{
  "artifacts": [
    {
      "file": "timestral-209m-c2048-h256.tflite",
      "sha256": "c016500177c84c6f03cf746d59d690baf35c9fafb9a4c720cba99e669fb66fca",
      "size_mb": 386.278
    }
  ],
  "benchmarks": [],
  "conversion": {
    "command": "python scripts/r4_export.py --bucket 64x8 --rows 2 --norm pow2 --norm-k scripts/normk_fixture.json --attn-k 4 --attn-clamp --dec-only-last --fp16  # in timestral/ of github.com/john-rocky/LiteRT-Models, the zoo sample the card links (github.com/john-rocky/LiteRT-Models timestral/README.md at main ceb8299 ('Rebuilding the graphs')), after hf download mistralai/Timestral-209M config.json model.safetensors --revision 9874d252ab0aae16f7f7c999f98f98d6ad7d8365; the float16-weight output timestral209m_r2_l64_h8_pow2knormk_fixture_att4c_dl_wfp16.tflite is published as timestral-209m-c2048-h256.tflite",
    "quantization": "float16 weights (cast with ai-edge-quantizer 0.9.0); inputs and output float32; made for the GPU's FP16 precision (card 'Files', 'How the graphs were made')",
    "tool": "litert-torch",
    "tool_version": "0.9.4 (ai-edge-quantizer 0.9.0 for the weights; torch 2.13.0; ai-edge-litert 2.2.0 for the desktop checks)"
  },
  "cross_runtime": [],
  "delegation": {
    "backend": "gpu_mldrift",
    "blocking_ops": [
      "DEQUANTIZE",
      "LOGISTIC",
      "EXP",
      "LOG"
    ],
    "coverage_ops_pct": 85.4,
    "lint_report_version": "1.1",
    "litert_version": "2.2.0",
    "matched_provenance_counts": {
      "measured": 1825,
      "unmatched": 58
    },
    "partitions": 276
  },
  "device": {
    "records": [
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "cpu_xnnpack",
          "context_length": null,
          "date": "2026-10-22",
          "decode_tokens_per_s": null,
          "delegated_ops": 1883,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-timestral-gate",
            "os_build": "Android 16 BP4A.251205.006.S942QOPS1AZH9",
            "runtime": "litert",
            "runtime_version": "2.2.0",
            "soc": "Qualcomm SM8850 (Adreno GPU, Hexagon HTP)",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "gate app com.mlboydaisuke.d1omni.gate in generic mode (Kotlin CompiledModel, LiteRT 2.2.0; debug build r9), one cold process per run, the fixture rows pushed as raw float32 tensors (the host steps done on the Mac); measured interval = the five input writes + run + the output read-back per call. Holds s4 and t4 of 2026-10-22.",
            "forecast check vs the official TimestralForecaster.forecast() (PyTorch, float32, CPU, every default) on the 55 fixture windows of this size (contexts of up to 2,048 steps, horizons of up to 256 steps; 16 groups): MASE 1.465494 vs 1.465583 (ratio x0.999939), CRPS 0.150539 vs 0.150539 (ratio x1.000002); NaN / inf 0, quantile order broken 0, graph inputs equal to the host's 55 / 55; the bar (both ratios <= 1.01, no NaN / inf, the 21 quantiles in order at every step): PASS.",
            "output_match = the shared element rule (|out - ref| <= max(1e-05, 0.001 * |ref|) per element, litert_compat.probe.runners.Tolerance) against the same file on the Mac CPU on the same rows (ai-edge-litert 2.2.0 XNNPACK, 8 threads; the vs_mac.cpu block): every one of the 55 rows is bit-equal (max |d| 0) -> True.",
            "latency_p50_ms = the timing run (ready: clean; rested 60s since 20:30:05; thermal before the run: Thermal Status: 0; 5 warm-ups then 20 calls on one window, tsl_l64h8_r2_att_one): median 76.1673 ms (min 73.5584, max 78.8476; run() alone 76.1168); one call is one forecast (the two rows, the series and its negation, in one graph call; the host steps around it not included); compile 415 ms. The 55-window check run's warm median was 80.8369 ms (thermal before it: Thermal Status: 0).",
            "peak_mem_mb = the app process's VmHWM at the end of the timing run (1355584 kB, read by the gate app from /proc/self/status); the check run's 1376852 kB.",
            "os_build = the device line of hold n5's npu_chain log (2026-10-22 21:03 JST); holds t4 and s4 ran on the same phone the same evening without a reboot (its uptime grew by the wall-clock gap); the gate app reads Build.VERSION.SDK_INT 36, Build.MODEL SM-S942Q.",
            "sha256 timestral209m_r2_l64_h8_pow2knormk_fixture_att4c_dl_wfp16.tflite phone=c016500177c84c6f03cf746d59d690baf35c9fafb9a4c720cba99e669fb66fca mac=c016500177c84c6f03cf746d59d690baf35c9fafb9a4c720cba99e669fb66fca MATCH",
            "sha256 timestral209m_r2_l64_h8_pow2knormk_fixture_att4c_dl_wfp16.tflite phone=c016500177c84c6f03cf746d59d690baf35c9fafb9a4c720cba99e669fb66fca mac=c016500177c84c6f03cf746d59d690baf35c9fafb9a4c720cba99e669fb66fca MATCH",
            "10-22 20:31:06.422 14307 14351 I tflite  : Replacing 1883 out of 1883 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 1 partitions for subgraph 0 (main).",
            "10-22 20:39:13.081 25927 25950 I tflite  : Replacing 1883 out of 1883 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 1 partitions for subgraph 0 (main)."
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": 76.1672915,
          "loads": true,
          "max_abs_diff": 0.0,
          "max_rel_diff": 0.0,
          "metrics": {
            "bar_pass": 1,
            "crps": 0.15053946621402436,
            "crps_ratio_vs_reference": 1.0000017854239174,
            "crps_reference": 0.15053919743774075,
            "gate_leg_compile_ms": 437.805625,
            "gate_leg_rows": 55,
            "gate_leg_vmhwm_kb": 1376852,
            "gate_leg_warm_median_write_run_read_ms": 80.8368745,
            "gate_leg_warm_min_write_run_read_ms": 74.707292,
            "gate_leg_warm_rows": 50,
            "inputs_equal": 55,
            "mac_same_graph_bit_equal_rows": 55,
            "mac_same_graph_max_abs": 0.0,
            "mac_same_graph_median_ch_max_abs": 0.0,
            "mac_same_graph_ratio_crps": 1.0000017854239174,
            "mac_same_graph_ratio_mase": 0.9999392801662534,
            "mac_same_graph_rows": 55,
            "mase": 1.4654941066094855,
            "mase_ratio_vs_reference": 0.9999392801662534,
            "mase_reference": 1.4655830965714511,
            "nonfinite": 0,
            "partitions": 1,
            "q_max_rel_vs_reference": 0.0008784518398197885,
            "quantile_order_broken": 0,
            "threads": 4,
            "timing_leg_compile_ms": 414.797552,
            "timing_leg_max_ms": 78.847552,
            "timing_leg_median_ms": 76.1672915,
            "timing_leg_min_ms": 73.558385,
            "timing_leg_n": 20,
            "timing_leg_run_only_median_ms": 76.1168235,
            "timing_leg_vmhwm_kb": 1355584,
            "timing_leg_warmups": 5,
            "window_groups": 16,
            "windows": 55
          },
          "output_match": true,
          "peak_mem_mb": 1323.812,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "CPU XNNPACK 4 threads; gate app com.mlboydaisuke.d1omni.gate (generic mode)",
          "total_ops": 1883,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.2.0/2026-10-22/timestral-209m__timestral-209m-c2048-h256__galaxy-s26.json"
      },
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "gpu_mldrift",
          "context_length": null,
          "date": "2026-10-22",
          "decode_tokens_per_s": null,
          "delegated_ops": 1883,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-timestral-gate",
            "os_build": "Android 16 BP4A.251205.006.S942QOPS1AZH9",
            "runtime": "litert",
            "runtime_version": "2.2.0",
            "soc": "Qualcomm SM8850 (Adreno GPU, Hexagon HTP)",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "gate app com.mlboydaisuke.d1omni.gate in generic mode (Kotlin CompiledModel, LiteRT 2.2.0; debug build r9), one cold process per run, the fixture rows pushed as raw float32 tensors (the host steps done on the Mac); measured interval = the five input writes + run + the output read-back per call. Holds s4 and t4 of 2026-10-22.",
            "forecast check vs the official TimestralForecaster.forecast() (PyTorch, float32, CPU, every default) on the 55 fixture windows of this size (contexts of up to 2,048 steps, horizons of up to 256 steps; 16 groups): MASE 1.458846 vs 1.465583 (ratio x0.995403), CRPS 0.150430 vs 0.150539 (ratio x0.999277); NaN / inf 0, quantile order broken 0, graph inputs equal to the host's 55 / 55; the bar (both ratios <= 1.01, no NaN / inf, the 21 quantiles in order at every step): PASS.",
            "output_match null: the work compares the graph output with the same file on the Mac CPU on the same rows (ai-edge-litert 2.2.0 XNNPACK, 8 threads; the vs_mac.cpu block): max |d| 1.06971 over 55 rows (the median quantile's max |d| 0.0246626), 0 rows bit-equal; against Mac Metal at its default (fp16) precision max |d| 0.244141, 0 rows bit-equal. That summary does not decide the shared element rule (|out - ref| <= max(1e-05, 0.001 * |ref|) per element), which needs |ref| at each element, so max_abs_diff / max_rel_diff stay null; the forecast check above is the bar.",
            "latency_p50_ms = the timing run (ready: clean; rested 64s since 20:22:34; thermal before the run: Thermal Status: 0; 5 warm-ups then 20 calls on one window, tsl_l64h8_r2_att_one): median 27.0027 ms (min 26.0166, max 29.4262; run() alone 0.5897); one call is one forecast (the two rows, the series and its negation, in one graph call; the host steps around it not included); compile 1599 ms. The 55-window check run's warm median was 26.8891 ms (thermal before it: Thermal Status: 0).",
            "peak_mem_mb = the app process's VmHWM at the end of the timing run (1915992 kB, read by the gate app from /proc/self/status); the check run's 1915300 kB.",
            "os_build = the device line of hold n5's npu_chain log (2026-10-22 21:03 JST); holds t4 and s4 ran on the same phone the same evening without a reboot (its uptime grew by the wall-clock gap); the gate app reads Build.VERSION.SDK_INT 36, Build.MODEL SM-S942Q.",
            "sha256 timestral209m_r2_l64_h8_pow2knormk_fixture_att4c_dl_wfp16.tflite phone=c016500177c84c6f03cf746d59d690baf35c9fafb9a4c720cba99e669fb66fca mac=c016500177c84c6f03cf746d59d690baf35c9fafb9a4c720cba99e669fb66fca MATCH",
            "sha256 timestral209m_r2_l64_h8_pow2knormk_fixture_att4c_dl_wfp16.tflite phone=c016500177c84c6f03cf746d59d690baf35c9fafb9a4c720cba99e669fb66fca mac=c016500177c84c6f03cf746d59d690baf35c9fafb9a4c720cba99e669fb66fca MATCH",
            "10-22 20:23:39.333  8864  8887 I tflite  : Replacing 1883 out of 1883 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 0 (main).",
            "10-22 20:38:57.934 25324 25348 I tflite  : Replacing 1883 out of 1883 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 0 (main)."
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": 27.0026565,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "bar_pass": 1,
            "crps": 0.1504303565601816,
            "crps_ratio_vs_reference": 0.999276993106037,
            "crps_reference": 0.15053919743774075,
            "gate_leg_compile_ms": 1648.84,
            "gate_leg_rows": 55,
            "gate_leg_vmhwm_kb": 1915300,
            "gate_leg_warm_median_write_run_read_ms": 26.889114499999998,
            "gate_leg_warm_min_write_run_read_ms": 26.30125,
            "gate_leg_warm_rows": 50,
            "inputs_equal": 55,
            "mac_metal_fp16_bit_equal_rows": 0,
            "mac_metal_fp16_max_abs": 0.244140625,
            "mac_same_graph_bit_equal_rows": 0,
            "mac_same_graph_max_abs": 1.0697097778320312,
            "mac_same_graph_median_ch_max_abs": 0.02466261386871338,
            "mac_same_graph_ratio_crps": 1.0000017854239174,
            "mac_same_graph_ratio_mase": 0.9999392801662534,
            "mac_same_graph_rows": 55,
            "mase": 1.4588455977256654,
            "mase_ratio_vs_reference": 0.9954028544259638,
            "mase_reference": 1.4655830965714511,
            "nonfinite": 0,
            "partitions": 1,
            "q_max_rel_vs_reference": 0.06169247906492436,
            "quantile_order_broken": 0,
            "timing_leg_compile_ms": 1598.825312,
            "timing_leg_gpu_max_clock_mhz": 1300,
            "timing_leg_max_ms": 29.426198,
            "timing_leg_median_ms": 27.0026565,
            "timing_leg_min_ms": 26.016562,
            "timing_leg_n": 20,
            "timing_leg_run_only_median_ms": 0.589687,
            "timing_leg_vmhwm_kb": 1915992,
            "timing_leg_warmups": 5,
            "window_groups": 16,
            "windows": 55
          },
          "output_match": null,
          "peak_mem_mb": 1871.086,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "GpuOptions precision=FP16; gate app com.mlboydaisuke.d1omni.gate (generic mode)",
          "total_ops": 1883,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.2.0/2026-10-22/timestral-209m__timestral-209m-c2048-h256__galaxy-s26.json"
      },
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "npu_qnn",
          "context_length": null,
          "date": "2026-10-22",
          "decode_tokens_per_s": null,
          "delegated_ops": 1883,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-timestral-npu-runner",
            "os_build": "Android 16 BP4A.251205.006.S942QOPS1AZH9",
            "runtime": "litert",
            "runtime_version": "2.2.0",
            "soc": "Qualcomm SM8850 (Adreno GPU, Hexagon HTP)",
            "vendor_sdk": "Qualcomm QAIRT QNN HTP libraries (libQnnHtp.so v2.47.0.260601114230, v81 skel) + the LiteRT 2.2.0 release's Qualcomm dispatch library and JIT compiler plugin"
          },
          "error": null,
          "evidence": [
            "the LiteRT C-API runner tsl_npu_runner (the r17 runner, LiteRT 2.2.0 C API; sha256 102a0f0a815df1a29a2db55cdcb71c9177f95de596c6f660f9596a27c22b3c94, the phone's read-back MATCH) with the Qualcomm libraries pushed to /data/local/tmp/tsl_npu (every file's sha256 OK) through benchmark script, one process per run: the HTP through the Qualcomm compiler plugin (JIT, a fresh cache dir per run). Hold n5 of 2026-10-22.",
            "forecast check vs the official TimestralForecaster.forecast() (PyTorch, float32, CPU, every default) on the 55 fixture windows of this size (contexts of up to 2,048 steps, horizons of up to 256 steps; 16 groups): MASE 1.472794 vs 1.465583 (ratio x1.004920), CRPS 0.150408 vs 0.150539 (ratio x0.999126); NaN / inf 0, quantile order broken 0, graph inputs equal to the host's 55 / 55; the bar (both ratios <= 1.01, no NaN / inf, the 21 quantiles in order at every step): PASS.",
            "output_match null: the work compares the graph output with the same file on the Mac CPU on the same rows (ai-edge-litert 2.2.0 XNNPACK, 8 threads; the vs_mac.cpu block, outputs... 054.f32): max |d| 3.94149 over 55 rows (the median quantile's max |d| 0.0263262), 0 rows bit-equal; against Mac Metal at its default (fp16) precision max |d| 3.67188, 0 rows bit-equal. That summary does not decide the shared element rule (|out - ref| <= max(1e-05, 0.001 * |ref|) per element), which needs |ref| at each element, so max_abs_diff / max_rel_diff stay null; the forecast check above is the bar.",
            "latency_p50_ms = the timing run's latency phase (ready: clean; rested 64s since 21:17:51; thermal before the run: Thermal Status: 0; 5 warm-ups then 20 rounds of one call on fixture 000): median 15.5695 ms (min 15.4820, max 16.0750; run alone 15.5595); one call is one forecast; compile 19500 ms (JIT, cold cache). The 55-window check run's median call was 15.7755 ms (JIT compile 20280 ms).",
            "peak_mem_mb null: the chain samples the runner pid's VmHWM during the run (max 1382964 kB in the timing run, 2402356 kB in the check run; in metrics), not at its exit.",
            "os_build = the device line of hold n5's npu_chain log (2026-10-22 21:03 JST); holds t4 and s4 ran on the same phone the same evening without a reboot (its uptime grew by the wall-clock gap); the gate app reads Build.VERSION.SDK_INT 36, Build.MODEL SM-S942Q.",
            "Push timestral209m_r2_l64_h8_pow2knormk_fixture_att4c_dl_wfp16.tflite rc=0 bytes 405041792/405041792 sha256 phone=c016500177c84c6f03cf746d59d690baf35c9fafb9a4c720cba99e669fb66fca mac=c016500177c84c6f03cf746d59d690baf35c9fafb9a4c720cba99e669fb66fca MATCH in 10s",
            "command: cd /data/local/tmp/tsl_npu && rm -rf /data/local/tmp/tsl_npu/out_N5_jit_b1h8_dlw && LD_LIBRARY_PATH=/data/local/tmp/tsl_npu ADSP_LIBRARY_PATH='/data/local/tmp/tsl_npu;/vendor/dsp/cdsp;/vendor/lib/rfsa/adsp;/system/lib/rfsa/adsp;/dsp' ./tsl_npu_runner --accel npu+cpu  --burst 1 --libdir /data/local/tmp/tsl_npu --cache-dir /data/local/tmp/tsl_npu/cache_t5h --fixtures /data/local/tmp/tsl_npu/fx_tsl_l64h8_r2_att --count 1 --warmup 5 --rounds 20 --ab-fixture 000 --dump full --out /data/local/tmp/tsl_npu/out_N5_jit_b1h8_dlw --model /data/local/tmp/tsl_npu/timestral209m_r2_l64_h8_pow2knormk_fixture_att4c_dl_wfp16.tflite 2>/data/local/tmp/tsl_npu/N5_jit_b1h8_dlw.stderr.txt",
            "INFO: [compiler_plugin.cc:741] Partitioned subgraph<0>, selected 1883 ops, from a total of 1883 ops. resulted in 1 partitions.",
            "VERBOSE: Replacing 1 out of 1 node(s) with delegate (DispatchDelegate) node, yielding 1 partitions for subgraph 0 (main)."
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": 15.569500000000001,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "bar_pass": 1,
            "crps": 0.1504076688897829,
            "crps_ratio_vs_reference": 0.9991262837174867,
            "crps_reference": 0.15053919743774075,
            "fully_accelerated": 1,
            "gate_leg_compile_ms": 20279.8,
            "gate_leg_parity_call_ms_median": 15.775500000000001,
            "gate_leg_runner_vmhwm_max_sampled_kb": 2402356,
            "inputs_equal": 55,
            "load_compile_ms": 19500.0,
            "mac_metal_fp16_bit_equal_rows": 0,
            "mac_metal_fp16_max_abs": 3.671875,
            "mac_same_graph_bit_equal_rows": 0,
            "mac_same_graph_max_abs": 3.94149112701416,
            "mac_same_graph_median_ch_max_abs": 0.02632620930671692,
            "mac_same_graph_ratio_crps": 1.0000017854239174,
            "mac_same_graph_ratio_mase": 0.9999392801662534,
            "mac_same_graph_rows": 55,
            "mase": 1.4727937950331047,
            "mase_ratio_vs_reference": 1.004920020214836,
            "mase_reference": 1.4655830965714511,
            "nonfinite": 0,
            "partitions": 1,
            "q_max_rel_vs_reference": 0.09260243344440322,
            "quantile_order_broken": 0,
            "runner_call_ms_max": 16.075,
            "runner_call_ms_median": 15.569500000000001,
            "runner_call_ms_min": 15.482,
            "runner_rounds": 20,
            "runner_run_only_ms_median": 15.5595,
            "runner_vmhwm_max_sampled_kb": 1382964,
            "runner_warmups": 5,
            "window_groups": 16,
            "windows": 55
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "QualcommOptions JIT (HTP, burst), cold compile cache per run; C-API runner --accel npu+cpu",
          "total_ops": 1883,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.2.0/2026-10-22/timestral-209m__timestral-209m-c2048-h256__galaxy-s26.json"
      }
    ]
  },
  "model": {
    "family": "timestral-209m",
    "id": "timestral-209m__timestral-209m-c2048-h256",
    "license": "other (modified-mit)",
    "source_url": "https://huggingface.co/mlboydaisuke/Timestral-209M-LiteRT",
    "task": "time-series-forecasting"
  },
  "pitfalls": [
    "Timestral-209M's attention logits (up to 242,875) and RMSNorm sums of squares (up to 164,602,176) leave the float16 range (65,504): each RMSNorm input is pre-scaled by a power of two, the logits are computed at 1/16 with the row maximum subtracted before the scale is restored, the softmax input is clamped at -3*10^4 and the mask value is -3*10^4; in float32 these rewrites leave the results unchanged (card 'How the graphs were made').",
    "Five float32 inputs, one signature `serving_default`: `patches` [2,64,64] (per patch 32 normalized values then 32 missing flags), `dec` [2,8,768] (`forecast_query.f32` for every decoder token), `bias` [1,1,72,72] (additive mask: 0 or -3*10^4), `cos` / `sin` [1,1,72,64] (rows cut from `rope_cos.f32` / `rope_sin.f32`) -> `output_0` [2,8,32,21], the normalized quantiles; the two rows are the series and its negation (card 'Inputs and output').",
    "The graph is one step of the forecast: the host builds the inputs and finishes the forecast the way the official forecaster does (causal float64 scaling and asinh, the attention mask, the inverse map and the advancing refine frame, the caps at the context's range widened by 10^4 times the scale, symmetric averaging with the negation, clipping at 0 for non-negative series); the repository's `host_forecast.py` and `android/TimestralHost.kt` do it (card 'Inputs and output').",
    "A file holds a context of up to 2,048 steps and a horizon of up to 256 steps (patches of 32): the host pads a shorter series in front with missing steps and keeps the requested steps; the smallest file that holds the series and the horizon takes the least time (card 'Files').",
    "One series per forecast without covariates; a horizon of up to 1,024 steps (the Python class and the Kotlin host stop there; no result covers the official block-by-block decoding of longer horizons); of a series longer than 8,192 steps the last 8,192 are read (the official forecaster with max_context = 8192; its default is 32,768); the Python class refuses fewer than 16 values (card 'Limits').",
    "The float16 files are made for the GPU's FP16 precision: set it explicitly, `CompiledModel.GpuOptions(precision = CompiledModel.GpuOptions.Precision.FP16)` (card 'Minimal usage').",
    "NPU (Qualcomm HTP): the graph is compiled on the phone at its first load (Galaxy S26: first JIT compile 19.5 s in the speed run; in the card's Kotlin run the first process took 18.1 s for the compile, one forecast and close(), the next 0.3 s from the cache); LiteRT keeps the result in the app's cache when the Environment is created with a Context. When the compiler plugin is missing, LiteRT runs the graph on the CPU without an error: look for `Replacing 1 out of 1 node(s) with delegate (DispatchDelegate)` in logcat (card 'Speed on phones', 'Minimal usage').",
    "LiteRT starts the NPU runtime once per process with the options of the first graph it compiles on any accelerator: compile the NPU graph first, or give every graph the same `qualcommOptions` (card 'Minimal usage').",
    "The Qualcomm libraries are not in the repository (the Qualcomm runtime may be distributed only inside an app); put them in `jniLibs/arm64-v8a/` with `useLegacyPackaging = true` so `libQnnHtpV81Skel.so` is extracted as a real file. The NPU path was run on one chip, the Snapdragon SM8850 of a Galaxy S26 (card 'Minimal usage', 'Limits')."
  ],
  "schema_version": "1.2"
}
