{
  "artifacts": [
    {
      "file": "mnn-pocket-0.3b-dec_L1024_fp16w.tflite",
      "sha256": "a76354c9b14333d4d55385c4d0cd86b3bfe8a6df30a99a9963b138ea71a8397c",
      "size_mb": 466.233
    }
  ],
  "benchmarks": [],
  "conversion": {
    "command": "the repository's conversion/ (conversion/README.md) holds the scripts and REPRODUCE.md every check (README L150); the card names the tools, not a command line",
    "quantization": "float16 linear-layer weights, the LoRA adapter folded into the weights first (README L20)",
    "tool": "litert-torch",
    "tool_version": "0.10.0 (for LiteRT 2.3.0)"
  },
  "cross_runtime": [],
  "delegation": null,
  "device": {
    "records": [
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "cpu_xnnpack",
          "context_length": null,
          "date": "2026-11-07",
          "decode_tokens_per_s": null,
          "delegated_ops": 2376,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-mnnpocketdec-capi-runner",
            "os_build": "Android 16 (BP4A.251205.006.S942QOPS1AZH9)",
            "runtime": "litert",
            "runtime_version": "2.3.0",
            "soc": "Qualcomm SM8850",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "(tag s26_cpu4_L1024, run r2): run DONE rc 0, start 2026-11-07 19:14:51, 127 questions",
            "Push mnn_pocket_dec_ns4_in_6-27_post_6-27_final_L1024_w16.tflite rc=0 bytes 488880528/488880528 sha256 phone=a76354c9b14333d4d55385c4d0cd86b3bfe8a6df30a99a9963b138ea71a8397c = mac = the Hub LFS sha256",
            "Runner 7e7539e5bd0c73ed… and libs: 12/12 OK",
            "Device: SM-S942Q 16 BP4A.251205.006.S942QOPS1AZH9",
            "The build 7e7539e5 asked for 4 threads (cpu_toml \"num_threads = 4\") and its calls ran on threads_in_calls_max 1; the corrected build ac2be1d3 at 4: 4. This run is entered as 1 thread",
            "runner.replacing: VERBOSE: Replacing 2376 out of 2432 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 57 partitions for subgraph 0 (main).",
            "line PASS: 127 questions, same winner 122/122 (not near ties), near ties kept 5/5, max |Δp| 0.000908779, mean |Δp| 9.71328e-05, control n/a (the file holds no request of the control pair), non-finite read-outs 0, winner = the Mac CPU's on the same file 122/122",
            "Run memory (22 samples): VmHWM_max_kB 1671268 VmSwap_max_kB 0 MemAvailable_min_kB 5970568 kgsl_page_alloc_max_B 219365376"
          ],
          "failure_class": null,
          "full_delegation": false,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "arm_files_passed": 8.0,
            "arm_files_run": 8.0,
            "compile_ms": 569.2,
            "line_bar_control_abs_dp": 0.02,
            "line_bar_max_abs_dp": 0.02,
            "line_bar_mean_abs_dp": 0.002,
            "line_bar_near_tie_gap": 0.02,
            "line_mac_same_file_non_near_tie": 122.0,
            "line_max_abs_dp": 0.0009087786688101729,
            "line_mean_abs_dp": 9.713278120176597e-05,
            "line_near_tie": 5.0,
            "line_non_near_tie": 122.0,
            "line_nonfinite_questions": 0.0,
            "line_pass": 1.0,
            "line_questions": 127.0,
            "line_winner_equal_mac_non_near_tie": 122.0,
            "line_winner_equal_near_tie": 5.0,
            "line_winner_equal_non_near_tie": 122.0,
            "threads": 1.0,
            "vmhwm_max_kb": 1671268.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "the line: 127 questions of the L1024 set against the source model's reference; the C-API runner build 7e7539e5 (LiteRT 2.3.0), CPU XNNPACK, 1 thread (this build passed the thread count under an option name LiteRT 2.3.0 does not read)",
          "total_ops": 2432,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.3.0/2026-11-07/mnn-pocket-0.3b-dec__mnn-pocket-0.3b-dec_l1024_fp16w__galaxy-s26.json"
      },
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "gpu_mldrift",
          "context_length": null,
          "date": "2026-11-07",
          "decode_tokens_per_s": null,
          "delegated_ops": 2432,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-mnnpocketdec-capi-runner",
            "os_build": "Android 16 (BP4A.251205.006.S942QOPS1AZH9)",
            "runtime": "litert",
            "runtime_version": "2.3.0",
            "soc": "Qualcomm SM8850",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "(tag s26_gpu_fp16acc32_L1024, run r2): run DONE rc 0, start 2026-11-07 19:17:43, 127 questions",
            "Push mnn_pocket_dec_ns4_in_6-27_post_6-27_final_L1024_w16.tflite rc=0 bytes 488880528/488880528 sha256 phone=a76354c9b14333d4d55385c4d0cd86b3bfe8a6df30a99a9963b138ea71a8397c = mac = the Hub LFS sha256",
            "Runner 7e7539e5bd0c73ed… and libs: 12/12 OK",
            "Device: SM-S942Q 16 BP4A.251205.006.S942QOPS1AZH9",
            "runner.replacing: VERBOSE: Replacing 2432 out of 2432 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 0 (main).",
            "line PASS: 127 questions, same winner 122/122 (not near ties), near ties kept 3/5, max |Δp| 0.00736313, mean |Δp| 0.000843399, control n/a (the file holds no request of the control pair), non-finite read-outs 0, winner = the Mac CPU's on the same file 122/122",
            "Run memory (8 samples): VmHWM_max_kB 2067588 VmSwap_max_kB 0 MemAvailable_min_kB 5552040 kgsl_page_alloc_max_B 836247552"
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "arm_files_passed": 8.0,
            "arm_files_run": 8.0,
            "compile_ms": 3770.8,
            "line_bar_control_abs_dp": 0.02,
            "line_bar_max_abs_dp": 0.02,
            "line_bar_mean_abs_dp": 0.002,
            "line_bar_near_tie_gap": 0.02,
            "line_mac_same_file_non_near_tie": 122.0,
            "line_max_abs_dp": 0.007363131112748944,
            "line_mean_abs_dp": 0.0008433992140805586,
            "line_near_tie": 5.0,
            "line_non_near_tie": 122.0,
            "line_nonfinite_questions": 0.0,
            "line_pass": 1.0,
            "line_questions": 127.0,
            "line_winner_equal_mac_non_near_tie": 122.0,
            "line_winner_equal_near_tie": 3.0,
            "line_winner_equal_non_near_tie": 122.0,
            "vmhwm_max_kb": 2067588.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "the line: 127 questions of the L1024 set against the source model's reference; the C-API runner build 7e7539e5 (LiteRT 2.3.0), GPU OpenCL, GpuOptions precision FP16_WITH_FP32_ACCUM",
          "total_ops": 2432,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.3.0/2026-11-07/mnn-pocket-0.3b-dec__mnn-pocket-0.3b-dec_l1024_fp16w__galaxy-s26.json"
      },
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "gpu_mldrift",
          "context_length": null,
          "date": "2026-11-07",
          "decode_tokens_per_s": null,
          "delegated_ops": 2432,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-mnnpocketdec-capi-runner",
            "os_build": "Android 16 (BP4A.251205.006.S942QOPS1AZH9)",
            "runtime": "litert",
            "runtime_version": "2.3.0",
            "soc": "Qualcomm SM8850",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "(tag s26_gpu_fp32_L1024, run r2): run DONE rc 0, start 2026-11-07 19:11:58, 127 questions",
            "Push mnn_pocket_dec_ns4_in_6-27_post_6-27_final_L1024_w16.tflite rc=0 bytes 488880528/488880528 sha256 phone=a76354c9b14333d4d55385c4d0cd86b3bfe8a6df30a99a9963b138ea71a8397c = mac = the Hub LFS sha256",
            "Runner 7e7539e5bd0c73ed… and libs: 12/12 OK",
            "Device: SM-S942Q 16 BP4A.251205.006.S942QOPS1AZH9",
            "runner.replacing: VERBOSE: Replacing 2432 out of 2432 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 0 (main).",
            "line PASS: 127 questions, same winner 122/122 (not near ties), near ties kept 5/5, max |Δp| 0.000909005, mean |Δp| 9.72565e-05, control n/a (the file holds no request of the control pair), non-finite read-outs 0, winner = the Mac CPU's on the same file 122/122",
            "Run memory (12 samples): VmHWM_max_kB 2543540 VmSwap_max_kB 0 MemAvailable_min_kB 5070024 kgsl_page_alloc_max_B 1389228032"
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "arm_files_passed": 8.0,
            "arm_files_run": 8.0,
            "compile_ms": 3058.8,
            "line_bar_control_abs_dp": 0.02,
            "line_bar_max_abs_dp": 0.02,
            "line_bar_mean_abs_dp": 0.002,
            "line_bar_near_tie_gap": 0.02,
            "line_mac_same_file_non_near_tie": 122.0,
            "line_max_abs_dp": 0.0009090049141375345,
            "line_mean_abs_dp": 9.725645663651394e-05,
            "line_near_tie": 5.0,
            "line_non_near_tie": 122.0,
            "line_nonfinite_questions": 0.0,
            "line_pass": 1.0,
            "line_questions": 127.0,
            "line_winner_equal_mac_non_near_tie": 122.0,
            "line_winner_equal_near_tie": 5.0,
            "line_winner_equal_non_near_tie": 122.0,
            "vmhwm_max_kb": 2543540.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "the line: 127 questions of the L1024 set against the source model's reference; the C-API runner build 7e7539e5 (LiteRT 2.3.0), GPU OpenCL, GpuOptions precision FP32",
          "total_ops": 2432,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.3.0/2026-11-07/mnn-pocket-0.3b-dec__mnn-pocket-0.3b-dec_l1024_fp16w__galaxy-s26.json"
      },
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "gpu_mldrift",
          "context_length": null,
          "date": "2026-11-08",
          "decode_tokens_per_s": null,
          "delegated_ops": 2432,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-mnnpocketdec-capi-runner",
            "os_build": "Android 16 (BP4A.251205.006.S942QOPS1AZH9)",
            "runtime": "litert",
            "runtime_version": "2.3.0",
            "soc": "Qualcomm SM8850",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "371.57 ms (min 367.32, max 373.81; median of 8 after 2 warm-ups, write+run+readback; load 1.32 s; capped 1/6 samples)… (the cell's value; parity cell: PASS; gate PASS, timing run = check run in every digit (V3 a); gate PASS: 122/122)",
            "(tag s26_t_gpu_fp16acc32_c_L1024_S, run r5): start 2026-11-08 03:09:22, request S/c_own_long (1024 tokens), write_run_read_ms median 371.5675 min 367.323 max 373.812 n 8, run_only_ms median 1.1815",
            "Push mnn_pocket_dec_ns4_in_6-27_post_6-27_final_L1024_w16.tflite rc=0 bytes 488880528/488880528 sha256 phone=a76354c9b14333d4d55385c4d0cd86b3bfe8a6df30a99a9963b138ea71a8397c = mac = the Hub LFS sha256",
            "Runner 7e7539e5bd0c73ed… and libs: 12/12 OK; extras: rest=180 sample=1",
            "Device: SM-S942Q 16 BP4A.251205.006.S942QOPS1AZH9",
            "clock samples: 6 samples, 1 capped (policy6 4339200/4742400), GPU level max 0",
            "parity: the timed request's 7 question(s) have the same probabilities in every digit as in the check run (line PASS)",
            "VERBOSE: Replacing 2432 out of 2432 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 0 (main).",
            "the cell's takes (take id: median, capped): 109: 381.8 ms capped; 110: 384.7 ms capped; 505: 371.6 ms capped; the cell is take 505, the fastest take (every take capped)",
            "other rows of this file, request and backend, not takes of this cell: 105: GPU precision fp16; 106: GPU precision fp16; 114: GPU precision fp32; 115: GPU precision fp32; 525: GPU precision fp32"
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": 371.5675,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "arm_files_passed": 8.0,
            "arm_files_run": 8.0,
            "capped": 1.0,
            "capped_samples": 1.0,
            "clock_samples": 6.0,
            "compile_ms": 1322.3,
            "latency_max_ms": 373.812,
            "latency_min_ms": 367.323,
            "request_questions": 3.0,
            "request_tokens": 1024.0,
            "run_only_p50_ms": 1.1815,
            "timed_calls": 8.0,
            "warmup_calls": 2.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "timing: `long_request` (the repo's 1,024-token example) (1,024 tokens, 3 questions), the median of 8 calls after 2 warm-up calls in one process, a call = write the four inputs, run, read `hidden` back; the C-API runner build 7e7539e5 (LiteRT 2.3.0), GPU OpenCL, GpuOptions precision FP16_WITH_FP32_ACCUM",
          "total_ops": 2432,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.3.0/2026-11-08/mnn-pocket-0.3b-dec__mnn-pocket-0.3b-dec_l1024_fp16w__galaxy-s26.json"
      },
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "gpu_mldrift",
          "context_length": null,
          "date": "2026-11-08",
          "decode_tokens_per_s": null,
          "delegated_ops": 2432,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-mnnpocketdec-capi-runner",
            "os_build": "Android 16 (BP4A.251205.006.S942QOPS1AZH9)",
            "runtime": "litert",
            "runtime_version": "2.3.0",
            "soc": "Qualcomm SM8850",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "530.28 ms (min 528.04, max 538.20; median of 8 after 2 warm-ups, write+run+readback; load 1.65 s; capped 2/8 samples)… (the cell's value; parity cell: PASS; gate PASS, timing run = check run in every digit (V3 a); gate PASS: 122/122)",
            "(tag s26_t_gpu_fp32_c_L1024_S, run r5): start 2026-11-08 03:06:11, request S/c_own_long (1024 tokens), write_run_read_ms median 530.277 min 528.039 max 538.201 n 8, run_only_ms median 0.9159999999999999",
            "Push mnn_pocket_dec_ns4_in_6-27_post_6-27_final_L1024_w16.tflite rc=0 bytes 488880528/488880528 sha256 phone=a76354c9b14333d4d55385c4d0cd86b3bfe8a6df30a99a9963b138ea71a8397c = mac = the Hub LFS sha256",
            "Runner 7e7539e5bd0c73ed… and libs: 12/12 OK; extras: rest=180 sample=1",
            "Device: SM-S942Q 16 BP4A.251205.006.S942QOPS1AZH9",
            "clock samples: 8 samples, 2 capped (policy0 3513600/3628800, policy6 4262400/4742400), GPU level max 0",
            "parity: the timed request's 7 question(s) have the same probabilities in every digit as in the check run (line PASS)",
            "VERBOSE: Replacing 2432 out of 2432 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 0 (main).",
            "the cell's takes (take id: median, capped): 114: 539.6 ms capped; 115: 536 ms capped; 525: 530.3 ms capped; the cell is take 525, the fastest take (every take capped)",
            "other rows of this file, request and backend, not takes of this cell: 105: GPU precision fp16; 106: GPU precision fp16; 109: GPU precision fp16acc32; 110: GPU precision fp16acc32; 505: GPU precision fp16acc32"
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": 530.277,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "arm_files_passed": 8.0,
            "arm_files_run": 8.0,
            "capped": 1.0,
            "capped_samples": 2.0,
            "clock_samples": 8.0,
            "compile_ms": 1645.9,
            "latency_max_ms": 538.201,
            "latency_min_ms": 528.039,
            "request_questions": 3.0,
            "request_tokens": 1024.0,
            "run_only_p50_ms": 0.9159999999999999,
            "timed_calls": 8.0,
            "warmup_calls": 2.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "timing: `long_request` (the repo's 1,024-token example) (1,024 tokens, 3 questions), the median of 8 calls after 2 warm-up calls in one process, a call = write the four inputs, run, read `hidden` back; the C-API runner build 7e7539e5 (LiteRT 2.3.0), GPU OpenCL, GpuOptions precision FP32",
          "total_ops": 2432,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.3.0/2026-11-08/mnn-pocket-0.3b-dec__mnn-pocket-0.3b-dec_l1024_fp16w__galaxy-s26.json"
      },
      {
        "device": "mac-studio-m4-max",
        "run": {
          "accelerator": "cpu_xnnpack",
          "context_length": null,
          "date": "2026-11-07",
          "decode_tokens_per_s": null,
          "delegated_ops": null,
          "env": {
            "device": "Mac Studio (M4 Max, Mac16,9)",
            "machine_label": "mac-studio-m4-max",
            "os_build": "macOS 27.0 (26A428)",
            "runtime": "litert",
            "runtime_version": "2.3.0",
            "soc": "Apple M4 Max",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "Name mnn_pocket_dec_ns4_in_6-27_post_6-27_final_L1024_w16, arm cpu4, child rc 0",
            "Runtime: api CompiledModel.from_file, accelerator CPU (XNNPACK), threads 4, is_fully_accelerated False (the API states no node count), compile 0.71 s",
            "date: the record's first commit, author date 2026-11-07 18:49:02 +0900, after the run's START 18:19:01 (the run log has times only and is not in git)",
            "line PASS: 2400 questions, same winner 2314/2314 (not near ties), near ties kept 86/86, max |Δp| 0.00209024, mean |Δp| 0.000110304, control 0.0401, non-finite read-outs 0"
          ],
          "failure_class": null,
          "full_delegation": false,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "arm_files_passed": 20.0,
            "arm_files_run": 20.0,
            "compile_s": 0.71,
            "line_bar_control_abs_dp": 0.02,
            "line_bar_max_abs_dp": 0.02,
            "line_bar_mean_abs_dp": 0.002,
            "line_bar_near_tie_gap": 0.02,
            "line_control_abs_dp": 0.04007461881503663,
            "line_max_abs_dp": 0.002090235550811559,
            "line_mean_abs_dp": 0.00011030421194566346,
            "line_near_tie": 86.0,
            "line_non_near_tie": 2314.0,
            "line_nonfinite_questions": 0.0,
            "line_pass": 1.0,
            "line_questions": 2400.0,
            "line_winner_equal_near_tie": 86.0,
            "line_winner_equal_non_near_tie": 2314.0,
            "threads": 4.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "the line: 2400 questions (the requests that fit L1024) against the source model's reference; Python CompiledModel (ai-edge-litert 2.3.0), CPU XNNPACK, 4 threads",
          "total_ops": null,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.3.0/2026-11-07/mnn-pocket-0.3b-dec__mnn-pocket-0.3b-dec_l1024_fp16w__mac-studio-m4-max.json"
      },
      {
        "device": "mac-studio-m4-max",
        "run": {
          "accelerator": "cpu_xnnpack",
          "context_length": null,
          "date": "2026-11-07",
          "decode_tokens_per_s": null,
          "delegated_ops": null,
          "env": {
            "device": "Mac Studio (M4 Max, Mac16,9)",
            "machine_label": "mac-studio-m4-max",
            "os_build": "macOS 27.0 (26A428)",
            "runtime": "litert",
            "runtime_version": "2.3.0",
            "soc": "Apple M4 Max",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "380.9 ms (the cell's value; parity cell: PASS 2314/2314 non-near-tie, max|Δp| 0.0021, mean 0.00011, red arm 0.0401 (2400 )",
            "(tag w16_fit, started 2026-11-07 18:50:27 JST, rule \"line L6: 5 warm-ups + median of 20, host write + run + read back; host ms apart\", ai-edge-litert 2.3.0): row file mnn_pocket_dec_ns4_in_6-27_post_6-27_final_L1024_w16, arm cpu4, request c_own_long, ms_write_run_read median 380.9292295045452 min 378.41670800116844 max 1103.4815839957446 p90 432.90037908009316 n 20, winners_equal_R True",
            "Reads (passes no --r1-graphs), the file of the gate record (sha256 a76354c9b14333d4d55385c4d0cd86b3bfe8a6df30a99a9963b138ea71a8397c = the Hub LFS entry); the file still holds those bytes, modified 2026-11-07 17:44:18 +0900, before this run",
            "the line on this file and arm, line PASS: 2400 questions, same winner 2314/2314 (not near ties), near ties kept 86/86, max |Δp| 0.00209024, mean |Δp| 0.000110304, control 0.0401, non-finite read-outs 0",
            "other rows of this file, request and backend, not takes of this cell: 21: arm cpu8; 24: arm cpu12; 27: arm cpu16"
          ],
          "failure_class": null,
          "full_delegation": false,
          "latency_p50_ms": 380.9292295045452,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "arm_files_passed": 20.0,
            "arm_files_run": 20.0,
            "compile_s": 0.7,
            "latency_max_ms": 1103.4815839957446,
            "latency_min_ms": 378.41670800116844,
            "latency_p90_ms": 432.90037908009316,
            "request_questions": 3.0,
            "request_tokens": 1024.0,
            "threads": 4.0,
            "timed_calls": 20.0,
            "warmup_calls": 5.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "timing: `long_request` (the repo's 1,024-token example) (1,024 tokens, 3 questions), the median of 20 calls after 5 warm-up calls in one process, a call = write the four inputs, run, read `hidden` back; Python CompiledModel (ai-edge-litert 2.3.0), CPU XNNPACK, 4 threads",
          "total_ops": null,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.3.0/2026-11-07/mnn-pocket-0.3b-dec__mnn-pocket-0.3b-dec_l1024_fp16w__mac-studio-m4-max.json"
      },
      {
        "device": "mac-studio-m4-max",
        "run": {
          "accelerator": "gpu_metal_mac",
          "context_length": null,
          "date": "2026-11-07",
          "decode_tokens_per_s": null,
          "delegated_ops": null,
          "env": {
            "device": "Mac Studio (M4 Max, Mac16,9)",
            "machine_label": "mac-studio-m4-max",
            "os_build": "macOS 27.0 (26A428)",
            "runtime": "litert",
            "runtime_version": "2.3.0",
            "soc": "Apple M4 Max",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "Name mnn_pocket_dec_ns4_in_6-27_post_6-27_final_L1024_w16, arm gpu_f32, child rc 0",
            "Runtime: api CompiledModel.from_file, accelerator GPU (Metal), gpu_options {\"enforce_f32\": true}, is_fully_accelerated True (the API states no node count), compile 2.94 s",
            "date: the record's first commit, author date 2026-11-07 18:49:02 +0900, after the run's START 18:11:54 (the run log has times only and is not in git)",
            "line PASS: 2400 questions, same winner 2314/2314 (not near ties), near ties kept 86/86, max |Δp| 0.00208065, mean |Δp| 0.000110316, control 0.0401, non-finite read-outs 0"
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "arm_files_passed": 20.0,
            "arm_files_run": 20.0,
            "compile_s": 2.94,
            "line_bar_control_abs_dp": 0.02,
            "line_bar_max_abs_dp": 0.02,
            "line_bar_mean_abs_dp": 0.002,
            "line_bar_near_tie_gap": 0.02,
            "line_control_abs_dp": 0.040074430422127605,
            "line_max_abs_dp": 0.002080653409673283,
            "line_mean_abs_dp": 0.00011031586296795013,
            "line_near_tie": 86.0,
            "line_non_near_tie": 2314.0,
            "line_nonfinite_questions": 0.0,
            "line_pass": 1.0,
            "line_questions": 2400.0,
            "line_winner_equal_near_tie": 86.0,
            "line_winner_equal_non_near_tie": 2314.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "the line: 2400 questions (the requests that fit L1024) against the source model's reference; Python CompiledModel (ai-edge-litert 2.3.0), GPU Metal, GpuOptions(enforce_f32=True)",
          "total_ops": null,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.3.0/2026-11-07/mnn-pocket-0.3b-dec__mnn-pocket-0.3b-dec_l1024_fp16w__mac-studio-m4-max.json"
      },
      {
        "device": "mac-studio-m4-max",
        "run": {
          "accelerator": "gpu_metal_mac",
          "context_length": null,
          "date": "2026-11-07",
          "decode_tokens_per_s": null,
          "delegated_ops": null,
          "env": {
            "device": "Mac Studio (M4 Max, Mac16,9)",
            "machine_label": "mac-studio-m4-max",
            "os_build": "macOS 27.0 (26A428)",
            "runtime": "litert",
            "runtime_version": "2.3.0",
            "soc": "Apple M4 Max",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "90.2 ms (the cell's value; parity cell: PASS 2314/2314 non-near-tie, max|Δp| 0.0021, mean 0.00011, red arm 0.0401 (2400 )",
            "(tag w16_fit, started 2026-11-07 18:50:27 JST, rule \"line L6: 5 warm-ups + median of 20, host write + run + read back; host ms apart\", ai-edge-litert 2.3.0): row file mnn_pocket_dec_ns4_in_6-27_post_6-27_final_L1024_w16, arm gpu_f32, request c_own_long, ms_write_run_read median 90.18472899333574 min 89.37354100635275 max 90.88108298601583 p90 90.55920829705428 n 20, winners_equal_R True",
            "Reads (passes no --r1-graphs), the file of the gate record (sha256 a76354c9b14333d4d55385c4d0cd86b3bfe8a6df30a99a9963b138ea71a8397c = the Hub LFS entry); the file still holds those bytes, modified 2026-11-07 17:44:18 +0900, before this run",
            "the line on this file and arm, line PASS: 2400 questions, same winner 2314/2314 (not near ties), near ties kept 86/86, max |Δp| 0.00208065, mean |Δp| 0.000110316, control 0.0401, non-finite read-outs 0",
            "other rows of this file, request and backend, not takes of this cell: 9: arm gpu_fp16"
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": 90.18472899333574,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "arm_files_passed": 20.0,
            "arm_files_run": 20.0,
            "compile_s": 2.4,
            "latency_max_ms": 90.88108298601583,
            "latency_min_ms": 89.37354100635275,
            "latency_p90_ms": 90.55920829705428,
            "request_questions": 3.0,
            "request_tokens": 1024.0,
            "timed_calls": 20.0,
            "warmup_calls": 5.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "timing: `long_request` (the repo's 1,024-token example) (1,024 tokens, 3 questions), the median of 20 calls after 5 warm-up calls in one process, a call = write the four inputs, run, read `hidden` back; Python CompiledModel (ai-edge-litert 2.3.0), GPU Metal, GpuOptions(enforce_f32=True)",
          "total_ops": null,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.3.0/2026-11-07/mnn-pocket-0.3b-dec__mnn-pocket-0.3b-dec_l1024_fp16w__mac-studio-m4-max.json"
      },
      {
        "device": "pixel-11-pro",
        "run": {
          "accelerator": "cpu_xnnpack",
          "context_length": null,
          "date": "2026-11-07",
          "decode_tokens_per_s": null,
          "delegated_ops": 2376,
          "env": {
            "device": "Pixel 11 Pro",
            "machine_label": "pixel-11-pro-mnnpocketdec-capi-runner",
            "os_build": "Android 17 (CD1A.260905.001.B1)",
            "runtime": "litert",
            "runtime_version": "2.3.0",
            "soc": "Google Tensor G6",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "(tag p11p_cpu4_L1024, run r2): run DONE rc 0, start 2026-11-07 19:16:55, 127 questions",
            "Push mnn_pocket_dec_ns4_in_6-27_post_6-27_final_L1024_w16.tflite rc=0 bytes 488880528/488880528 sha256 phone=a76354c9b14333d4d55385c4d0cd86b3bfe8a6df30a99a9963b138ea71a8397c = mac = the Hub LFS sha256",
            "Runner 7e7539e5bd0c73ed… and libs: 2/2 OK",
            "Device: Pixel 11 Pro 17 CD1A.260905.001.B1",
            "The build 7e7539e5 asked for 4 threads (cpu_toml \"num_threads = 4\") and its calls ran on threads_in_calls_max 1; the corrected build ac2be1d3 at 4: 4. This run is entered as 1 thread",
            "runner.replacing: VERBOSE: Replacing 2376 out of 2432 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 57 partitions for subgraph 0 (main).",
            "line PASS: 127 questions, same winner 122/122 (not near ties), near ties kept 5/5, max |Δp| 0.000908779, mean |Δp| 9.71328e-05, control n/a (the file holds no request of the control pair), non-finite read-outs 0, winner = the Mac CPU's on the same file 122/122",
            "Run memory (123 samples): VmHWM_max_kB 1665760 VmSwap_max_kB 0 MemAvailable_min_kB 8711880 kgsl_page_alloc_max_B "
          ],
          "failure_class": null,
          "full_delegation": false,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "arm_files_passed": 8.0,
            "arm_files_run": 8.0,
            "compile_ms": 1741.1,
            "line_bar_control_abs_dp": 0.02,
            "line_bar_max_abs_dp": 0.02,
            "line_bar_mean_abs_dp": 0.002,
            "line_bar_near_tie_gap": 0.02,
            "line_mac_same_file_non_near_tie": 122.0,
            "line_max_abs_dp": 0.0009087786688101729,
            "line_mean_abs_dp": 9.713278120176597e-05,
            "line_near_tie": 5.0,
            "line_non_near_tie": 122.0,
            "line_nonfinite_questions": 0.0,
            "line_pass": 1.0,
            "line_questions": 127.0,
            "line_winner_equal_mac_non_near_tie": 122.0,
            "line_winner_equal_near_tie": 5.0,
            "line_winner_equal_non_near_tie": 122.0,
            "threads": 1.0,
            "vmhwm_max_kb": 1665760.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "the line: 127 questions of the L1024 set against the source model's reference; the C-API runner build 7e7539e5 (LiteRT 2.3.0), CPU XNNPACK, 1 thread (this build passed the thread count under an option name LiteRT 2.3.0 does not read)",
          "total_ops": 2432,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.3.0/2026-11-07/mnn-pocket-0.3b-dec__mnn-pocket-0.3b-dec_l1024_fp16w__pixel-11-pro.json"
      },
      {
        "device": "pixel-11-pro",
        "run": {
          "accelerator": "cpu_xnnpack",
          "context_length": null,
          "date": "2026-11-08",
          "decode_tokens_per_s": null,
          "delegated_ops": 2376,
          "env": {
            "device": "Pixel 11 Pro",
            "machine_label": "pixel-11-pro-mnnpocketdec-capi-runner",
            "os_build": "Android 17 (CD1A.260905.001.B1)",
            "runtime": "litert",
            "runtime_version": "2.3.0",
            "soc": "Google Tensor G6",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "2871.72 ms (min 2753.15, max 4025.65; median of 8 after 2 warm-ups, write+run+readback; load 0.85 s; capped 6/34 samples… (the cell's value; parity cell: PASS; gate PASS, timing run = check run in every digit (V3 a); gate PASS: 122/122)",
            "(tag p11p_t_xt5_c_L1024_S, run r5): start 2026-11-08 00:08:36, request S/c_own_long (1024 tokens), write_run_read_ms median 2871.7195 min 2753.152 max 4025.653 n 8, run_only_ms median 2868.9785",
            "Push mnn_pocket_dec_ns4_in_6-27_post_6-27_final_L1024_w16.tflite rc=0 bytes 488880528/488880528 sha256 phone=a76354c9b14333d4d55385c4d0cd86b3bfe8a6df30a99a9963b138ea71a8397c = mac = the Hub LFS sha256",
            "Runner ac2be1d3f7b7a78f… and libs: 2/2 OK; extras: rest=180 sample=1 args=--cpu-threads,5",
            "Device: Pixel 11 Pro 17 CD1A.260905.001.B1",
            "clock samples: 34 samples, 6 capped (policy2 2899000/3379000, policy2 3148000/3379000, policy6 3898000/4109000, policy6 4013000/4109000), GPU level max 0",
            "parity: the timed request's 7 question(s) have the same probabilities in every digit as in the check run (line PASS)",
            "VERBOSE: Replacing 2376 out of 2432 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 57 partitions for subgraph 0 (main).",
            "the cell's takes (take id: median, capped): 480: 3407 ms capped; 482: 2872 ms capped; 489: 4553 ms capped; the cell is take 482, the fastest take (every take capped)",
            "other rows of this file, request and backend, not takes of this cell: 82: 1 thread; 83: 1 thread",
            "threads: the build ac2be1d3 passes --cpu-threads to LiteRT"
          ],
          "failure_class": null,
          "full_delegation": false,
          "latency_p50_ms": 2871.7195,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "arm_files_passed": 8.0,
            "arm_files_run": 8.0,
            "capped": 1.0,
            "capped_samples": 6.0,
            "clock_samples": 34.0,
            "compile_ms": 854.4,
            "latency_max_ms": 4025.653,
            "latency_min_ms": 2753.152,
            "request_questions": 3.0,
            "request_tokens": 1024.0,
            "run_only_p50_ms": 2868.9785,
            "threads": 5.0,
            "timed_calls": 8.0,
            "warmup_calls": 2.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "timing: `long_request` (the repo's 1,024-token example) (1,024 tokens, 3 questions), the median of 8 calls after 2 warm-up calls in one process, a call = write the four inputs, run, read `hidden` back; the C-API runner build ac2be1d3 (LiteRT 2.3.0), CPU XNNPACK, 5 threads (--cpu-threads 5)",
          "total_ops": 2432,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.3.0/2026-11-08/mnn-pocket-0.3b-dec__mnn-pocket-0.3b-dec_l1024_fp16w__pixel-11-pro.json"
      },
      {
        "device": "pixel-11-pro",
        "run": {
          "accelerator": "gpu_mldrift",
          "context_length": null,
          "date": "2026-11-07",
          "decode_tokens_per_s": null,
          "delegated_ops": 2432,
          "env": {
            "device": "Pixel 11 Pro",
            "machine_label": "pixel-11-pro-mnnpocketdec-capi-runner",
            "os_build": "Android 17 (CD1A.260905.001.B1)",
            "runtime": "litert",
            "runtime_version": "2.3.0",
            "soc": "Google Tensor G6",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "(tag p11p_gpu_fp32_L1024, run r2): run DONE rc 0, start 2026-11-07 19:22:30, 127 questions",
            "Push mnn_pocket_dec_ns4_in_6-27_post_6-27_final_L1024_w16.tflite rc=0 bytes 488880528/488880528 sha256 phone=a76354c9b14333d4d55385c4d0cd86b3bfe8a6df30a99a9963b138ea71a8397c = mac = the Hub LFS sha256",
            "Runner 7e7539e5bd0c73ed… and libs: 2/2 OK",
            "Device: Pixel 11 Pro 17 CD1A.260905.001.B1",
            "runner.replacing: VERBOSE: Replacing 2432 out of 2432 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 0 (main).",
            "line PASS: 127 questions, same winner 122/122 (not near ties), near ties kept 5/5, max |Δp| 0.000906431, mean |Δp| 9.71865e-05, control n/a (the file holds no request of the control pair), non-finite read-outs 0, winner = the Mac CPU's on the same file 122/122",
            "Run memory (25 samples): VmHWM_max_kB 2504732 VmSwap_max_kB 0 MemAvailable_min_kB 7812508 kgsl_page_alloc_max_B "
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": null,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "arm_files_passed": 8.0,
            "arm_files_run": 8.0,
            "compile_ms": 28893.8,
            "line_bar_control_abs_dp": 0.02,
            "line_bar_max_abs_dp": 0.02,
            "line_bar_mean_abs_dp": 0.002,
            "line_bar_near_tie_gap": 0.02,
            "line_mac_same_file_non_near_tie": 122.0,
            "line_max_abs_dp": 0.0009064306168956593,
            "line_mean_abs_dp": 9.71865125192646e-05,
            "line_near_tie": 5.0,
            "line_non_near_tie": 122.0,
            "line_nonfinite_questions": 0.0,
            "line_pass": 1.0,
            "line_questions": 127.0,
            "line_winner_equal_mac_non_near_tie": 122.0,
            "line_winner_equal_near_tie": 5.0,
            "line_winner_equal_non_near_tie": 122.0,
            "vmhwm_max_kb": 2504732.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "the line: 127 questions of the L1024 set against the source model's reference; the C-API runner build 7e7539e5 (LiteRT 2.3.0), GPU OpenCL, GpuOptions precision FP32",
          "total_ops": 2432,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.3.0/2026-11-07/mnn-pocket-0.3b-dec__mnn-pocket-0.3b-dec_l1024_fp16w__pixel-11-pro.json"
      },
      {
        "device": "pixel-11-pro",
        "run": {
          "accelerator": "gpu_mldrift",
          "context_length": null,
          "date": "2026-11-08",
          "decode_tokens_per_s": null,
          "delegated_ops": 2432,
          "env": {
            "device": "Pixel 11 Pro",
            "machine_label": "pixel-11-pro-mnnpocketdec-capi-runner",
            "os_build": "Android 17 (CD1A.260905.001.B1)",
            "runtime": "litert",
            "runtime_version": "2.3.0",
            "soc": "Google Tensor G6",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "1020.71 ms (min 1007.52, max 1034.56; median of 8 after 2 warm-ups, write+run+readback; load 15.54 s; no cap in 20 sampl… (the cell's value; parity cell: PASS; gate PASS, timing run = check run in every digit (V3 a); gate PASS: 122/122)",
            "(tag p11p_t_gpu_fp32_c_L1024_S, run r5): start 2026-11-08 01:14:23, request S/c_own_long (1024 tokens), write_run_read_ms median 1020.7149999999999 min 1007.521 max 1034.557 n 8, run_only_ms median 2.2735000000000003",
            "Push mnn_pocket_dec_ns4_in_6-27_post_6-27_final_L1024_w16.tflite rc=0 bytes 488880528/488880528 sha256 phone=a76354c9b14333d4d55385c4d0cd86b3bfe8a6df30a99a9963b138ea71a8397c = mac = the Hub LFS sha256",
            "Runner 7e7539e5bd0c73ed… and libs: 2/2 OK; extras: rest=180 sample=1",
            "Device: Pixel 11 Pro 17 CD1A.260905.001.B1",
            "clock samples: 20 samples, 0 capped, GPU level max 0",
            "parity: the timed request's 7 question(s) have the same probabilities in every digit as in the check run (line PASS)",
            "VERBOSE: Replacing 2432 out of 2432 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 0 (main).",
            "the cell's takes (take id: median, capped): 93: 1213 ms capped; 94: 1259 ms capped; 476: 1021 ms; 628: 1083 ms capped; 629: 1067 ms; the cell is take 476, the fastest uncapped take",
            "other rows of this file, request and backend, not takes of this cell: 86: GPU precision fp16; 89: GPU precision fp16acc32; 626: GPU precision fp16acc32; 627: GPU precision fp16acc32"
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": 1020.7149999999999,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "arm_files_passed": 8.0,
            "arm_files_run": 8.0,
            "capped": 0.0,
            "capped_samples": 0.0,
            "clock_samples": 20.0,
            "compile_ms": 15542.5,
            "latency_max_ms": 1034.557,
            "latency_min_ms": 1007.521,
            "request_questions": 3.0,
            "request_tokens": 1024.0,
            "run_only_p50_ms": 2.2735000000000003,
            "timed_calls": 8.0,
            "warmup_calls": 2.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "timing: `long_request` (the repo's 1,024-token example) (1,024 tokens, 3 questions), the median of 8 calls after 2 warm-up calls in one process, a call = write the four inputs, run, read `hidden` back; the C-API runner build 7e7539e5 (LiteRT 2.3.0), GPU OpenCL, GpuOptions precision FP32",
          "total_ops": 2432,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.3.0/2026-11-08/mnn-pocket-0.3b-dec__mnn-pocket-0.3b-dec_l1024_fp16w__pixel-11-pro.json"
      }
    ]
  },
  "model": {
    "family": "mnn-pocket",
    "id": "mnn-pocket-0.3b-dec__mnn-pocket-0.3b-dec_l1024_fp16w",
    "license": "apache-2.0",
    "source_url": "https://huggingface.co/litert-community/MNN-Pocket-0.3B-Dec-LiteRT",
    "task": "text-classification"
  },
  "pitfalls": [
    "A decision model, not a chat model: it takes a `state` and typed `questions` (`choice`, `noul`, `score`) and returns the probability of every candidate in one forward pass; it never generates text (README L18).",
    "The graph is the model body only: it takes the embeddings of the request's tokens, the rotary tables and the attention mask and returns the hidden states after the final norm; the Python host in `host/` builds the request, takes the embeddings from `embedding.safetensors` (the source model's token table) and reads the answers with the source model's head and temperature, as the source model's `inference.py` does (README L20, L147-L148).",
    "Each file is a static window: `<n>` in the name is the longest request it takes; run the smallest file that holds the request. A request longer than 2,048 tokens raises `ValueError` in the host, which never cuts it; the source model's own limit is 32,768 tokens (README L71, L84).",
    "The checks measure agreement with the source model's `inference.py`, unchanged, on its own files (float32, CPU, one thread): 2,409 questions on the Mac, a fixed subset of 16 to 322 questions per file on the phones; the bar is the same winner on every question whose two most likely options are more than 0.02 apart, a largest |dp| of at most 0.02 and a mean of at most 0.002 (README L88, L102).",
    "The repository has no Android host yet: an Android app has to port `host/pocket_dec_litert.py`; the phones ran LiteRT 2.3.0 through its C API with inputs the Mac computed with the host's code (README L137).",
    "On a GPU set float32 precision (`GpuOptions(enforce_f32=True)` in the Python host, `CompiledModel.GpuOptions.Precision.FP32` in Kotlin, `kLiteRtDelegatePrecisionFp32` in the C API): an app that sets no precision gets FP16 on both phones, and FP16 changes the answers: among questions whose top two options are more than 0.02 apart the winner changed on up to 5 of 305 on the Galaxy S26, 1 of 305 on the Pixel 11 Pro and 9 of 2,314 on Apple Metal at its default precision, and a probability moved by up to 0.068, 0.029 and 0.058 (README L73).",
    "FP16 with FP32 accumulation passes the bar on the Galaxy S26 (44 of 54 near ties keep the reference's winner, largest |dp| 0.0126) and misses the mean on 3 of the 6 files it ran on the Pixel 11 Pro; it can flip a near tie: in the source model's default example (`department`: `shipping` 0.41064 against `returns` 0.40709) it answers `returns` (README L96, L102, L104).",
    "Set the CPU thread count: with no CPU options LiteRT 2.3.0's C API ran on one thread, and the 128-token file took 1.3 times as long as at 4 threads on the Galaxy S26 and 2.2 times as long on the Pixel 11 Pro; the card's phone CPU rows use 4 threads (Galaxy S26) and 6 (Pixel 11 Pro), whose probabilities equal those of one thread to every digit (README L133).",
    "LiteRT's GPU cache (`serializationDir`, `modelCacheKey`, `serializeProgramCache` and `serializeExternalTensors` in `CompiledModel.GpuOptions`) cut the Pixel 11 Pro's creation of the 128-token model from 15.5 s to 1.8 to 2.0 s, but each call then took 12 to 19 % longer; on the Galaxy S26 it cut 1.6 s to 1.3 s at the same speed; the cache takes 0.93 GiB on disk and raises the process's peak by about 0.9 GiB (README L135)."
  ],
  "schema_version": "1.2"
}
