{
  "artifacts": [
    {
      "file": "LFM2-2.6B_int8.litertlm",
      "sha256": "3dda920ce47e5ef2597c1a43732daeecc5d4b9b91fe29599cb9eaafb874b29e8",
      "size_mb": 2597.627
    }
  ],
  "benchmarks": [],
  "conversion": {
    "command": "LiquidAI/LFM2-2.6B rev 36ed799f: unquantized export (11 prefill signatures, 1024 down to 1, plus decode, over a 4096-position KV cache) -> post-hoc int8 quantization -> metadata repack with LiteRT-LM's lfm2 configuration (card L110-L111)",
    "quantization": "int8 dynamic on linears + int8 embedding, convs float (card L27)",
    "tool": "litert-torch 0.9.4; repro = hf-to-litertlm lfm_work/convert_lfm2_26b.sh (card L110)",
    "tool_version": "litert-torch 0.9.4, litert-converter 0.4.0, ai-edge-quantizer 0.9.0, transformers 5.14.1, torch 2.13.0; packed by the litert-lm 0.18.0 CLI"
  },
  "cross_runtime": [],
  "delegation": null,
  "device": {
    "records": [
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "cpu",
          "context_length": null,
          "date": "2026-10-12",
          "decode_tokens_per_s": 17.07,
          "delegated_ops": 1394,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-aab1",
            "os_build": "Android 16 (BP4A.251205.006)",
            "runtime": "litert-lm",
            "runtime_version": "prebuilt-adac974b",
            "soc": "Qualcomm SM8850",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "binary: Google's prebuilt litert_lm_advanced_main + 6 .so (gs://litert/binaries/latest/android_arm64/litert_lm/, object dated 2026-09-18, 40,382,152 B, sha256 adac974b…)",
            "r1 hold 2026-10-12 19:25:49-19:42:28 JST: [19:27:10] bundle on device: 3dda920ce47e5ef2597c1a43732daeecc5d4b9b91fe29599cb9eaafb874b29e8 (2723809744 B, push 72.2 s) = Mac",
            "bench args: --backend=cpu --model_path=/data/local/tmp/aa3_26b/LFM2-2.6B_int8.litertlm --benchmark=true --benchmark_prefill_tokens=1024 --benchmark_decode_tokens=256 --max_num_tokens=2048 --num_iterations=3 --report_peak_memory_footprint=true --metric_proto_file_path=/data/local/tmp/aa3_26b/metrics.pb",
            "VERBOSE: Replacing 869 out of 978 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 152 partitions for subgraph 0 (prefill_1024).",
            "VERBOSE: Replacing 795 out of 895 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 156 partitions for subgraph 11 (decode).",
            "VERBOSE: Replacing 1394 out of 1428 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 18 partitions for subgraph 0 (prefill_1024).",
            "VERBOSE: Replacing 1273 out of 1296 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 18 partitions for subgraph 11 (decode).",
            "[19:29:54] parsed: s26_i8_cpu_timing cpu: gate PASS 1/1 | bench prefill 322.56 decode 17.07 init 221.5 cold 3100.79 peak 4158.28 capped 24/25 | answers: 'The capital of France is Paris.'",
            "timing run prompt process: \"The capital of France is Paris.\"",
            "check run (hold 2026-10-12 19:25:49-19:42:28 JST): 2 of 2 prompt processes answered with the expected word (\"The capital of France is Paris.\")"
          ],
          "failure_class": null,
          "full_delegation": false,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "boot_epoch_s": 1791590918.97,
            "capped_samples": 24.0,
            "cold_init_ms": 3100.79,
            "cpu_capped_samples": 24.0,
            "decode_tok_s_it1": 20.59,
            "decode_tok_s_it2": 16.46,
            "decode_tok_s_it3": 17.68,
            "decode_tokens": 256.0,
            "init_ms": 221.5,
            "iterations": 3.0,
            "kgsl_ceiling_min_mhz": 1300.0,
            "kgsl_temp_max_c": 65.1,
            "max_num_tokens": 2048.0,
            "memavail_at_bench_start_kib": 7119236.0,
            "memavail_min_kib": 5509836.0,
            "prefill_tok_s_it1": 412.79,
            "prefill_tok_s_it2": 350.24,
            "prefill_tok_s_it3": 294.88,
            "prefill_tokens": 1024.0,
            "sampler_samples": 25.0,
            "skin_c_at_bench_start": 37.9,
            "t_1024_256_s": 18.1717,
            "thermal_status_at_bench_start": 0.0,
            "ttft_s_it1": 2.53,
            "ttft_s_it2": 2.98,
            "ttft_s_it3": 3.53
          },
          "output_match": null,
          "peak_mem_mb": 4158.28,
          "prefill_tokens_per_s": 322.56,
          "provenance": "measured",
          "runs": true,
          "total_ops": 1428,
          "ttft_ms": 3255.0
        },
        "source": "data/device_runs/prebuilt-adac974b/2026-10-12/lfm2-2.6b-int8__galaxy-s26.json"
      },
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "gpu",
          "context_length": null,
          "date": "2026-10-12",
          "decode_tokens_per_s": 20.39,
          "delegated_ops": 978,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-aab1",
            "os_build": "Android 16 (BP4A.251205.006)",
            "runtime": "litert-lm",
            "runtime_version": "prebuilt-adac974b",
            "soc": "Qualcomm SM8850",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "binary: Google's prebuilt litert_lm_advanced_main + 6 .so (gs://litert/binaries/latest/android_arm64/litert_lm/, object dated 2026-09-18, 40,382,152 B, sha256 adac974b…)",
            "r1 hold 2026-10-12 19:25:49-19:42:28 JST: [19:35:07] bundle on device: 3dda920ce47e5ef2597c1a43732daeecc5d4b9b91fe29599cb9eaafb874b29e8 (2723809744 B, push 0 s) = Mac",
            "bench args: --backend=gpu --model_path=/data/local/tmp/aa3_26b/LFM2-2.6B_int8.litertlm --benchmark=true --benchmark_prefill_tokens=1024 --benchmark_decode_tokens=256 --max_num_tokens=2048 --num_iterations=3 --report_peak_memory_footprint=true --metric_proto_file_path=/data/local/tmp/aa3_26b/metrics.pb",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 0 (prefill_1024).",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 1 (prefill_512).",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 2 (prefill_256).",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 3 (prefill_128).",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 4 (prefill_64).",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 5 (prefill_32).",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 6 (prefill_16).",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 7 (prefill_8).",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 8 (prefill_4).",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 9 (prefill_2).",
            "VERBOSE: Replacing 877 out of 877 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 10 (prefill_1).",
            "VERBOSE: Replacing 895 out of 895 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 11 (decode).",
            "[19:38:11] parsed: s26_i8_gpu_timing gpu: gate PASS 1/1 | bench prefill 1122.455 decode 20.39 init 5284.47 cold 9050.21 peak 784.406 capped 19/23 | answers: 'The capital of France is Paris.'",
            "timing run prompt process: \"The capital of France is Paris.\"",
            "check run (hold 2026-10-12 19:25:49-19:42:28 JST): 2 of 2 prompt processes answered with the expected word (\"The capital of France is Paris.\")"
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "boot_epoch_s": 1791590918.85,
            "capped_samples": 19.0,
            "cold_init_ms": 9050.21,
            "cpu_capped_samples": 20.0,
            "decode_tok_s_it1": 20.09,
            "decode_tok_s_it2": 20.48,
            "decode_tok_s_it3": 20.3,
            "decode_tokens": 256.0,
            "init_ms": 5284.47,
            "iterations": 3.0,
            "kgsl_ceiling_min_mhz": 902.0,
            "kgsl_temp_max_c": 93.4,
            "max_num_tokens": 2048.0,
            "memavail_at_bench_start_kib": 7635960.0,
            "memavail_min_kib": 3543620.0,
            "prefill_tok_s_it1": 1289.08,
            "prefill_tok_s_it2": 1114.46,
            "prefill_tok_s_it3": 1130.45,
            "prefill_tokens": 1024.0,
            "sampler_samples": 23.0,
            "skin_c_at_bench_start": 39.9,
            "t_1024_256_s": 13.4675,
            "thermal_status_at_bench_start": 1.0,
            "ttft_s_it1": 0.84,
            "ttft_s_it2": 0.97,
            "ttft_s_it3": 0.96
          },
          "output_match": null,
          "peak_mem_mb": 784.406,
          "prefill_tokens_per_s": 1122.455,
          "provenance": "measured",
          "runs": true,
          "total_ops": 978,
          "ttft_ms": 965.0
        },
        "source": "data/device_runs/prebuilt-adac974b/2026-10-12/lfm2-2.6b-int8__galaxy-s26.json"
      },
      {
        "device": "mac-studio-m4-max",
        "run": {
          "accelerator": "cpu",
          "context_length": null,
          "date": "2026-10-12",
          "decode_tokens_per_s": null,
          "delegated_ops": null,
          "env": {
            "device": "Mac Studio (M4 Max)",
            "machine_label": "mac-studio-m4-max",
            "os_build": "macOS 27.0",
            "runtime": "litert-lm",
            "runtime_version": "0.17.1",
            "soc": "Apple M4 Max",
            "vendor_sdk": null
          },
          "error": "It seems like your message might be incomplete. Could you please provide more details or clarify what you need assistance with? Whether it's a question, a topic you'd like to discuss, or something else, I'm here to help!",
          "evidence": [
            "command: litert-lm run LFM2-2.6B_int8.litertlm --prompt \"What is the capital of France?\" --cache no < /dev/null",
            "litert-lm 0.17.1 (the log's version line)",
            "stdout: It seems like your message might be incomplete. Could you please provide more details or clarify what you need assistance with? Whether it's a question, a topic you'd like to discuss, or something else, I'm here to help!",
            "### EXIT: 0",
            "2026-10-12 20:13:10 JST 0.17.1 run int8"
          ],
          "failure_class": "degenerate_output",
          "full_delegation": null,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "processes": 1.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": false,
          "total_ops": null,
          "ttft_ms": null
        },
        "source": "data/device_runs/0.17.1/2026-10-12/lfm2-2.6b-int8__mac-studio-m4-max.json"
      },
      {
        "device": "mac-studio-m4-max",
        "run": {
          "accelerator": "cpu",
          "context_length": null,
          "date": "2026-10-12",
          "decode_tokens_per_s": 39.18,
          "delegated_ops": null,
          "env": {
            "device": "Mac Studio (M4 Max)",
            "machine_label": "mac-studio-m4-max",
            "os_build": "macOS 27.0",
            "runtime": "litert-lm",
            "runtime_version": "0.18.0",
            "soc": "Apple M4 Max",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "Prefill 399.41 / decode 39.18 tok/s, TTFT 2.5896 s, init 30.2006 s",
            "settle_run: quiet after 9 s; run: litert-lm benchmark LFM2-2.6B_int8.litertlm --backend cpu -p 1024 -d 256 --max-num-tokens 2048 --runs 3 --cache no",
            "2026-10-12 18:58:51 JST cpu int8 exit 0"
          ],
          "failure_class": null,
          "full_delegation": null,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "decode_tokens": 256.0,
            "init_s": 30.2006,
            "iterations": 3.0,
            "max_num_tokens": 2048.0,
            "prefill_tokens": 1024.0,
            "processes": 1.0,
            "t_1024_256_s": 9.0977
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": 399.41,
          "provenance": "measured",
          "runs": true,
          "total_ops": null,
          "ttft_ms": 2589.6
        },
        "source": "data/device_runs/0.18.0/2026-10-12/lfm2-2.6b-int8__mac-studio-m4-max.json"
      },
      {
        "device": "mac-studio-m4-max",
        "run": {
          "accelerator": "gpu",
          "context_length": null,
          "date": "2026-10-12",
          "decode_tokens_per_s": 125.63,
          "delegated_ops": null,
          "env": {
            "device": "Mac Studio (M4 Max)",
            "machine_label": "mac-studio-m4-max",
            "os_build": "macOS 27.0",
            "runtime": "litert-lm",
            "runtime_version": "0.18.0",
            "soc": "Apple M4 Max",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "Prefill 2466.0 / decode 125.63 tok/s, TTFT 0.4232 s, init 2.6537 s",
            "settle_run: quiet after 37 s; run: litert-lm benchmark LFM2-2.6B_int8.litertlm --backend gpu -p 1024 -d 256 --max-num-tokens 2048 --runs 3 --cache no",
            "2026-10-12 19:10:32 JST gpu int8 verbose exit 0 validation_error=0 shape_mismatch=0 replacing=24",
            "2026-10-12 19:09:26 JST gpu int8 exit 0"
          ],
          "failure_class": null,
          "full_delegation": null,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "decode_tokens": 256.0,
            "init_s": 2.6537,
            "iterations": 3.0,
            "max_num_tokens": 2048.0,
            "prefill_tokens": 1024.0,
            "processes": 1.0,
            "t_1024_256_s": 2.453
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": 2466.0,
          "provenance": "measured",
          "runs": true,
          "total_ops": null,
          "ttft_ms": 423.2
        },
        "source": "data/device_runs/0.18.0/2026-10-12/lfm2-2.6b-int8__mac-studio-m4-max.json"
      },
      {
        "device": "pixel-8a",
        "run": {
          "accelerator": "cpu",
          "context_length": null,
          "date": "2026-10-12",
          "decode_tokens_per_s": 6.655,
          "delegated_ops": 1394,
          "env": {
            "device": "Pixel 8a",
            "machine_label": "pixel-8a-aab1",
            "os_build": "Android 16 (CP1A.260505.005)",
            "runtime": "litert-lm",
            "runtime_version": "prebuilt-adac974b",
            "soc": "Tensor G3",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "binary: Google's prebuilt litert_lm_advanced_main + 6 .so (gs://litert/binaries/latest/android_arm64/litert_lm/, object dated 2026-09-18, 40,382,152 B, sha256 adac974b…)",
            "r1 hold 2026-10-12 19:25:49-19:52:57 JST: [19:27:00] bundle on device: 3dda920ce47e5ef2597c1a43732daeecc5d4b9b91fe29599cb9eaafb874b29e8 (2723809744 B, push 58.4 s) = Mac",
            "bench args: --backend=cpu --model_path=/data/local/tmp/aa3_26b/LFM2-2.6B_int8.litertlm --benchmark=true --benchmark_prefill_tokens=1024 --benchmark_decode_tokens=256 --max_num_tokens=2048 --num_iterations=3 --report_peak_memory_footprint=true --metric_proto_file_path=/data/local/tmp/aa3_26b/metrics.pb",
            "VERBOSE: Replacing 869 out of 978 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 152 partitions for subgraph 0 (prefill_1024).",
            "VERBOSE: Replacing 795 out of 895 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 156 partitions for subgraph 11 (decode).",
            "VERBOSE: Replacing 1394 out of 1428 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 18 partitions for subgraph 0 (prefill_1024).",
            "VERBOSE: Replacing 1273 out of 1296 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 18 partitions for subgraph 11 (decode).",
            "[19:31:58] parsed: p8a_i8_cpu_timing cpu: gate PASS 1/1 | bench prefill 87.335 decode 6.654999999999999 init 727.12 cold 12240.92 peak 3647.4 capped 40/48 | answers: 'The capital of France is Paris.'",
            "timing run prompt process: \"The capital of France is Paris.\"",
            "check run (hold 2026-10-12 19:25:49-19:52:57 JST): 1 of 1 prompt processes answered with the expected word (\"The capital of France is Paris.\")"
          ],
          "failure_class": null,
          "full_delegation": false,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "boot_epoch_s": 1791442099.87,
            "capped_samples": 40.0,
            "cold_init_ms": 12240.92,
            "cpu_capped_samples": 40.0,
            "decode_tok_s_it1": 8.8,
            "decode_tok_s_it2": 7.3,
            "decode_tok_s_it3": 6.01,
            "decode_tokens": 256.0,
            "init_ms": 727.12,
            "iterations": 3.0,
            "max_num_tokens": 2048.0,
            "memavail_at_bench_start_kib": 4028148.0,
            "memavail_min_kib": 2953380.0,
            "prefill_tok_s_it1": 106.42,
            "prefill_tok_s_it2": 93.07,
            "prefill_tok_s_it3": 81.6,
            "prefill_tokens": 1024.0,
            "sampler_samples": 48.0,
            "t_1024_256_s": 50.1923,
            "thermal_status_at_bench_start": 0.0,
            "ttft_s_it1": 9.74,
            "ttft_s_it2": 11.14,
            "ttft_s_it3": 12.72
          },
          "output_match": null,
          "peak_mem_mb": 3647.4,
          "prefill_tokens_per_s": 87.335,
          "provenance": "measured",
          "runs": true,
          "total_ops": 1428,
          "ttft_ms": 11930.0
        },
        "source": "data/device_runs/prebuilt-adac974b/2026-10-12/lfm2-2.6b-int8__pixel-8a.json"
      },
      {
        "device": "pixel-8a",
        "run": {
          "accelerator": "gpu",
          "context_length": null,
          "date": "2026-10-12",
          "decode_tokens_per_s": 11.105,
          "delegated_ops": 978,
          "env": {
            "device": "Pixel 8a",
            "machine_label": "pixel-8a-aab1",
            "os_build": "Android 16 (CP1A.260505.005)",
            "runtime": "litert-lm",
            "runtime_version": "prebuilt-adac974b",
            "soc": "Tensor G3",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "binary: Google's prebuilt litert_lm_advanced_main + 6 .so (gs://litert/binaries/latest/android_arm64/litert_lm/, object dated 2026-09-18, 40,382,152 B, sha256 adac974b…)",
            "r4 hold 2026-10-12 21:06:18-21:25:50 JST: [21:17:22] bundle on device: 3dda920ce47e5ef2597c1a43732daeecc5d4b9b91fe29599cb9eaafb874b29e8 (2723809744 B, push 0 s) = Mac",
            "bench args: --backend=gpu --model_path=/data/local/tmp/aa3_26b/LFM2-2.6B_int8.litertlm --benchmark=true --benchmark_prefill_tokens=1024 --benchmark_decode_tokens=256 --max_num_tokens=2048 --num_iterations=3 --report_peak_memory_footprint=true --metric_proto_file_path=/data/local/tmp/aa3_26b/metrics.pb",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 0 (prefill_1024).",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 1 (prefill_512).",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 2 (prefill_256).",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 3 (prefill_128).",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 4 (prefill_64).",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 5 (prefill_32).",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 6 (prefill_16).",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 7 (prefill_8).",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 8 (prefill_4).",
            "VERBOSE: Replacing 978 out of 978 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 9 (prefill_2).",
            "VERBOSE: Replacing 877 out of 877 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 10 (prefill_1).",
            "VERBOSE: Replacing 895 out of 895 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 11 (decode).",
            "[21:25:47] parsed: p8a_i8_gpu_timing gpu: gate PASS 1/1 | bench prefill 504.39 decode 11.105 init 12549.64 cold 116225.34 peak 1654.26 capped 0/34 | answers: 'The capital of France is Paris.'",
            "timing run prompt process: \"The capital of France is Paris.\"",
            "check run (hold 2026-10-12 21:06:18-21:25:50 JST): 1 of 1 prompt processes answered with the expected word (\"The capital of France is Paris.\")"
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "boot_epoch_s": 1791442099.71,
            "capped_samples": 0.0,
            "cold_init_ms": 116225.34,
            "cpu_capped_samples": 21.0,
            "decode_tok_s_it1": 10.84,
            "decode_tok_s_it2": 11.25,
            "decode_tok_s_it3": 10.96,
            "decode_tokens": 256.0,
            "init_ms": 12549.64,
            "iterations": 3.0,
            "max_num_tokens": 2048.0,
            "memavail_at_bench_start_kib": 4268892.0,
            "memavail_min_kib": 419100.0,
            "prefill_tok_s_it1": 513.2,
            "prefill_tok_s_it2": 507.7,
            "prefill_tok_s_it3": 501.08,
            "prefill_tokens": 1024.0,
            "sampler_samples": 34.0,
            "t_1024_256_s": 25.0829,
            "thermal_status_at_bench_start": 1.0,
            "ttft_s_it1": 2.09,
            "ttft_s_it2": 2.11,
            "ttft_s_it3": 2.13
          },
          "output_match": null,
          "peak_mem_mb": 1654.26,
          "prefill_tokens_per_s": 504.39,
          "provenance": "measured",
          "runs": true,
          "total_ops": 978,
          "ttft_ms": 2120.0
        },
        "source": "data/device_runs/prebuilt-adac974b/2026-10-12/lfm2-2.6b-int8__pixel-8a.json"
      }
    ]
  },
  "model": {
    "family": "lfm2",
    "id": "lfm2-2.6b-int8",
    "license": "other (lfm1.0)",
    "source_url": "https://huggingface.co/litert-community/LFM2-2.6B",
    "task": "text-generation"
  },
  "pitfalls": [
    "Runtime: tested on litert-lm 0.18.0 on a Mac; on litert-lm 0.17.1 this file answered \"What is the capital of France?\" as if no question had been asked (one CPU run) (card L21).",
    "Reasoning trace: a trace between <think> and </think> goes to the thought channel (the CLI prints it between [thought] and [/thought]); traces appeared for \"What is the capital of France?\" (this file, CPU and GPU) and 1 of 120 IFEval-lite prompts, none for GSM8K or the checks; --thinking-budget 0 and 32 did not suppress the trace on that prompt (card L71).",
    "Template: Liquid AI's chat template fails at render on litert-lm 0.18.0 (map has no method named get), so the bundle carries LiteRT-LM's lfm2 template (byte-identical to models/lfm2/chat_template.jinja at commit 86a7e6d0), which reads message content as a list of parts; stops <|im_end|> and <|endoftext|>, start token <|startoftext|> (card L71, L112).",
    "Tool calling is declared (model type lfm2): the file returned a parsed get_weather call (Python API, CPU, greedy) and used its result with automatic tool calling; with tools the template omits Liquid AI's tool-list and tool-response markers, wraps the tool result in [ … ] and quotes call arguments with double quotes (card L73).",
    "One Engine per conversation: in a probe with two different system prompts a shared Engine gave the same factual answers as fresh Engines (3 runs each); LiteRT-LM issue #3165 reports recurrent state reused across conversations that share prompt tokens, and the probe's two conversations started with different system prompts; the card advises a new Engine for each conversation (card L75).",
    "Conversation preface with tools and no messages: LiteRT-LM renders the chat template with an empty message list, and the template reads messages[0] without checking that the list is non-empty, so the render fails with 'Failed to apply template: undefined value' when the preface is prefilled at creation (prefillPrefaceOnInit, default false; creating the conversation fails) and in the preface render (renderPrefaceIntoString and its C++ and C API forms); the same with extra context (such as enable_thinking) and no messages, and in the preface render with an empty preface (run in the runtime on one Qwen3 template; the other templates rendered with minijinja 2.14 and jinja2); read in the source of LiteRT-LM 0.17.0, 0.18.1 and main, measured with the 0.18.1 PyPI package through its C API on these files' template text; the default path, where the first message is rendered with the preface, is not affected (the 0.18.1 Python API and CLI expose neither the flag nor the preface render); a system message or any initial message in the conversation config avoids it (card L116).",
    "Sampler: the bundle's default sampler is temperature 0.1, top-k 50, top-p 1.0; Liquid AI recommends temperature 0.3, min_p 0.15 and repetition_penalty 1.05, which the bundle does not carry; --top-k 1 gives greedy output (card L69).",
    "Cache: without --cache, `run` writes a compiled-weights cache next to the model (2.58 GB for this file on CPU); without --backend, `run` uses the CPU (card L67).",
    "Speed: int4 is the faster file on every GPU and this file on every CPU (Mac 9.10 vs 10.75 s, Galaxy S26 18.17 vs 23.54 s, Pixel 8a 50.17 vs 68.48 s for a 1024-token prefill plus 256 decoded tokens); not run on the NPU (card L32-L39, L106).",
    "iOS not measured: this file's one weights section is 2,722,843,088 bytes (2.54 GiB); the LFM2.5-2.6B card reports an int8 file with a 2.87 GB weights section running on an iPhone 17 Pro only with the extended-virtual-addressing and increased-memory entitlements (card L39).",
    "Phone conditions: on the Galaxy S26 the GPU clock ceiling fell from 1300 to 902 MHz inside this file's GPU benchmark and the CPU clock was lowered in 24 of 25 samples of its CPU run; on the Pixel 8a the CPU clock was lowered in 40 of 48 samples of the CPU run, and in the GPU benchmark the lowest MemAvailable sample was 419,100 kB (card L97, L106).",
    "Accuracy (greedy, at most 2048 output tokens, litert-lm on the Mac GPU, comparable only within the card's table): GSM8K 89 of 100 (bf16 88), IFEval-lite 85 of 120 (bf16 85); 8-question check 7/8 on CPU and GPU, bf16 7/8 (card L43, L49-L57).",
    "License: the LFM Open License v1.0's section 5 conditions commercial use on an annual revenue below 10 million US dollars (card L120)."
  ],
  "schema_version": "1.2"
}
