
    {
      "backend": "webgpu_mldrift",
      "date": "2026-10-18",
      "delegation_evidence": [
        "error: WARNING: [npu_registry.cc:34] NPU accelerator could not be loaded and registered: kLiteRtStatusErrorInvalidArgument.",
        "error: WARNING: [npu_registry.cc:34] NPU accelerator could not be loaded and registered: kLiteRtStatusErrorInvalidArgument."
      ],
      "env": {
        "browser": "chromium",
        "browser_version": "151.0.7922.34",
        "headless": true,
        "jspi": true,
        "litertjs_core_version": "2.5.3",
        "machine_label": "mac-studio-m4-max",
        "os": "macOS",
        "os_version": "27.0.0",
        "webgpu_adapter": {
          "architecture": "metal-3",
          "description": "",
          "device": "",
          "vendor": "apple"
        }
      },
      "error": null,
      "failure_class": null,
      "full_delegation": true,
      "latency_p50_ms": 14.09,
      "loads": true,
      "max_abs_diff": 0.000001214444637298584,
      "max_rel_diff": 0.00173499573591404,
      "output_match": true,
      "provenance": "measured",
      "runs": true
    }
  ],
  "schema_version": "1.0",
  "source": "https://huggingface.co/litert-community/Vela-1.0-Encoder-307M-Embedding-LiteRT/resolve/main/vela_L64_embeds_fp16.tflite"
}
