{
  "artifacts": [
    {
      "file": "LFM2.5-VL-3B_int8.litertlm",
      "sha256": "392eeae18ddbe7a851d5b37feae8ee13819def03abcd5a64cf56e5a2f302b298",
      "size_mb": 3387.193
    }
  ],
  "benchmarks": [],
  "conversion": {
    "command": "lfm_work/convert_lfm25_vl.py --task image_text_to_text (hf-to-litertlm; HF card Conversion)",
    "quantization": "int8 dynamic on text linears + convs + embedding, vision tower int8 (HF card file table)",
    "tool": "hf-to-litertlm lfm_work/convert_lfm25_vl.py (litert-torch 0.9.3, --task image_text_to_text) — exact recipe, int4 post-processing and executor metadata all in the open pipeline (HF card Conversion)",
    "tool_version": "0.9.3 (HF card Conversion)"
  },
  "cross_runtime": [],
  "delegation": null,
  "device": {
    "records": [
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "cpu",
          "context_length": null,
          "date": "2026-09-05",
          "decode_tokens_per_s": 19.87,
          "delegated_ops": 1330,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-cold-cache-cooled",
            "os_build": "Android 16",
            "runtime": "litert-lm",
            "runtime_version": "0.16.0",
            "soc": "Qualcomm SM8850",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "VERBOSE: Replacing 805 out of 937 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 192 partitions for subgraph 0 (prefill_1024).",
            "VERBOSE: Replacing 805 out of 937 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 192 partitions for subgraph 1 (prefill_512).",
            "VERBOSE: Replacing 805 out of 937 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 192 partitions for subgraph 2 (prefill_256).",
            "VERBOSE: Replacing 805 out of 937 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 192 partitions for subgraph 3 (prefill_128).",
            "results block: prefill=277.36 decode=19.87 tokens/s"
          ],
          "failure_class": null,
          "full_delegation": false,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "decode_tokens": 100.0,
            "init_s": 5.41489,
            "prefill_tokens": 205.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": 277.36,
          "provenance": "measured",
          "runs": true,
          "total_ops": 1387,
          "ttft_ms": 790.0
        },
        "source": "data/device_runs/0.16.0/2026-09-05/lfm2.5-vl-3b-int8__galaxy-s26.json"
      },
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "gpu",
          "context_length": null,
          "date": "2026-08-24",
          "decode_tokens_per_s": 20.15,
          "delegated_ops": 937,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-cl-pinned",
            "os_build": "Android 16",
            "runtime": "litert-lm",
            "runtime_version": "0.16.0",
            "soc": "Qualcomm SM8850",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "VERBOSE: Replacing 937 out of 937 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 0 (prefill_1024).",
            "VERBOSE: Replacing 937 out of 937 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 1 (prefill_512).",
            "VERBOSE: Replacing 937 out of 937 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 2 (prefill_256).",
            "VERBOSE: Replacing 937 out of 937 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 3 (prefill_128).",
            "results block: prefill=743.88 decode=20.15 tokens/s"
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "decode_tokens": 109.0,
            "init_s": 12.24928,
            "prefill_tokens": 205.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": 743.88,
          "provenance": "measured",
          "runs": true,
          "total_ops": 937,
          "ttft_ms": 330.0
        },
        "source": "data/device_runs/0.16.0/2026-08-24/lfm2.5-vl-3b-int8__galaxy-s26.json"
      }
    ]
  },
  "model": {
    "family": "lfm2.5",
    "id": "lfm2.5-vl-3b-int8",
    "license": "other (lfm-open-license-v1.0)",
    "source_url": "https://huggingface.co/litert-community/LFM2.5-VL-3B",
    "task": "image-text-to-text"
  },
  "pitfalls": [
    "Android: Pixel 8a (Tensor G3), litert_lm_main built from the v0.16.0 release tag, 292-token prompt, decode run to EOS (≈1.1k tokens sustained), --disable_cache worst-case load — measured on the int4 file (HF card Speed); this repo's S26 CPU row for the int8 file is from the S7 backfill (device runs).",
    "License: LFM Open License v1.0 (HF tag license:other; card license_name lfm-open-license-v1.0, LICENSE file in the repo) (HF card front matter).",
    "LiteRT-LM .litertlm bundle: LiteRT.js cannot run it, so delegation stays null and there is no browser block; device rows come from data/device_runs/ (Pi 5 / S26 / Pixel 8a / Mac / iPhone as measured)."
  ],
  "schema_version": "1.2"
}