{
  "artifacts": [
    {
      "file": "timestral-209m-c8192-h256.tflite",
      "sha256": "62a92115fb81e16f7b235d402c74df6175c568540b6c0ac14f354d92567763d7",
      "size_mb": 386.278
    }
  ],
  "benchmarks": [],
  "conversion": {
    "command": "python scripts/r4_export.py --bucket 256x8 --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_l256_h8_pow2knormk_fixture_att4c_dl_wfp16.tflite is published as timestral-209m-c8192-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 5 fixture windows of this size (contexts of 2,049 to 8,192 steps, horizons of up to 256 steps; 2 groups): MASE 0.432961 vs 0.432953 (ratio x1.000020), CRPS 0.125659 vs 0.125656 (ratio x1.000017); NaN / inf 0, quantile order broken 0, graph inputs equal to the host's 5 / 5; 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 5 rows is bit-equal (max |d| 0) -> True.",
            "latency_p50_ms = the timing run (ready: clean; rested 63s since 20:31:14; thermal before the run: Thermal Status: 0; 5 warm-ups then 20 calls on one window, tsl_l256h8_r2_att_one): median 294.1965 ms (min 282.0252, max 306.4591; run() alone 294.1070); 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 417 ms. The 5-window check run's warm median was 289.6192 ms (thermal before it: Thermal Status: 0).",
            "peak_mem_mb = the app process's VmHWM at the end of the timing run (1376384 kB, read by the gate app from /proc/self/status); the check run's 1380888 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_l256_h8_pow2knormk_fixture_att4c_dl_wfp16.tflite phone=62a92115fb81e16f7b235d402c74df6175c568540b6c0ac14f354d92567763d7 mac=62a92115fb81e16f7b235d402c74df6175c568540b6c0ac14f354d92567763d7 MATCH",
            "sha256 timestral209m_r2_l256_h8_pow2knormk_fixture_att4c_dl_wfp16.tflite phone=62a92115fb81e16f7b235d402c74df6175c568540b6c0ac14f354d92567763d7 mac=62a92115fb81e16f7b235d402c74df6175c568540b6c0ac14f354d92567763d7 MATCH",
            "10-22 20:32:18.984 15200 15222 I tflite  : Replacing 1883 out of 1883 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 1 partitions for subgraph 0 (main).",
            "10-22 20:39:59.548 26964 26985 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": 294.1965365,
          "loads": true,
          "max_abs_diff": 0.0,
          "max_rel_diff": 0.0,
          "metrics": {
            "bar_pass": 1,
            "crps": 0.12565858247433664,
            "crps_ratio_vs_reference": 1.00001657589097,
            "crps_reference": 0.1256564996058995,
            "gate_leg_compile_ms": 434.312135,
            "gate_leg_rows": 5,
            "gate_leg_vmhwm_kb": 1380888,
            "gate_leg_warm_median_write_run_read_ms": 289.619219,
            "gate_leg_warm_min_write_run_read_ms": 277.588646,
            "gate_leg_warm_rows": 5,
            "inputs_equal": 5,
            "mac_same_graph_bit_equal_rows": 5,
            "mac_same_graph_max_abs": 0.0,
            "mac_same_graph_median_ch_max_abs": 0.0,
            "mac_same_graph_ratio_crps": 1.00001657589097,
            "mac_same_graph_ratio_mase": 1.0000196786182198,
            "mac_same_graph_rows": 5,
            "mase": 0.4329611119051438,
            "mase_ratio_vs_reference": 1.0000196786182198,
            "mase_reference": 0.4329525919963786,
            "nonfinite": 0,
            "partitions": 1,
            "q_max_rel_vs_reference": 0.00021003734720978083,
            "quantile_order_broken": 0,
            "threads": 4,
            "timing_leg_compile_ms": 416.960156,
            "timing_leg_max_ms": 306.459062,
            "timing_leg_median_ms": 294.1965365,
            "timing_leg_min_ms": 282.025156,
            "timing_leg_n": 20,
            "timing_leg_run_only_median_ms": 294.106953,
            "timing_leg_vmhwm_kb": 1376384,
            "timing_leg_warmups": 5,
            "window_groups": 2,
            "windows": 5
          },
          "output_match": true,
          "peak_mem_mb": 1344.125,
          "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-c8192-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 5 fixture windows of this size (contexts of 2,049 to 8,192 steps, horizons of up to 256 steps; 2 groups): MASE 0.433429 vs 0.432953 (ratio x1.001101), CRPS 0.125839 vs 0.125656 (ratio x1.001453); NaN / inf 0, quantile order broken 0, graph inputs equal to the host's 5 / 5; 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| 0.0289216 over 5 rows (the median quantile's max |d| 0.0279772), 0 rows bit-equal; against Mac Metal at its default (fp16) precision max |d| 0.0175781, 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 80s since 20:23:47; thermal before the run: Thermal Status: 0; 5 warm-ups then 20 calls on one window, tsl_l256h8_r2_att_one): median 68.7149 ms (min 68.0065, max 70.3365; run() alone 0.3583); 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 1633 ms. The 5-window check run's warm median was 69.4403 ms (thermal before it: Thermal Status: 0).",
            "peak_mem_mb = the app process's VmHWM at the end of the timing run (1919520 kB, read by the gate app from /proc/self/status); the check run's 1916616 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_l256_h8_pow2knormk_fixture_att4c_dl_wfp16.tflite phone=62a92115fb81e16f7b235d402c74df6175c568540b6c0ac14f354d92567763d7 mac=62a92115fb81e16f7b235d402c74df6175c568540b6c0ac14f354d92567763d7 MATCH",
            "sha256 timestral209m_r2_l256_h8_pow2knormk_fixture_att4c_dl_wfp16.tflite phone=62a92115fb81e16f7b235d402c74df6175c568540b6c0ac14f354d92567763d7 mac=62a92115fb81e16f7b235d402c74df6175c568540b6c0ac14f354d92567763d7 MATCH",
            "10-22 20:25:08.085  9733  9754 I tflite  : Replacing 1883 out of 1883 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 0 (main).",
            "10-22 20:39:43.858 26517 26544 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": 68.714922,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "bar_pass": 1,
            "crps": 0.12583904967166695,
            "crps_ratio_vs_reference": 1.0014527705796357,
            "crps_reference": 0.1256564996058995,
            "gate_leg_compile_ms": 1673.063281,
            "gate_leg_rows": 5,
            "gate_leg_vmhwm_kb": 1916616,
            "gate_leg_warm_median_write_run_read_ms": 69.440312,
            "gate_leg_warm_min_write_run_read_ms": 68.448125,
            "gate_leg_warm_rows": 5,
            "inputs_equal": 5,
            "mac_metal_fp16_bit_equal_rows": 0,
            "mac_metal_fp16_max_abs": 0.017578125,
            "mac_same_graph_bit_equal_rows": 0,
            "mac_same_graph_max_abs": 0.02892160415649414,
            "mac_same_graph_median_ch_max_abs": 0.02797722816467285,
            "mac_same_graph_ratio_crps": 1.00001657589097,
            "mac_same_graph_ratio_mase": 1.0000196786182198,
            "mac_same_graph_rows": 5,
            "mase": 0.43342947962900297,
            "mase_ratio_vs_reference": 1.001101477717053,
            "mase_reference": 0.4329525919963786,
            "nonfinite": 0,
            "partitions": 1,
            "q_max_rel_vs_reference": 0.004934782064638312,
            "quantile_order_broken": 0,
            "timing_leg_compile_ms": 1633.312969,
            "timing_leg_gpu_max_clock_mhz": 1300,
            "timing_leg_max_ms": 70.336458,
            "timing_leg_median_ms": 68.714922,
            "timing_leg_min_ms": 68.00651,
            "timing_leg_n": 20,
            "timing_leg_run_only_median_ms": 0.358255,
            "timing_leg_vmhwm_kb": 1919520,
            "timing_leg_warmups": 5,
            "window_groups": 2,
            "windows": 5
          },
          "output_match": null,
          "peak_mem_mb": 1874.531,
          "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-c8192-h256__galaxy-s26.json"
      }
    ]
  },
  "model": {
    "family": "timestral-209m",
    "id": "timestral-209m__timestral-209m-c8192-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` [2,256,64] (per patch 32 normalized values then 32 missing flags), `dec` [2,8,768] (`forecast_query.f32` for every decoder token), `bias` [1,1,264,264] (additive mask: 0 or -3*10^4), `cos` / `sin` [1,1,264,64] (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 8,192 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').",
    "The `c8192` files are not used on the NPU: on the Galaxy S26, other forms of these graphs took 82.0 to 96.5 ms per forecast on the NPU against 68.7 to 73.4 ms for these files on its GPU (card 'Speed on phones')."
  ],
  "schema_version": "1.2"
}
