{
  "artifacts": [
    {
      "file": "gliclass_edge_v3_s256_fp32.tflite",
      "sha256": "aff0ee9795a2ffc2d4d32385e8f804b30bac4282d96d19c1e53cac88dcc97708",
      "size_mb": 51.465
    }
  ],
  "benchmarks": [],
  "browser": {
    "backends": [
      {
        "backend": "wasm_xnnpack",
        "date": "2026-10-02",
        "env": {
          "browser": "chromium",
          "browser_version": "151.0.7922.34",
          "headless": true,
          "jspi": true,
          "litertjs_core_version": "2.5.3",
          "machine_label": "mac-studio-m4-max",
          "os": "macOS",
          "os_version": "27.0.0",
          "webgpu_adapter": {
            "architecture": "metal-3",
            "description": "",
            "device": "",
            "vendor": "apple"
          }
        },
        "full_delegation": null,
        "latency_p50_ms": 101.72,
        "loads": true,
        "max_rel_diff": null,
        "output_match": null,
        "provenance": "measured",
        "runs": true
      },
      {
        "backend": "webgpu_mldrift",
        "date": "2026-10-02",
        "env": {
          "browser": "chromium",
          "browser_version": "151.0.7922.34",
          "headless": true,
          "jspi": true,
          "litertjs_core_version": "2.5.3",
          "machine_label": "mac-studio-m4-max",
          "os": "macOS",
          "os_version": "27.0.0",
          "webgpu_adapter": {
            "architecture": "metal-3",
            "description": "",
            "device": "",
            "vendor": "apple"
          }
        },
        "full_delegation": true,
        "latency_p50_ms": 7.918,
        "loads": true,
        "max_rel_diff": 1.636965062852597e-05,
        "output_match": true,
        "provenance": "measured",
        "runs": true
      }
    ],
    "demo_url": null,
    "sweep_source": "data/sweep/2.5.3/2026-10-02/gliclass-edge-v3.0__gliclass_edge_v3_s256_fp32.json"
  },
  "conversion": {
    "command": "conversion/export.py --seq 256 --matmul rank4 --ln-shift audit --suffix sln (litert-torch convert of the classification-path graph; the fp16-weight file by ai-edge-quantizer FLOAT_CASTING in the same script); card 'Provenance, conversion and license', REPRODUCE.md of the Hub repo",
    "quantization": "float32 weights: 670 operators, no quantization (card 'Files': the default download).",
    "tool": "litert-torch",
    "tool_version": "0.9.3 (torch 2.12.1; transformers 5.17.0; gliclass 0.1.20; ai-edge-litert 2.1.6; ai-edge-quantizer 0.8.0 for the float16 weight storage)"
  },
  "cross_runtime": [],
  "delegation": {
    "backend": "gpu_mldrift",
    "blocking_ops": [
      "SQUARED_DIFFERENCE"
    ],
    "coverage_ops_pct": 96.9,
    "lint_report_version": "1.1",
    "litert_version": "2.2.0",
    "matched_provenance_counts": {
      "measured": 649,
      "unmatched": 21
    },
    "partitions": 22
  },
  "device": {
    "records": [
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "cpu_xnnpack",
          "context_length": null,
          "date": "2026-10-04",
          "decode_tokens_per_s": null,
          "delegated_ops": 670,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-hfmodels-decide-gate",
            "os_build": "Android 16 (build BP4A.251205.006.S942QOPS1AZH9)",
            "runtime": "litert",
            "runtime_version": "2.2.0",
            "soc": "Qualcomm SM8850",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "hfmodels-android SDK device check: androidTest GliclassDeviceTest of the litert module (git 29bc04c), run by tools/decide_gate.sh on the Galaxy S26, one instrumentation process per run; the SDK (hfmodels-litert EncoderDecisions, family gliclass) resolved litert-community/GLiClass-Edge-v3.0-LiteRT through the dev descriptor catalog/dev/litert-community__GLiClass-Edge-v3.0-LiteRT.hfmodels.json at that commit (the log's descriptor sha256 prefix afe7d983 = `git show 29bc04c:catalog/dev/litert-community__GLiClass-Edge-v3.0-LiteRT.hfmodels.json`); graph gliclass_edge_v3_s256_fp32.tflite sha256 aff0ee9795a2ffc2d4d32385e8f804b30bac4282d96d19c1e53cac88dcc97708 = the published gliclass_edge_v3_s256_fp32.tflite (litert-community/GLiClass-Edge-v3.0-LiteRT@88c90950; other records hold rows of the same name and digest).",
            "# hfmodels typed-decisions device check  date=2026-10-04T01:29:34Z serial=RFGL80R6A6H device=SM-S942Q build=BP4A.251205.006.S942QOPS1AZH9 android=16",
            "# git=29bc04c test=GliclassDeviceTest variant=s256_fp32 backend=cpu thermal=Thermal Status: 0 battery_level= level: 79 battery_temp_x10C= temperature: 361 screen= mWakefulness=Dozing",
            "# gradle rc=0 end=2026-10-04T01:29:43Z thermal_end=Thermal Status: 0 battery_temp_end_x10C= temperature: 360",
            "resolve litert-community/GLiClass-Edge-v3.0-LiteRT: commit=88c90950 via EXPLICIT_REVISION, descriptor=caller@explicit/hfmodels.json sha=afe7d983",
            "no import line for the graph: import_ms=0, the file already in the SDK's store from an earlier run of the same bytes (imported there with 'sha256 ok')",
            "ready litert-community/GLiClass-Edge-v3.0-LiteRT@88c90950 variant=s256_fp32 profile=cpu inference=CPU prepare_ms=678",
            "litert  : [compiled_model.cc:1250] 0 compiler plugins were applied successfully: ",
            "tflite  : Replacing 670 out of 670 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 1 partitions for subgraph 0 (main).",
            "RESULT step=load ok=true variant=s256_fp32 backend=cpu import_ms=0 load_ms=689 compile_ms=181 profile=cpu window=256 head=256 max_options=25 notes=family=gliclass compile_ms=181 on cpu (4 threads) | observed backend is UNKNOWN: litert 2.2.0 reports no per-op execution target; 'initialized' is the accelerator the CompiledModel was created with | host token lookup: tok_embeddings_fp16.bin (float16, 50370 rows x 384) feeds inputs_embeds | one gliclass request per question: labels = a choice's descriptions (the key when there is none), a score's levels, a noul's false / true descriptions or no / yes; prompt = the instructions plus one space, right before the text; a request longer than the 256-token window or with more than 25 labels is refused, never cut; probabilities are gliclass's single-label float32 softmax of the label logits (T=1)",
            "RESULT step=sequence ok=true variant=s256_fp32 backend=cpu rows=152 of 152 mismatches=0 window=256",
            "RESULT step=parity ok=true variant=s256_fp32 backend=cpu rows=152 argmax_match=152/152 single_label_equal_official=152/152 multi_label_equal_official=152/152 max_prob_abs_err=0.001450374722480774 (worst semif_135) max_logit_abs_err_vs_oracle=0.009016335010528564 dp_tol=0.01 forward_ms_median=24.105937 p90=24.743177 min=23.791198 max=31.640886",
            "RESULT step=demo ok=true variant=s256_fp32 backend=cpu form=described rows=30 same_answer_as_mac=30/30 max_prob_abs_err_vs_mac=0.0 (worst ) encoded_length_equal_host=30/30 dp_tol=0.01 correct_device=10/30 correct_mac=10/30 question_ms_median=26.85276 p90=27.541197999999998 min=25.658646 max=27.675626 lookup_ms_median=0.243698",
            "RESULT step=demo ok=true variant=s256_fp32 backend=cpu form=keys rows=30 same_answer_as_mac=30/30 max_prob_abs_err_vs_mac=0.0 (worst ) encoded_length_equal_host=30/30 dp_tol=0.01 correct_device=10/30 correct_mac=10/30 question_ms_median=25.073697 p90=25.686301999999998 min=24.918021 max=26.340625 lookup_ms_median=0.222135",
            "RESULT step=timing ok=true variant=s256_fp32 backend=cpu repeats=5 decide_4q_total_ms_median=102.827604 four_decide_1q_total_ms_median=102.415052 state_tokens=21 per_question_ms=25.371823,25.184687,26.501458000000003,26.018229 answers={department=billing 0.9799988865852356, urgency=0.11925152130424976, refund_requested=0.33966735005378723, tone=angry 0.7151762247085571}",
            "RESULT step=lookup ok=true variant=s256_fp32 backend=cpu rows=128 hidden=384 zoo_style=lut zoo_ms_median=1.458125 zoo_ms_min=1.443594 sdk_ms_median=0.086145 sdk_ms_min=0.085573 repeats=20 outputs_identical=true",
            "RESULT step=release ok=true variant=s256_fp32 backend=cpu close_ms=6",
            "RESULT ok=true model=litert-community/GLiClass-Edge-v3.0-LiteRT variant=s256_fp32 backend=cpu device=SM-S942Q build=BP4A.251205.006.S942QOPS1AZH9 failures=",
            "latency_p50_ms = the parity step's forward_ms_median: one LiteRtDecisionModel.scores call per captured request (host float16 table lookup, input writes, CompiledModel run, output readback on the model thread), after one warm forward.",
            "output_match null: the gate compares with the publisher's outputs under its own criterion (argmax and max |dp| <= 0.01, in metrics) and keeps no raw per-element outputs. full_delegation from the delegate line the gate's logcat filter now keeps ('Replacing 670 out of 670 node(s) with delegate (TfLiteXNNPackDelegate)').",
            "conditions (the gate's header and footer): thermal status 0 -> 0, battery 36.1 -> 36.0 C, battery level 79 %, display mWakefulness=Dozing at the start.",
            "source: hfmodels-android/litert/results/2026-10-04-1029-RFGL80R6A6H-litert2.2.0-decide-gliclass-s256_fp32-cpu.log (main 3e1fcd8)."
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": 24.105937,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "battery_level_start": 79,
            "battery_temp_c_end": 36.0,
            "battery_temp_c_start": 36.1,
            "demo_described_correct_device": 10.0,
            "demo_described_correct_device_of": 30.0,
            "demo_described_correct_mac": 10.0,
            "demo_described_correct_mac_of": 30.0,
            "demo_described_dp_tol": 0.01,
            "demo_described_encoded_length_equal_host": 30.0,
            "demo_described_encoded_length_equal_host_of": 30.0,
            "demo_described_lookup_ms_median": 0.243698,
            "demo_described_max": 27.675626,
            "demo_described_max_prob_abs_err_vs_mac": 0.0,
            "demo_described_min": 25.658646,
            "demo_described_p90": 27.541197999999998,
            "demo_described_question_ms_median": 26.85276,
            "demo_described_rows": 30.0,
            "demo_described_same_answer_as_mac": 30.0,
            "demo_described_same_answer_as_mac_of": 30.0,
            "demo_keys_correct_device": 10.0,
            "demo_keys_correct_device_of": 30.0,
            "demo_keys_correct_mac": 10.0,
            "demo_keys_correct_mac_of": 30.0,
            "demo_keys_dp_tol": 0.01,
            "demo_keys_encoded_length_equal_host": 30.0,
            "demo_keys_encoded_length_equal_host_of": 30.0,
            "demo_keys_lookup_ms_median": 0.222135,
            "demo_keys_max": 26.340625,
            "demo_keys_max_prob_abs_err_vs_mac": 0.0,
            "demo_keys_min": 24.918021,
            "demo_keys_p90": 25.686301999999998,
            "demo_keys_question_ms_median": 25.073697,
            "demo_keys_rows": 30.0,
            "demo_keys_same_answer_as_mac": 30.0,
            "demo_keys_same_answer_as_mac_of": 30.0,
            "load_compile_ms": 181.0,
            "load_head": 256.0,
            "load_import_ms": 0.0,
            "load_load_ms": 689.0,
            "load_max_options": 25.0,
            "load_window": 256.0,
            "lookup_hidden": 384.0,
            "lookup_repeats": 20.0,
            "lookup_rows": 128.0,
            "lookup_sdk_ms_median": 0.086145,
            "lookup_sdk_ms_min": 0.085573,
            "lookup_zoo_ms_median": 1.458125,
            "lookup_zoo_ms_min": 1.443594,
            "parity_argmax_match": 152.0,
            "parity_argmax_match_of": 152.0,
            "parity_dp_tol": 0.01,
            "parity_forward_ms_median": 24.105937,
            "parity_max": 31.640886,
            "parity_max_logit_abs_err_vs_oracle": 0.009016335010528564,
            "parity_max_prob_abs_err": 0.001450374722480774,
            "parity_min": 23.791198,
            "parity_multi_label_equal_official": 152.0,
            "parity_multi_label_equal_official_of": 152.0,
            "parity_p90": 24.743177,
            "parity_rows": 152.0,
            "parity_single_label_equal_official": 152.0,
            "parity_single_label_equal_official_of": 152.0,
            "prepare_ms": 678.0,
            "release_close_ms": 6.0,
            "sequence_mismatches": 0.0,
            "sequence_rows": 152.0,
            "sequence_window": 256.0,
            "thermal_status_end": 0,
            "thermal_status_start": 0,
            "timing_decide_4q_total_ms_median": 102.827604,
            "timing_four_decide_1q_total_ms_median": 102.415052,
            "timing_repeats": 5.0,
            "timing_state_tokens": 21.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "CpuOptions 4 threads; hfmodels-android SDK device test (EncoderDecisions gliclass)",
          "total_ops": 670,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.2.0/2026-10-04/gliclass-edge-v3.0__gliclass_edge_v3_s256_fp32__galaxy-s26.json"
      },
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "cpu_xnnpack",
          "context_length": null,
          "date": "2026-10-02",
          "decode_tokens_per_s": null,
          "delegated_ops": 670,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-gliclass-app",
            "os_build": "Android 16 (SDK 36)",
            "runtime": "litert",
            "runtime_version": "2.2.0",
            "soc": "Qualcomm SM8850",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "Android sample app (zoo main 7cf7a55 gliclass/, mirrored as android/sample/ in the Hub repo; Kotlin host GliclassTokenizer.kt + GliclassInputs.kt + GliclassClassifier.kt + GliclassDecoder.kt), debug build, gate launched by intent extra; one process for both windows (the smallest window that fits each request); graph ms = first input write through logits readback.",
            "on-device token ids, <<LABEL>> positions and padding equal the captured official inputs on 70/70 requests of this window; single-label 70/70 and multi-label 70/70 answers equal the official pipeline; max |dlogit| 0.004972.",
            "output_match null: the app gate is a text-to-answer check against the official pipeline, not a same-bytes element comparison (adapter rule); the debug-gate rows above carry the element rule.",
            "conditions: thermal status 1 before and after (state files beside the report), USB powered, top-app.",
            "10-02 12:55:10.912 23279 23375 I tflite  : Replacing 670 out of 670 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 1 partitions for subgraph 0 (main).",
            "10-02 12:55:10.995 23279 23375 I tflite  : Replacing 670 out of 670 node(s) with delegate (TfLiteXNNPackDelegate) node, yielding 1 partitions for subgraph 0 (main)."
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": 15.649375,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "argmax_equal_official": 70.0,
            "fixtures": 70.0,
            "graph_max_ms": 22.490677,
            "graph_median_ms": 15.649375,
            "graph_min_ms": 13.434271,
            "inputs_identical_to_official": 70.0,
            "label_set_equal_official": 70.0,
            "max_abs_dlogit_vs_official": 0.004971981048583984,
            "whole_gate_requests": 552.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "CpuOptions numThreads=4 (XNNPACK); Android sample app com.gliclass (zoo gliclass/, debug build, on-device tokenizer)",
          "total_ops": 670,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.2.0/2026-10-02/gliclass-edge-v3.0__gliclass_edge_v3_s256_fp32__galaxy-s26.json"
      },
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "gpu_mldrift",
          "context_length": null,
          "date": "2026-10-02",
          "decode_tokens_per_s": null,
          "delegated_ops": 670,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-gliclass-app",
            "os_build": "Android 16 (SDK 36)",
            "runtime": "litert",
            "runtime_version": "2.2.0",
            "soc": "Qualcomm SM8850",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "Android sample app (zoo main 7cf7a55 gliclass/, mirrored as android/sample/ in the Hub repo; Kotlin host GliclassTokenizer.kt + GliclassInputs.kt + GliclassClassifier.kt + GliclassDecoder.kt), debug build, gate launched by intent extra; one process for both windows (the smallest window that fits each request); graph ms = first input write through logits readback.",
            "on-device token ids, <<LABEL>> positions and padding equal the captured official inputs on 70/70 requests of this window; single-label 70/70 and multi-label 70/70 answers equal the official pipeline; max |dlogit| 0.004961.",
            "output_match null: the app gate is a text-to-answer check against the official pipeline, not a same-bytes element comparison (adapter rule); the debug-gate rows above carry the element rule.",
            "conditions: thermal status 1 before and after (state files beside the report), USB powered, top-app.",
            "10-02 12:54:21.100 22777 22866 I tflite  : Replacing 670 out of 670 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 0 (main).",
            "10-02 12:54:21.604 22777 22866 I tflite  : Replacing 670 out of 670 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 0 (main)."
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": 8.230103999999999,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "argmax_equal_official": 70.0,
            "fixtures": 70.0,
            "graph_max_ms": 8.673282,
            "graph_median_ms": 8.230103999999999,
            "graph_min_ms": 8.019635,
            "inputs_identical_to_official": 70.0,
            "label_set_equal_official": 70.0,
            "max_abs_dlogit_vs_official": 0.0049610137939453125,
            "whole_gate_requests": 552.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "GpuOptions precision=FP32; Android sample app com.gliclass (zoo gliclass/, debug build, on-device tokenizer)",
          "total_ops": 670,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.2.0/2026-10-02/gliclass-edge-v3.0__gliclass_edge_v3_s256_fp32__galaxy-s26.json"
      },
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "gpu_mldrift",
          "context_length": null,
          "date": "2026-10-02",
          "decode_tokens_per_s": null,
          "delegated_ops": 670,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-gliclass-gate",
            "os_build": "Android 16 (SDK 36)",
            "runtime": "litert",
            "runtime_version": "2.2.0",
            "soc": "Qualcomm SM8850",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "Debug gate app com.gliclass.gate (android_gate/ of the run dir, Kotlin CompiledModel 2.2.0, Environment.create(context, DispatchLibraryDir + CompilerPluginLibraryDir = nativeLibraryDir with the Qualcomm dispatch/compiler libraries in jniLibs), one process per run, the official token ids + <<LABEL>> positions pushed as JSON, host fp16 table lookup (android.util.Half) outside the timed interval; timed interval = the three input writes + run + logits readback per request, warm median after 5 warm-up requests. Source, logcat, state / .state_after.txt.",
            "conditions: before thermal status 0, battery 37.4 C; after thermal status 0, battery 37.5 C; screen on (FLAG_KEEP_SCREEN_ON, setShowWhenLocked), USB powered, gate activity top-resumed (cgroup top-app).",
            "parity vs the official gliclass 0.1.20 pipeline (CPU fp32, transformers 5.17.0): same top label (softmax argmax) on 552/552 requests and same multi-label set (sigmoid >= 0.5) on 552/552 (419/419 boundary = reference top-1 < 0.9); max |dlogit| 0.009066; max |dp| softmax 0.00146 / sigmoid 0.00181; non-finite requests 0.",
            "output_match = shared element rule |out - ref| <= max(1e-5, 1e-3 * |ref|) over the valid label logits (6275 elements over 552 requests) against the Mac ai-edge-litert 2.1.6 CPU run of the same graph bytes on the same fp16-table inputs (recomputed by scripts/litert_compat_records.py 2026-10-02): 0 violating elements, max |d| 6.092e-05, max |d|/|ref| 0.01637.",
            "10-02 13:49:00.446  6330  6404 I tflite  : Replacing 670 out of 670 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 0 (main).",
            "10-02 13:49:00.896  6330  6404 I GliclassGate: GLICLASS_GATE compiled in 574.349062 ms"
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": 8.322448,
          "loads": true,
          "max_abs_diff": 6.091594696044922e-05,
          "max_rel_diff": 0.01637054378484428,
          "metrics": {
            "argmax_equal_official": 552.0,
            "battery_temp_c_end": 37.5,
            "battery_temp_c_start": 37.4,
            "boundary_argmax_equal_official": 419.0,
            "boundary_fixtures": 419.0,
            "element_rule_elements": 6275.0,
            "element_rule_violations": 0.0,
            "finite_fixtures": 552.0,
            "first_call_ms": 8.883854,
            "fixtures": 552.0,
            "label_set_equal_official": 552.0,
            "load_compile_ms": 574.349062,
            "max_abs_dlogit_vs_official": 0.009065747261047363,
            "max_abs_dp_sigmoid_vs_official": 0.0018071854740951965,
            "max_abs_dp_softmax_vs_official": 0.0014623336813603394,
            "thermal_status_end": 0.0,
            "thermal_status_start": 0.0,
            "warm_max_ms": 12.217187,
            "warm_median_ms": 8.322448,
            "warm_min_ms": 7.932448,
            "warmup_rows_excluded": 5.0
          },
          "output_match": true,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "GpuOptions precision=FP32; debug gate app com.gliclass.gate",
          "total_ops": 670,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.2.0/2026-10-02/gliclass-edge-v3.0__gliclass_edge_v3_s256_fp32__galaxy-s26.json"
      },
      {
        "device": "galaxy-s26",
        "run": {
          "accelerator": "gpu_mldrift",
          "context_length": null,
          "date": "2026-10-04",
          "decode_tokens_per_s": null,
          "delegated_ops": 670,
          "env": {
            "device": "Galaxy S26 (SM-S942Q)",
            "machine_label": "galaxy-s26-hfmodels-decide-gate",
            "os_build": "Android 16 (build BP4A.251205.006.S942QOPS1AZH9)",
            "runtime": "litert",
            "runtime_version": "2.2.0",
            "soc": "Qualcomm SM8850",
            "vendor_sdk": null
          },
          "error": null,
          "evidence": [
            "hfmodels-android SDK device check: androidTest GliclassDeviceTest of the litert module (git 29bc04c), run by tools/decide_gate.sh on the Galaxy S26, one instrumentation process per run; the SDK (hfmodels-litert EncoderDecisions, family gliclass) resolved litert-community/GLiClass-Edge-v3.0-LiteRT through the dev descriptor catalog/dev/litert-community__GLiClass-Edge-v3.0-LiteRT.hfmodels.json at that commit (the log's descriptor sha256 prefix afe7d983 = `git show 29bc04c:catalog/dev/litert-community__GLiClass-Edge-v3.0-LiteRT.hfmodels.json`); graph gliclass_edge_v3_s256_fp32.tflite sha256 aff0ee9795a2ffc2d4d32385e8f804b30bac4282d96d19c1e53cac88dcc97708 = the published gliclass_edge_v3_s256_fp32.tflite (litert-community/GLiClass-Edge-v3.0-LiteRT@88c90950; other records hold rows of the same name and digest).",
            "# hfmodels typed-decisions device check  date=2026-10-04T01:29:06Z serial=RFGL80R6A6H device=SM-S942Q build=BP4A.251205.006.S942QOPS1AZH9 android=16",
            "# git=29bc04c test=GliclassDeviceTest variant=s256_fp32 backend=gpu thermal=Thermal Status: 0 battery_level= level: 79 battery_temp_x10C= temperature: 361 screen= mWakefulness=Dozing",
            "# gradle rc=0 end=2026-10-04T01:29:14Z thermal_end=Thermal Status: 0 battery_temp_end_x10C= temperature: 359",
            "resolve litert-community/GLiClass-Edge-v3.0-LiteRT: commit=88c90950 via EXPLICIT_REVISION, descriptor=caller@explicit/hfmodels.json sha=afe7d983",
            "imported gliclass_edge_v3_s256_fp32.tflite: 53965476 bytes, sha256 ok, from /data/local/tmp/hfmodels/gliclass/gliclass_edge_v3_s256_fp32.tflite",
            "ready litert-community/GLiClass-Edge-v3.0-LiteRT@88c90950 variant=s256_fp32 profile=gpu inference=GPU prepare_ms=1385",
            "litert  : [compiled_model.cc:1250] 0 compiler plugins were applied successfully: ",
            "tflite  : Replacing 670 out of 670 node(s) with delegate (LITERT_CL) node, yielding 1 partitions for subgraph 0 (main).",
            "RESULT step=load ok=true variant=s256_fp32 backend=gpu import_ms=168 load_ms=1397 compile_ms=887 profile=gpu window=256 head=256 max_options=25 notes=family=gliclass compile_ms=887 on gpu (precision fp32) | observed backend is UNKNOWN: litert 2.2.0 reports no per-op execution target; 'initialized' is the accelerator the CompiledModel was created with | host token lookup: tok_embeddings_fp16.bin (float16, 50370 rows x 384) feeds inputs_embeds | one gliclass request per question: labels = a choice's descriptions (the key when there is none), a score's levels, a noul's false / true descriptions or no / yes; prompt = the instructions plus one space, right before the text; a request longer than the 256-token window or with more than 25 labels is refused, never cut; probabilities are gliclass's single-label float32 softmax of the label logits (T=1)",
            "RESULT step=sequence ok=true variant=s256_fp32 backend=gpu rows=152 of 152 mismatches=0 window=256",
            "RESULT step=parity ok=true variant=s256_fp32 backend=gpu rows=152 argmax_match=152/152 single_label_equal_official=152/152 multi_label_equal_official=152/152 max_prob_abs_err=0.0014623254537582397 (worst semif_135) max_logit_abs_err_vs_oracle=0.009065747261047363 dp_tol=0.01 forward_ms_median=10.000468 p90=10.397083 min=8.353646 max=13.288438",
            "RESULT step=demo ok=true variant=s256_fp32 backend=gpu form=described rows=30 same_answer_as_mac=30/30 max_prob_abs_err_vs_mac=4.738569259643555E-6 (worst b06) encoded_length_equal_host=30/30 dp_tol=0.01 correct_device=10/30 correct_mac=10/30 question_ms_median=14.675626000000001 p90=15.405000000000001 min=9.565052 max=17.843854999999998 lookup_ms_median=0.904792",
            "RESULT step=demo ok=true variant=s256_fp32 backend=gpu form=keys rows=30 same_answer_as_mac=30/30 max_prob_abs_err_vs_mac=1.5497207641601562E-6 (worst b25) encoded_length_equal_host=30/30 dp_tol=0.01 correct_device=10/30 correct_mac=10/30 question_ms_median=11.808177 p90=12.905365 min=10.82177 max=15.454843 lookup_ms_median=0.887917",
            "RESULT step=timing ok=true variant=s256_fp32 backend=gpu repeats=5 decide_4q_total_ms_median=48.985677 four_decide_1q_total_ms_median=50.260677 state_tokens=21 per_question_ms=12.907656,12.314688,11.415626,11.736719 answers={department=billing 0.9799990653991699, urgency=0.11925097089260817, refund_requested=0.33966851234436035, tone=angry 0.7151775360107422}",
            "RESULT step=lookup ok=true variant=s256_fp32 backend=gpu rows=128 hidden=384 zoo_style=lut zoo_ms_median=1.927136 zoo_ms_min=1.593333 sdk_ms_median=0.094896 sdk_ms_min=0.094427 repeats=20 outputs_identical=true",
            "RESULT step=release ok=true variant=s256_fp32 backend=gpu close_ms=10",
            "RESULT ok=true model=litert-community/GLiClass-Edge-v3.0-LiteRT variant=s256_fp32 backend=gpu device=SM-S942Q build=BP4A.251205.006.S942QOPS1AZH9 failures=",
            "latency_p50_ms = the parity step's forward_ms_median: one LiteRtDecisionModel.scores call per captured request (host float16 table lookup, input writes, CompiledModel run, output readback on the model thread), after one warm forward.",
            "output_match null: the gate compares with the publisher's outputs under its own criterion (argmax and max |dp| <= 0.01, in metrics) and keeps no raw per-element outputs. full_delegation from the delegate line the gate's logcat filter now keeps ('Replacing 670 out of 670 node(s) with delegate (LITERT_CL)').",
            "conditions (the gate's header and footer): thermal status 0 -> 0, battery 36.1 -> 35.9 C, battery level 79 %, display mWakefulness=Dozing at the start.",
            "source: hfmodels-android/litert/results/2026-10-04-1029-RFGL80R6A6H-litert2.2.0-decide-gliclass-s256_fp32-gpu.log (main 3e1fcd8)."
          ],
          "failure_class": null,
          "full_delegation": true,
          "latency_p50_ms": 10.000468,
          "loads": true,
          "max_abs_diff": null,
          "max_rel_diff": null,
          "metrics": {
            "battery_level_start": 79,
            "battery_temp_c_end": 35.9,
            "battery_temp_c_start": 36.1,
            "demo_described_correct_device": 10.0,
            "demo_described_correct_device_of": 30.0,
            "demo_described_correct_mac": 10.0,
            "demo_described_correct_mac_of": 30.0,
            "demo_described_dp_tol": 0.01,
            "demo_described_encoded_length_equal_host": 30.0,
            "demo_described_encoded_length_equal_host_of": 30.0,
            "demo_described_lookup_ms_median": 0.904792,
            "demo_described_max": 17.843854999999998,
            "demo_described_max_prob_abs_err_vs_mac": 4.738569259643555e-06,
            "demo_described_min": 9.565052,
            "demo_described_p90": 15.405000000000001,
            "demo_described_question_ms_median": 14.675626000000001,
            "demo_described_rows": 30.0,
            "demo_described_same_answer_as_mac": 30.0,
            "demo_described_same_answer_as_mac_of": 30.0,
            "demo_keys_correct_device": 10.0,
            "demo_keys_correct_device_of": 30.0,
            "demo_keys_correct_mac": 10.0,
            "demo_keys_correct_mac_of": 30.0,
            "demo_keys_dp_tol": 0.01,
            "demo_keys_encoded_length_equal_host": 30.0,
            "demo_keys_encoded_length_equal_host_of": 30.0,
            "demo_keys_lookup_ms_median": 0.887917,
            "demo_keys_max": 15.454843,
            "demo_keys_max_prob_abs_err_vs_mac": 1.5497207641601562e-06,
            "demo_keys_min": 10.82177,
            "demo_keys_p90": 12.905365,
            "demo_keys_question_ms_median": 11.808177,
            "demo_keys_rows": 30.0,
            "demo_keys_same_answer_as_mac": 30.0,
            "demo_keys_same_answer_as_mac_of": 30.0,
            "load_compile_ms": 887.0,
            "load_head": 256.0,
            "load_import_ms": 168.0,
            "load_load_ms": 1397.0,
            "load_max_options": 25.0,
            "load_window": 256.0,
            "lookup_hidden": 384.0,
            "lookup_repeats": 20.0,
            "lookup_rows": 128.0,
            "lookup_sdk_ms_median": 0.094896,
            "lookup_sdk_ms_min": 0.094427,
            "lookup_zoo_ms_median": 1.927136,
            "lookup_zoo_ms_min": 1.593333,
            "parity_argmax_match": 152.0,
            "parity_argmax_match_of": 152.0,
            "parity_dp_tol": 0.01,
            "parity_forward_ms_median": 10.000468,
            "parity_max": 13.288438,
            "parity_max_logit_abs_err_vs_oracle": 0.009065747261047363,
            "parity_max_prob_abs_err": 0.0014623254537582397,
            "parity_min": 8.353646,
            "parity_multi_label_equal_official": 152.0,
            "parity_multi_label_equal_official_of": 152.0,
            "parity_p90": 10.397083,
            "parity_rows": 152.0,
            "parity_single_label_equal_official": 152.0,
            "parity_single_label_equal_official_of": 152.0,
            "prepare_ms": 1385.0,
            "release_close_ms": 10.0,
            "sequence_mismatches": 0.0,
            "sequence_rows": 152.0,
            "sequence_window": 256.0,
            "thermal_status_end": 0,
            "thermal_status_start": 0,
            "timing_decide_4q_total_ms_median": 48.985677,
            "timing_four_decide_1q_total_ms_median": 50.260677,
            "timing_repeats": 5.0,
            "timing_state_tokens": 21.0
          },
          "output_match": null,
          "peak_mem_mb": null,
          "prefill_tokens_per_s": null,
          "provenance": "measured",
          "runs": true,
          "signature": "GpuOptions precision=FP32; hfmodels-android SDK device test (EncoderDecisions gliclass)",
          "total_ops": 670,
          "ttft_ms": null
        },
        "source": "data/device_runs/2.2.0/2026-10-04/gliclass-edge-v3.0__gliclass_edge_v3_s256_fp32__galaxy-s26.json"
      }
    ]
  },
  "model": {
    "family": "gliclass-edge-v3.0",
    "id": "gliclass-edge-v3.0__gliclass_edge_v3_s256_fp32",
    "license": "apache-2.0",
    "source_url": "https://huggingface.co/litert-community/GLiClass-Edge-v3.0-LiteRT",
    "task": "zero-shot-classification"
  },
  "pitfalls": [
    "Weight storage does not set the GPU compute precision: request GpuOptions(precision = FP32). At the default precision the outputs stay finite but the answers move: Galaxy S26 475/482 top labels and 455/482 label sets, Mac Metal 479/469 (card 'Galaxy S26, gate app' and 'fp16, the GPU default precision and the NPU').",
    "The graph takes host-built inputs: inputs_embeds [1,256,384] from host_assets/tok_embeddings_fp16.bin ([50370,384] float16 rows upcast to float32, [PAD] 50283 rows at padded positions), attention_mask [1,256] (1 real / 0 pad), label_routing [1,25,256] (row k = 1 at the k-th <<LABEL>> token); output logits [1,1,1,25], slots past the request's labels are ignored; softmax (single-label) or sigmoid >= threshold (multi-label) runs on the host (card 'Host contract').",
    "The request string is <<LABEL>>label 1<<LABEL>>label 2…<<SEP>> + prompt + text with nothing between the prompt and the text; tokenize with the source tokenizer.json ([CLS] … [SEP], ByteLevel prefix space per piece between added tokens); at most 25 labels and 256 tokens, longer requests raise an error, nothing is truncated (card 'Host contract', 'Limits').",
    "The shipped graph computes 15 LayerNorms on x*2^-k with eps*2^-2k (k = 5 in layers 3-6, 6 in layers 7-9 and the final norm) and uses a -1e4 mask; this is exact in FP32 (bit-identical logits to the unmodified graph) and keeps fp16 computation finite but not exact: the Galaxy S26 Hexagon NPU (JIT, one partition, 2.51 ms) keeps 480/482 top labels and 462/482 label sets, every change a near tie, so the card reports the NPU and does not call it a validated path (card 'fp16, the GPU default precision and the NPU').",
    "The checkpoint is float32: the float16 host table rounds values by up to 1.2e-4 (max logit difference 0.0090 against 0.000036 with a float32 table, same answers on all 552 requests) (card 'fp16 …').",
    "Validated on one desktop (Apple M4 Max, ai-edge-litert 2.1.6 CPU and Metal explicit FP32) and one phone (Galaxy S26, LiteRT 2.2.0 GPU explicit FP32 and CPU); English only; conversion fidelity to the official gliclass pipeline (552/552 answers), not a task-accuracy claim — own-subset gold agreement 67/144 SemIf, 115/200 ag_news, 71/200 banking77 (25-way) is not the publisher's benchmark (card 'Agreement with the dataset labels', 'Limits')."
  ],
  "schema_version": "1.2"
}
