{ "config": { "input_seed": 42, "timed_runs": 10, "tolerance_abs": 0.00001, "tolerance_rel": 0.001, "warmup_runs": 3 }, "input_spec": "1x1x896:float32;1:int32;1:float32;1:int32;1x1x1x10:float32;4x1x2x10x64:float32;4x1x2x10x64:float32", "license": "apache-2.0", "model_id": "audio8-tts-preview-0.6b__fast_ar_int8", "notes": "Fast AR step graph of the Audio8 DualAR TTS (4 layers, one codebook token per call, 10 calls per frame): inputs slow hidden state, token id, use_hidden flag, position, additive mask [1,1,1,10] and two rank-5 KV caches [4,1,2,10,64]; random int32 token/position inputs and random caches measure the graph, not speech; dynamic int8 FULLY_CONNECTED + int8 embedding tables; 68 MB", "results": [ { "backend": "wasm_xnnpack", "date": "2026-09-28", "delegation_evidence": [ "error: WARNING: [npu_registry.cc:34] NPU accelerator could not be loaded and registered: kLiteRtStatusErrorInvalidArgument.", "error: WARNING: [npu_registry.cc:34] NPU accelerator could not be loaded and registered: kLiteRtStatusErrorInvalidArgument." ], "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" } }, "error": null, "failure_class": null, "full_delegation": null, "latency_p50_ms": 2.48, "loads": true, "max_abs_diff": null, "max_rel_diff": null, "output_match": null, "provenance": "measured", "runs": true }, { "backend": "webgpu_mldrift", "date": "2026-09-28", "delegation_evidence": [ "error: WARNING: [npu_registry.cc:34] NPU accelerator could not be loaded and registered: kLiteRtStatusErrorInvalidArgument.", "error: WARNING: [npu_registry.cc:34] NPU accelerator could not be loaded and registered: kLiteRtStatusErrorInvalidArgument.", "error: ERROR: Following operations are not supported by GPU delegate:", "error: ADD: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m/arktts_port.ArBlock_0;39", "error: ADD: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m/arktts_port.ArBlock_1;32", "error: ADD: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m/arktts_port.ArBlock_2;32", "error: ADD: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m/arktts_port.ArBlock_3;32", "error: GATHER_ND: Operation is not supported.", "error: RESHAPE: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m/arktts_port.ArBlock_0;39", "error: RESHAPE: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m/arktts_port.ArBlock_0;40", "error: RESHAPE: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m/arktts_port.ArBlock_0;41", "error: RESHAPE: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m/arktts_port.ArBlock_1;32", "error: RESHAPE: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m/arktts_port.ArBlock_1;;__main__.Wrap/arktts_port.FastAR_m/arktts_port.ArBlock_0;4", "error: RESHAPE: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m/arktts_port.ArBlock_2;32", "error: RESHAPE: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m/arktts_port.ArBlock_2;;__main__.Wrap/arktts_port.FastAR_m/arktts_port.ArBlock_0;4", "error: RESHAPE: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m/arktts_port.ArBlock_3;32", "error: RESHAPE: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m/arktts_port.ArBlock_3;;__main__.Wrap/arktts_port.FastAR_m/arktts_port.ArBlock_0;4", "error: RESHAPE: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m;24", "error: RESHAPE: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m;26", "error: RESHAPE: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m;28", "error: RESHAPE: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m;30", "error: RESHAPE: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m;32", "error: RESHAPE: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m;34", "error: RESHAPE: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m;36", "error: RESHAPE: Tensor dimensions must be less than 5. __main__.Wrap/arktts_port.FastAR_m;38", "error: RESHAPE: Tensor dimensions must be less than 5. k_all", "error: RESHAPE: Tensor dimensions must be less than 5. v_all", "error: SLICE: Slicing is unsupported for input with rank > 4", "error: 58 operations will run on the GPU, and the remaining 225 operations will run on the CPU.", "warning: %c[LiteRT]%c Model not fully compiled for webgpu. Partially delegating to WASM execution. background: #FFA000; color: black; font-weight: bold; padding: 2px 5px; border-radius: 3px; font-weight: bold;" ], "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" } }, "error": null, "failure_class": null, "full_delegation": false, "latency_p50_ms": 3.423, "loads": true, "max_abs_diff": 0.8894538879394531, "max_rel_diff": 168.63736206304307, "output_match": false, "provenance": "measured", "runs": true } ], "schema_version": "1.0", "source": "https://huggingface.co/litert-community/Audio8-TTS-Preview-0.6b/resolve/main/fast_ar_int8.tflite" }