@elizaos/plugin-local-inference 2.0.0-beta.1 → 2.0.11-beta.7
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/LICENSE +21 -0
- package/README.md +83 -0
- package/package.json +81 -15
- package/src/actions/generate-media.d.ts +59 -0
- package/src/actions/generate-media.d.ts.map +1 -0
- package/src/actions/generate-media.ts +647 -0
- package/src/actions/identify-speaker.d.ts +23 -0
- package/src/actions/identify-speaker.d.ts.map +1 -0
- package/src/actions/identify-speaker.ts +171 -0
- package/src/adapters/capacitor-llama/__tests__/compat-behavior.test.ts +218 -0
- package/src/adapters/capacitor-llama/__tests__/index.test.ts +68 -0
- package/src/adapters/capacitor-llama/__tests__/structured-output.test.ts +215 -0
- package/src/adapters/capacitor-llama/__tests__/text-streaming.test.ts +174 -0
- package/src/adapters/capacitor-llama/environment.ts +71 -0
- package/src/adapters/capacitor-llama/index.browser.ts +83 -0
- package/src/adapters/capacitor-llama/index.ts +807 -0
- package/src/adapters/capacitor-llama/loader.ts +109 -0
- package/src/adapters/capacitor-llama/structured-output.ts +165 -0
- package/src/adapters/capacitor-llama/text-streaming.ts +227 -0
- package/src/adapters/capacitor-llama/types.ts +374 -0
- package/src/backends/apple-foundation.ts +127 -0
- package/src/index.d.ts +7 -0
- package/src/index.d.ts.map +1 -0
- package/src/index.ts +54 -0
- package/src/local-inference-routes.d.ts +38 -0
- package/src/local-inference-routes.d.ts.map +1 -0
- package/src/local-inference-routes.test.ts +344 -0
- package/src/local-inference-routes.ts +1543 -0
- package/src/provider.d.ts +21 -0
- package/src/provider.d.ts.map +1 -0
- package/src/provider.ts +1171 -0
- package/src/routes/compat-helpers.d.ts +18 -0
- package/src/routes/compat-helpers.d.ts.map +1 -0
- package/src/routes/compat-helpers.ts +274 -0
- package/src/routes/family-member-route.d.ts +62 -0
- package/src/routes/family-member-route.d.ts.map +1 -0
- package/src/routes/family-member-route.ts +353 -0
- package/src/routes/index.d.ts +19 -0
- package/src/routes/index.d.ts.map +1 -0
- package/src/routes/index.ts +60 -0
- package/src/routes/live-diarization-route.d.ts +26 -0
- package/src/routes/live-diarization-route.d.ts.map +1 -0
- package/src/routes/live-diarization-route.test.ts +213 -0
- package/src/routes/live-diarization-route.ts +122 -0
- package/src/routes/local-inference-asr-route.d.ts +4 -0
- package/src/routes/local-inference-asr-route.d.ts.map +1 -0
- package/src/routes/local-inference-asr-route.test.ts +190 -0
- package/src/routes/local-inference-asr-route.ts +213 -0
- package/src/routes/local-inference-compat-routes.d.ts +16 -0
- package/src/routes/local-inference-compat-routes.d.ts.map +1 -0
- package/src/routes/local-inference-compat-routes.test.ts +423 -0
- package/src/routes/local-inference-compat-routes.ts +782 -0
- package/src/routes/local-inference-tts-route.d.ts +7 -0
- package/src/routes/local-inference-tts-route.d.ts.map +1 -0
- package/src/routes/local-inference-tts-route.test.ts +179 -0
- package/src/routes/local-inference-tts-route.ts +230 -0
- package/src/routes/voice-first-run-routes.d.ts +62 -0
- package/src/routes/voice-first-run-routes.d.ts.map +1 -0
- package/src/routes/voice-first-run-routes.ts +524 -0
- package/src/routes/voice-models-routes.d.ts +62 -0
- package/src/routes/voice-models-routes.d.ts.map +1 -0
- package/src/routes/voice-models-routes.ts +554 -0
- package/src/routes/voice-profile-plugin-routes.d.ts +19 -0
- package/src/routes/voice-profile-plugin-routes.d.ts.map +1 -0
- package/src/routes/voice-profile-plugin-routes.ts +138 -0
- package/src/routes/voice-profiles-management-routes.d.ts +52 -0
- package/src/routes/voice-profiles-management-routes.d.ts.map +1 -0
- package/src/routes/voice-profiles-management-routes.ts +476 -0
- package/src/routes/voice-speaker-profile-routes.d.ts +57 -0
- package/src/routes/voice-speaker-profile-routes.d.ts.map +1 -0
- package/src/routes/voice-speaker-profile-routes.ts +199 -0
- package/src/runtime/aosp-llama-loader-selection.test.ts +80 -0
- package/src/runtime/capacitor-llama.d.ts +25 -0
- package/src/runtime/embedding-manager-support.d.ts +77 -0
- package/src/runtime/embedding-manager-support.d.ts.map +1 -0
- package/src/runtime/embedding-manager-support.ts +497 -0
- package/src/runtime/embedding-presets.d.ts +16 -0
- package/src/runtime/embedding-presets.d.ts.map +1 -0
- package/src/runtime/embedding-presets.ts +81 -0
- package/src/runtime/embedding-warmup-policy.d.ts +14 -0
- package/src/runtime/embedding-warmup-policy.d.ts.map +1 -0
- package/src/runtime/embedding-warmup-policy.test.ts +53 -0
- package/src/runtime/embedding-warmup-policy.ts +48 -0
- package/src/runtime/ensure-local-inference-handler.d.ts +53 -0
- package/src/runtime/ensure-local-inference-handler.d.ts.map +1 -0
- package/src/runtime/ensure-local-inference-handler.test.ts +528 -0
- package/src/runtime/ensure-local-inference-handler.ts +1398 -0
- package/src/runtime/index.d.ts +14 -0
- package/src/runtime/index.d.ts.map +1 -0
- package/src/runtime/index.ts +27 -0
- package/src/runtime/mobile-local-inference-gate.d.ts +31 -0
- package/src/runtime/mobile-local-inference-gate.d.ts.map +1 -0
- package/src/runtime/mobile-local-inference-gate.test.ts +69 -0
- package/src/runtime/mobile-local-inference-gate.ts +44 -0
- package/src/runtime/voice-entity-binding.d.ts +103 -0
- package/src/runtime/voice-entity-binding.d.ts.map +1 -0
- package/src/runtime/voice-entity-binding.transcript.test.ts +69 -0
- package/src/runtime/voice-entity-binding.ts +328 -0
- package/src/services/README.md +71 -0
- package/src/services/__tests__/backend-selector.test.ts +101 -0
- package/src/services/__tests__/checkpoint-manager.test.ts +376 -0
- package/src/services/__tests__/gpu-autotune.test.ts +400 -0
- package/src/services/__tests__/llm-streaming-binding.test.ts +85 -0
- package/src/services/__tests__/planner-grammar.test.ts +372 -0
- package/src/services/__tests__/runtime-target.test.ts +176 -0
- package/src/services/active-model-switch-rollback.test.ts +183 -0
- package/src/services/active-model.d.ts +282 -0
- package/src/services/active-model.d.ts.map +1 -0
- package/src/services/active-model.ts +1213 -0
- package/src/services/asr/errors.d.ts +21 -0
- package/src/services/asr/errors.d.ts.map +1 -0
- package/src/services/asr/errors.ts +50 -0
- package/src/services/asr/hash.d.ts +28 -0
- package/src/services/asr/hash.d.ts.map +1 -0
- package/src/services/asr/hash.ts +49 -0
- package/src/services/asr/index.d.ts +76 -0
- package/src/services/asr/index.d.ts.map +1 -0
- package/src/services/asr/index.ts +178 -0
- package/src/services/asr/types.d.ts +91 -0
- package/src/services/asr/types.d.ts.map +1 -0
- package/src/services/asr/types.ts +95 -0
- package/src/services/assignments.d.ts +71 -0
- package/src/services/assignments.d.ts.map +1 -0
- package/src/services/assignments.test.ts +80 -0
- package/src/services/assignments.ts +230 -0
- package/src/services/backend-selector.ts +95 -0
- package/src/services/backend.d.ts +346 -0
- package/src/services/backend.d.ts.map +1 -0
- package/src/services/backend.ts +612 -0
- package/src/services/bundled-models.d.ts +34 -0
- package/src/services/bundled-models.d.ts.map +1 -0
- package/src/services/bundled-models.ts +129 -0
- package/src/services/cache-bridge.d.ts +206 -0
- package/src/services/cache-bridge.d.ts.map +1 -0
- package/src/services/cache-bridge.test.ts +516 -0
- package/src/services/cache-bridge.ts +423 -0
- package/src/services/catalog.d.ts +10 -0
- package/src/services/catalog.d.ts.map +1 -0
- package/src/services/catalog.test.ts +240 -0
- package/src/services/catalog.ts +27 -0
- package/src/services/checkpoint-client.d.ts +109 -0
- package/src/services/checkpoint-client.d.ts.map +1 -0
- package/src/services/checkpoint-client.ts +258 -0
- package/src/services/checkpoint-manager.ts +474 -0
- package/src/services/cloud-fallback.d.ts +102 -0
- package/src/services/cloud-fallback.d.ts.map +1 -0
- package/src/services/cloud-fallback.ts +230 -0
- package/src/services/conversation-registry.d.ts +142 -0
- package/src/services/conversation-registry.d.ts.map +1 -0
- package/src/services/conversation-registry.test.ts +235 -0
- package/src/services/conversation-registry.ts +264 -0
- package/src/services/desktop-fused-ffi-backend-runtime.d.ts +92 -0
- package/src/services/desktop-fused-ffi-backend-runtime.d.ts.map +1 -0
- package/src/services/desktop-fused-ffi-backend-runtime.ts +333 -0
- package/src/services/device-bridge.d.ts +188 -0
- package/src/services/device-bridge.d.ts.map +1 -0
- package/src/services/device-bridge.ts +1237 -0
- package/src/services/device-resource-metrics.d.ts +149 -0
- package/src/services/device-resource-metrics.d.ts.map +1 -0
- package/src/services/device-resource-metrics.test.ts +98 -0
- package/src/services/device-resource-metrics.ts +346 -0
- package/src/services/device-tier.d.ts +115 -0
- package/src/services/device-tier.d.ts.map +1 -0
- package/src/services/device-tier.test.ts +371 -0
- package/src/services/device-tier.ts +410 -0
- package/src/services/downloader.d.ts +82 -0
- package/src/services/downloader.d.ts.map +1 -0
- package/src/services/downloader.test.ts +724 -0
- package/src/services/downloader.ts +899 -0
- package/src/services/engine-direct-bundle.test.ts +58 -0
- package/src/services/engine-streaming.test.ts +80 -0
- package/src/services/engine.d.ts +534 -0
- package/src/services/engine.d.ts.map +1 -0
- package/src/services/engine.ts +1891 -0
- package/src/services/ensure-local-artifacts.integration.test.ts +273 -0
- package/src/services/ensure-local-artifacts.test.ts +368 -0
- package/src/services/ensure-local-artifacts.ts +351 -0
- package/src/services/external-scanner.d.ts +17 -0
- package/src/services/external-scanner.d.ts.map +1 -0
- package/src/services/external-scanner.ts +312 -0
- package/src/services/ffi-llm-mock.ts +354 -0
- package/src/services/ffi-llm-streaming-abi.ts +442 -0
- package/src/services/ffi-streaming-backend.d.ts +180 -0
- package/src/services/ffi-streaming-backend.d.ts.map +1 -0
- package/src/services/ffi-streaming-backend.ts +382 -0
- package/src/services/ffi-streaming-runner.d.ts +122 -0
- package/src/services/ffi-streaming-runner.d.ts.map +1 -0
- package/src/services/ffi-streaming-runner.test.ts +60 -0
- package/src/services/ffi-streaming-runner.ts +354 -0
- package/src/services/ffi-unload-ordering.test.ts +162 -0
- package/src/services/gpu-autotune.ts +534 -0
- package/src/services/gpu-detect.ts +139 -0
- package/src/services/handler-registry.d.ts +72 -0
- package/src/services/handler-registry.d.ts.map +1 -0
- package/src/services/handler-registry.ts +240 -0
- package/src/services/hardware.d.ts +63 -0
- package/src/services/hardware.d.ts.map +1 -0
- package/src/services/hardware.test.ts +183 -0
- package/src/services/hardware.ts +404 -0
- package/src/services/hf-search.d.ts +26 -0
- package/src/services/hf-search.d.ts.map +1 -0
- package/src/services/hf-search.test.ts +69 -0
- package/src/services/hf-search.ts +420 -0
- package/src/services/image-description-runtime.d.ts +14 -0
- package/src/services/image-description-runtime.d.ts.map +1 -0
- package/src/services/image-description-runtime.test.ts +61 -0
- package/src/services/image-description-runtime.ts +118 -0
- package/src/services/imagegen/aosp-unavailable.d.ts +134 -0
- package/src/services/imagegen/aosp-unavailable.d.ts.map +1 -0
- package/src/services/imagegen/aosp-unavailable.ts +229 -0
- package/src/services/imagegen/backend-selector.d.ts +118 -0
- package/src/services/imagegen/backend-selector.d.ts.map +1 -0
- package/src/services/imagegen/backend-selector.ts +281 -0
- package/src/services/imagegen/coreml-unavailable.d.ts +105 -0
- package/src/services/imagegen/coreml-unavailable.d.ts.map +1 -0
- package/src/services/imagegen/coreml-unavailable.ts +237 -0
- package/src/services/imagegen/errors.d.ts +16 -0
- package/src/services/imagegen/errors.d.ts.map +1 -0
- package/src/services/imagegen/errors.ts +40 -0
- package/src/services/imagegen/index.d.ts +58 -0
- package/src/services/imagegen/index.d.ts.map +1 -0
- package/src/services/imagegen/index.ts +144 -0
- package/src/services/imagegen/mflux.d.ts +74 -0
- package/src/services/imagegen/mflux.d.ts.map +1 -0
- package/src/services/imagegen/mflux.ts +313 -0
- package/src/services/imagegen/sd-cpp.d.ts +180 -0
- package/src/services/imagegen/sd-cpp.d.ts.map +1 -0
- package/src/services/imagegen/sd-cpp.ts +718 -0
- package/src/services/imagegen/tensorrt-unavailable.d.ts +83 -0
- package/src/services/imagegen/tensorrt-unavailable.d.ts.map +1 -0
- package/src/services/imagegen/tensorrt-unavailable.ts +295 -0
- package/src/services/imagegen/types.d.ts +181 -0
- package/src/services/imagegen/types.d.ts.map +1 -0
- package/src/services/imagegen/types.ts +193 -0
- package/src/services/index.d.ts +30 -0
- package/src/services/index.d.ts.map +1 -0
- package/src/services/index.ts +225 -0
- package/src/services/inference-capabilities.d.ts +132 -0
- package/src/services/inference-capabilities.d.ts.map +1 -0
- package/src/services/inference-capabilities.test.ts +75 -0
- package/src/services/inference-capabilities.ts +204 -0
- package/src/services/inference-telemetry.d.ts +59 -0
- package/src/services/inference-telemetry.d.ts.map +1 -0
- package/src/services/inference-telemetry.ts +143 -0
- package/src/services/ios-llama-streaming.ts +248 -0
- package/src/services/kv-spill.d.ts +189 -0
- package/src/services/kv-spill.d.ts.map +1 -0
- package/src/services/kv-spill.test.ts +222 -0
- package/src/services/kv-spill.ts +356 -0
- package/src/services/latency-trace.d.ts +346 -0
- package/src/services/latency-trace.d.ts.map +1 -0
- package/src/services/latency-trace.test.ts +266 -0
- package/src/services/latency-trace.ts +844 -0
- package/src/services/llama-server-metrics.ts +304 -0
- package/src/services/llm-streaming-binding.d.ts +96 -0
- package/src/services/llm-streaming-binding.d.ts.map +1 -0
- package/src/services/llm-streaming-binding.ts +136 -0
- package/src/services/load-args.d.ts +82 -0
- package/src/services/load-args.d.ts.map +1 -0
- package/src/services/load-args.ts +81 -0
- package/src/services/manifest/eliza-1.manifest.v1.json +708 -0
- package/src/services/manifest/index.d.ts +4 -0
- package/src/services/manifest/index.d.ts.map +1 -0
- package/src/services/manifest/index.ts +66 -0
- package/src/services/manifest/manifest.test.ts +693 -0
- package/src/services/manifest/schema.d.ts +715 -0
- package/src/services/manifest/schema.d.ts.map +1 -0
- package/src/services/manifest/schema.ts +655 -0
- package/src/services/manifest/types.d.ts +30 -0
- package/src/services/manifest/types.d.ts.map +1 -0
- package/src/services/manifest/types.ts +55 -0
- package/src/services/manifest/validator.d.ts +66 -0
- package/src/services/manifest/validator.d.ts.map +1 -0
- package/src/services/manifest/validator.ts +569 -0
- package/src/services/memory-arbiter.d.ts +343 -0
- package/src/services/memory-arbiter.d.ts.map +1 -0
- package/src/services/memory-arbiter.test.ts +419 -0
- package/src/services/memory-arbiter.ts +1000 -0
- package/src/services/memory-monitor.d.ts +119 -0
- package/src/services/memory-monitor.d.ts.map +1 -0
- package/src/services/memory-monitor.test.ts +208 -0
- package/src/services/memory-monitor.ts +296 -0
- package/src/services/memory-pressure.d.ts +127 -0
- package/src/services/memory-pressure.d.ts.map +1 -0
- package/src/services/memory-pressure.ts +413 -0
- package/src/services/mtp-doctor.d.ts +13 -0
- package/src/services/mtp-doctor.d.ts.map +1 -0
- package/src/services/mtp-doctor.ts +78 -0
- package/src/services/network-policy.d.ts +127 -0
- package/src/services/network-policy.d.ts.map +1 -0
- package/src/services/network-policy.ts +346 -0
- package/src/services/paths.d.ts +6 -0
- package/src/services/paths.d.ts.map +1 -0
- package/src/services/paths.ts +25 -0
- package/src/services/planner-skeleton.d.ts +124 -0
- package/src/services/planner-skeleton.d.ts.map +1 -0
- package/src/services/planner-skeleton.ts +175 -0
- package/src/services/providers.d.ts +38 -0
- package/src/services/providers.d.ts.map +1 -0
- package/src/services/providers.ts +507 -0
- package/src/services/ram-budget-cache.test.ts +163 -0
- package/src/services/ram-budget.d.ts +110 -0
- package/src/services/ram-budget.d.ts.map +1 -0
- package/src/services/ram-budget.ts +0 -0
- package/src/services/readiness.d.ts +9 -0
- package/src/services/readiness.d.ts.map +1 -0
- package/src/services/readiness.test.ts +87 -0
- package/src/services/readiness.ts +238 -0
- package/src/services/recommendation.d.ts +111 -0
- package/src/services/recommendation.d.ts.map +1 -0
- package/src/services/recommendation.ts +672 -0
- package/src/services/registry.d.ts +35 -0
- package/src/services/registry.d.ts.map +1 -0
- package/src/services/registry.ts +151 -0
- package/src/services/router-handler.d.ts +92 -0
- package/src/services/router-handler.d.ts.map +1 -0
- package/src/services/router-handler.test.ts +45 -0
- package/src/services/router-handler.ts +376 -0
- package/src/services/routing-policy.d.ts +55 -0
- package/src/services/routing-policy.d.ts.map +1 -0
- package/src/services/routing-policy.ts +228 -0
- package/src/services/routing-preferences.d.ts +8 -0
- package/src/services/routing-preferences.d.ts.map +1 -0
- package/src/services/routing-preferences.ts +15 -0
- package/src/services/runtime-target.d.ts +98 -0
- package/src/services/runtime-target.d.ts.map +1 -0
- package/src/services/runtime-target.ts +154 -0
- package/src/services/service.d.ts +128 -0
- package/src/services/service.d.ts.map +1 -0
- package/src/services/service.test.ts +223 -0
- package/src/services/service.ts +735 -0
- package/src/services/session-pool.d.ts +72 -0
- package/src/services/session-pool.d.ts.map +1 -0
- package/src/services/session-pool.ts +153 -0
- package/src/services/structured-output/deterministic-repair.d.ts +23 -0
- package/src/services/structured-output/deterministic-repair.d.ts.map +1 -0
- package/src/services/structured-output/deterministic-repair.test.ts +169 -0
- package/src/services/structured-output/deterministic-repair.ts +443 -0
- package/src/services/structured-output/index.ts +4 -0
- package/src/services/structured-output.d.ts +311 -0
- package/src/services/structured-output.d.ts.map +1 -0
- package/src/services/structured-output.test.ts +483 -0
- package/src/services/structured-output.ts +712 -0
- package/src/services/transcription-priority.test.ts +211 -0
- package/src/services/tts/errors.ts +46 -0
- package/src/services/tts/index.ts +214 -0
- package/src/services/tts/tts-audio-cache.ts +235 -0
- package/src/services/tts/types.ts +157 -0
- package/src/services/types.d.ts +19 -0
- package/src/services/types.d.ts.map +1 -0
- package/src/services/types.ts +55 -0
- package/src/services/verify-on-device.d.ts +34 -0
- package/src/services/verify-on-device.d.ts.map +1 -0
- package/src/services/verify-on-device.test.ts +87 -0
- package/src/services/verify-on-device.ts +127 -0
- package/src/services/verify.d.ts +8 -0
- package/src/services/verify.d.ts.map +1 -0
- package/src/services/verify.ts +13 -0
- package/src/services/vision/aosp-unavailable.d.ts +115 -0
- package/src/services/vision/aosp-unavailable.d.ts.map +1 -0
- package/src/services/vision/aosp-unavailable.ts +163 -0
- package/src/services/vision/capacitor-llama.d.ts +99 -0
- package/src/services/vision/capacitor-llama.d.ts.map +1 -0
- package/src/services/vision/capacitor-llama.ts +255 -0
- package/src/services/vision/cloud-fallback.d.ts +47 -0
- package/src/services/vision/cloud-fallback.d.ts.map +1 -0
- package/src/services/vision/cloud-fallback.test.ts +243 -0
- package/src/services/vision/cloud-fallback.ts +268 -0
- package/src/services/vision/fallback-chain.test.ts +86 -0
- package/src/services/vision/hash.d.ts +71 -0
- package/src/services/vision/hash.d.ts.map +1 -0
- package/src/services/vision/hash.ts +157 -0
- package/src/services/vision/index.d.ts +95 -0
- package/src/services/vision/index.d.ts.map +1 -0
- package/src/services/vision/index.ts +251 -0
- package/src/services/vision/llama-server.d.ts +73 -0
- package/src/services/vision/llama-server.d.ts.map +1 -0
- package/src/services/vision/llama-server.ts +177 -0
- package/src/services/vision/types.d.ts +153 -0
- package/src/services/vision/types.d.ts.map +1 -0
- package/src/services/vision/types.ts +154 -0
- package/src/services/vision/vast-fallback.d.ts +18 -0
- package/src/services/vision/vast-fallback.d.ts.map +1 -0
- package/src/services/vision/vast-fallback.ts +127 -0
- package/src/services/vision-embedding-cache.d.ts +98 -0
- package/src/services/vision-embedding-cache.d.ts.map +1 -0
- package/src/services/vision-embedding-cache.ts +189 -0
- package/src/services/voice/VOICE_WORKBENCH.md +88 -0
- package/src/services/voice/__test-helpers__/fake-ffi.ts +92 -0
- package/src/services/voice/__test-helpers__/synthetic-speech.ts +124 -0
- package/src/services/voice/__tests__/checkpoint-manager.test.ts +241 -0
- package/src/services/voice/__tests__/checkpoint-policy.test.ts +270 -0
- package/src/services/voice/__tests__/eager-context-builder.test.ts +257 -0
- package/src/services/voice/__tests__/eliza1-eot-scorer.test.ts +288 -0
- package/src/services/voice/__tests__/eot-classifier.test.ts +431 -0
- package/src/services/voice/__tests__/optimistic-rollback.test.ts +312 -0
- package/src/services/voice/__tests__/prefill-client.test.ts +266 -0
- package/src/services/voice/__tests__/prefix-preserving-queue.test.ts +208 -0
- package/src/services/voice/__tests__/streaming-asr.test.ts +450 -0
- package/src/services/voice/__tests__/streaming-transcriber.test.ts +339 -0
- package/src/services/voice/__tests__/turn-detector-resolver.test.ts +197 -0
- package/src/services/voice/__tests__/voice-state-machine-prefill.test.ts +275 -0
- package/src/services/voice/__tests__/voice-state-machine.test.ts +354 -0
- package/src/services/voice/audio-frame-consumer.d.ts +212 -0
- package/src/services/voice/audio-frame-consumer.d.ts.map +1 -0
- package/src/services/voice/audio-frame-consumer.test.ts +343 -0
- package/src/services/voice/audio-frame-consumer.ts +491 -0
- package/src/services/voice/barge-in.d.ts +112 -0
- package/src/services/voice/barge-in.d.ts.map +1 -0
- package/src/services/voice/barge-in.test.ts +244 -0
- package/src/services/voice/barge-in.ts +336 -0
- package/src/services/voice/cancellation-coordinator.d.ts +127 -0
- package/src/services/voice/cancellation-coordinator.d.ts.map +1 -0
- package/src/services/voice/cancellation-coordinator.test.ts +196 -0
- package/src/services/voice/cancellation-coordinator.ts +269 -0
- package/src/services/voice/checkpoint-manager.d.ts +199 -0
- package/src/services/voice/checkpoint-manager.d.ts.map +1 -0
- package/src/services/voice/checkpoint-manager.ts +401 -0
- package/src/services/voice/checkpoint-policy.ts +336 -0
- package/src/services/voice/composite-eot-classifier.test.ts +59 -0
- package/src/services/voice/e2e-harness.test.ts +182 -0
- package/src/services/voice/e2e-harness.ts +743 -0
- package/src/services/voice/eager-context-builder.d.ts +170 -0
- package/src/services/voice/eager-context-builder.d.ts.map +1 -0
- package/src/services/voice/eager-context-builder.ts +262 -0
- package/src/services/voice/eliza1-eot-scorer.d.ts +124 -0
- package/src/services/voice/eliza1-eot-scorer.d.ts.map +1 -0
- package/src/services/voice/eliza1-eot-scorer.ts +242 -0
- package/src/services/voice/embedding-server.ts +200 -0
- package/src/services/voice/embedding.d.ts +133 -0
- package/src/services/voice/embedding.d.ts.map +1 -0
- package/src/services/voice/embedding.test.ts +148 -0
- package/src/services/voice/embedding.ts +244 -0
- package/src/services/voice/emotion-attribution.d.ts +68 -0
- package/src/services/voice/emotion-attribution.d.ts.map +1 -0
- package/src/services/voice/emotion-attribution.test.ts +129 -0
- package/src/services/voice/emotion-attribution.ts +361 -0
- package/src/services/voice/engine-bridge-cancellation.test.ts +422 -0
- package/src/services/voice/engine-bridge.d.ts +746 -0
- package/src/services/voice/engine-bridge.d.ts.map +1 -0
- package/src/services/voice/engine-bridge.test.ts +384 -0
- package/src/services/voice/engine-bridge.ts +2226 -0
- package/src/services/voice/eot-classifier-ggml.d.ts +179 -0
- package/src/services/voice/eot-classifier-ggml.d.ts.map +1 -0
- package/src/services/voice/eot-classifier-ggml.ts +566 -0
- package/src/services/voice/eot-classifier.d.ts +214 -0
- package/src/services/voice/eot-classifier.d.ts.map +1 -0
- package/src/services/voice/eot-classifier.ts +533 -0
- package/src/services/voice/errors.d.ts +20 -0
- package/src/services/voice/errors.d.ts.map +1 -0
- package/src/services/voice/errors.ts +32 -0
- package/src/services/voice/expressive-tags.d.ts +158 -0
- package/src/services/voice/expressive-tags.d.ts.map +1 -0
- package/src/services/voice/expressive-tags.ts +405 -0
- package/src/services/voice/ffi-bindings.d.ts +636 -0
- package/src/services/voice/ffi-bindings.d.ts.map +1 -0
- package/src/services/voice/ffi-bindings.test.ts +671 -0
- package/src/services/voice/ffi-bindings.ts +3050 -0
- package/src/services/voice/first-line-cache.d.ts +181 -0
- package/src/services/voice/first-line-cache.d.ts.map +1 -0
- package/src/services/voice/first-line-cache.ts +725 -0
- package/src/services/voice/fused-eot-scorer.d.ts +51 -0
- package/src/services/voice/fused-eot-scorer.d.ts.map +1 -0
- package/src/services/voice/fused-eot-scorer.ts +135 -0
- package/src/services/voice/index.d.ts +91 -0
- package/src/services/voice/index.d.ts.map +1 -0
- package/src/services/voice/index.ts +481 -0
- package/src/services/voice/kokoro/__tests__/kokoro-backend.test.ts +151 -0
- package/src/services/voice/kokoro/__tests__/kokoro-engine-bridge.real.test.ts +151 -0
- package/src/services/voice/kokoro/__tests__/kokoro-engine-bridge.test.ts +60 -0
- package/src/services/voice/kokoro/__tests__/kokoro-engine-discovery.test.ts +277 -0
- package/src/services/voice/kokoro/__tests__/kokoro-ffi-runtime.test.ts +235 -0
- package/src/services/voice/kokoro/__tests__/kokoro-runtime.test.ts +95 -0
- package/src/services/voice/kokoro/__tests__/phonemizer.test.ts +53 -0
- package/src/services/voice/kokoro/__tests__/runtime-selection.test.ts +231 -0
- package/src/services/voice/kokoro/__tests__/voices.test.ts +57 -0
- package/src/services/voice/kokoro/index.ts +79 -0
- package/src/services/voice/kokoro/kokoro-backend.d.ts +72 -0
- package/src/services/voice/kokoro/kokoro-backend.d.ts.map +1 -0
- package/src/services/voice/kokoro/kokoro-backend.ts +207 -0
- package/src/services/voice/kokoro/kokoro-engine-discovery.d.ts +58 -0
- package/src/services/voice/kokoro/kokoro-engine-discovery.d.ts.map +1 -0
- package/src/services/voice/kokoro/kokoro-engine-discovery.ts +177 -0
- package/src/services/voice/kokoro/kokoro-ffi-runtime.d.ts +75 -0
- package/src/services/voice/kokoro/kokoro-ffi-runtime.d.ts.map +1 -0
- package/src/services/voice/kokoro/kokoro-ffi-runtime.ts +233 -0
- package/src/services/voice/kokoro/kokoro-runtime.d.ts +100 -0
- package/src/services/voice/kokoro/kokoro-runtime.d.ts.map +1 -0
- package/src/services/voice/kokoro/kokoro-runtime.ts +170 -0
- package/src/services/voice/kokoro/phoneme-stream.ts +123 -0
- package/src/services/voice/kokoro/phonemizer.d.ts +50 -0
- package/src/services/voice/kokoro/phonemizer.d.ts.map +1 -0
- package/src/services/voice/kokoro/phonemizer.ts +344 -0
- package/src/services/voice/kokoro/pick-runtime.d.ts +61 -0
- package/src/services/voice/kokoro/pick-runtime.d.ts.map +1 -0
- package/src/services/voice/kokoro/pick-runtime.test.ts +91 -0
- package/src/services/voice/kokoro/pick-runtime.ts +130 -0
- package/src/services/voice/kokoro/runtime-selection.d.ts +92 -0
- package/src/services/voice/kokoro/runtime-selection.d.ts.map +1 -0
- package/src/services/voice/kokoro/runtime-selection.ts +237 -0
- package/src/services/voice/kokoro/types.d.ts +82 -0
- package/src/services/voice/kokoro/types.d.ts.map +1 -0
- package/src/services/voice/kokoro/types.ts +95 -0
- package/src/services/voice/kokoro/voice-presets.d.ts +23 -0
- package/src/services/voice/kokoro/voice-presets.d.ts.map +1 -0
- package/src/services/voice/kokoro/voice-presets.ts +129 -0
- package/src/services/voice/kokoro/voices.d.ts +30 -0
- package/src/services/voice/kokoro/voices.d.ts.map +1 -0
- package/src/services/voice/kokoro/voices.ts +64 -0
- package/src/services/voice/lifecycle.d.ts +135 -0
- package/src/services/voice/lifecycle.d.ts.map +1 -0
- package/src/services/voice/lifecycle.test.ts +315 -0
- package/src/services/voice/lifecycle.ts +301 -0
- package/src/services/voice/live-diarization-session.d.ts +96 -0
- package/src/services/voice/live-diarization-session.d.ts.map +1 -0
- package/src/services/voice/live-diarization-session.ts +289 -0
- package/src/services/voice/mic-source.d.ts +136 -0
- package/src/services/voice/mic-source.d.ts.map +1 -0
- package/src/services/voice/mic-source.test.ts +210 -0
- package/src/services/voice/mic-source.ts +503 -0
- package/src/services/voice/optimistic-policy.d.ts +109 -0
- package/src/services/voice/optimistic-policy.d.ts.map +1 -0
- package/src/services/voice/optimistic-policy.test.ts +101 -0
- package/src/services/voice/optimistic-policy.ts +192 -0
- package/src/services/voice/optimistic-rollback.ts +343 -0
- package/src/services/voice/partial-stabilizer.d.ts +73 -0
- package/src/services/voice/partial-stabilizer.d.ts.map +1 -0
- package/src/services/voice/partial-stabilizer.test.ts +68 -0
- package/src/services/voice/partial-stabilizer.ts +140 -0
- package/src/services/voice/phoneme-tokenizer.d.ts +49 -0
- package/src/services/voice/phoneme-tokenizer.d.ts.map +1 -0
- package/src/services/voice/phoneme-tokenizer.ts +158 -0
- package/src/services/voice/phrase-cache.d.ts +76 -0
- package/src/services/voice/phrase-cache.d.ts.map +1 -0
- package/src/services/voice/phrase-cache.test.ts +242 -0
- package/src/services/voice/phrase-cache.ts +186 -0
- package/src/services/voice/phrase-chunker.d.ts +62 -0
- package/src/services/voice/phrase-chunker.d.ts.map +1 -0
- package/src/services/voice/phrase-chunker.test.ts +239 -0
- package/src/services/voice/phrase-chunker.ts +281 -0
- package/src/services/voice/pipeline-impls.d.ts +151 -0
- package/src/services/voice/pipeline-impls.d.ts.map +1 -0
- package/src/services/voice/pipeline-impls.l6.test.ts +110 -0
- package/src/services/voice/pipeline-impls.test.ts +292 -0
- package/src/services/voice/pipeline-impls.ts +315 -0
- package/src/services/voice/pipeline.d.ts +216 -0
- package/src/services/voice/pipeline.d.ts.map +1 -0
- package/src/services/voice/pipeline.ts +505 -0
- package/src/services/voice/prefill-client.d.ts +123 -0
- package/src/services/voice/prefill-client.d.ts.map +1 -0
- package/src/services/voice/prefill-client.ts +316 -0
- package/src/services/voice/prefix-preserving-queue.d.ts +113 -0
- package/src/services/voice/prefix-preserving-queue.d.ts.map +1 -0
- package/src/services/voice/prefix-preserving-queue.ts +162 -0
- package/src/services/voice/profile-store.d.ts +248 -0
- package/src/services/voice/profile-store.d.ts.map +1 -0
- package/src/services/voice/profile-store.ts +887 -0
- package/src/services/voice/ring-buffer.d.ts +40 -0
- package/src/services/voice/ring-buffer.d.ts.map +1 -0
- package/src/services/voice/ring-buffer.ts +105 -0
- package/src/services/voice/rollback-queue.d.ts +24 -0
- package/src/services/voice/rollback-queue.d.ts.map +1 -0
- package/src/services/voice/rollback-queue.ts +74 -0
- package/src/services/voice/samantha-preset-placeholder.d.ts +67 -0
- package/src/services/voice/samantha-preset-placeholder.d.ts.map +1 -0
- package/src/services/voice/samantha-preset-placeholder.test.ts +97 -0
- package/src/services/voice/samantha-preset-placeholder.ts +148 -0
- package/src/services/voice/samantha-preset-regenerator.d.ts +87 -0
- package/src/services/voice/samantha-preset-regenerator.d.ts.map +1 -0
- package/src/services/voice/samantha-preset-regenerator.ts +393 -0
- package/src/services/voice/scheduler.d.ts +146 -0
- package/src/services/voice/scheduler.d.ts.map +1 -0
- package/src/services/voice/scheduler.t2.test.ts +141 -0
- package/src/services/voice/scheduler.ts +927 -0
- package/src/services/voice/shared-resources.d.ts +190 -0
- package/src/services/voice/shared-resources.d.ts.map +1 -0
- package/src/services/voice/shared-resources.ts +320 -0
- package/src/services/voice/speaker/attribution-pipeline.d.ts +74 -0
- package/src/services/voice/speaker/attribution-pipeline.d.ts.map +1 -0
- package/src/services/voice/speaker/attribution-pipeline.ts +386 -0
- package/src/services/voice/speaker/diarizer-fused.d.ts +59 -0
- package/src/services/voice/speaker/diarizer-fused.d.ts.map +1 -0
- package/src/services/voice/speaker/diarizer-fused.real.test.ts +100 -0
- package/src/services/voice/speaker/diarizer-fused.ts +154 -0
- package/src/services/voice/speaker/diarizer.d.ts +75 -0
- package/src/services/voice/speaker/diarizer.d.ts.map +1 -0
- package/src/services/voice/speaker/diarizer.ts +218 -0
- package/src/services/voice/speaker/encoder-fused.d.ts +60 -0
- package/src/services/voice/speaker/encoder-fused.d.ts.map +1 -0
- package/src/services/voice/speaker/encoder-fused.real.test.ts +113 -0
- package/src/services/voice/speaker/encoder-fused.ts +138 -0
- package/src/services/voice/speaker/encoder-ggml.d.ts +33 -0
- package/src/services/voice/speaker/encoder-ggml.d.ts.map +1 -0
- package/src/services/voice/speaker/encoder-ggml.ts +79 -0
- package/src/services/voice/speaker/encoder.d.ts +37 -0
- package/src/services/voice/speaker/encoder.d.ts.map +1 -0
- package/src/services/voice/speaker/encoder.ts +105 -0
- package/src/services/voice/speaker-imprint.d.ts +83 -0
- package/src/services/voice/speaker-imprint.d.ts.map +1 -0
- package/src/services/voice/speaker-imprint.test.ts +185 -0
- package/src/services/voice/speaker-imprint.ts +312 -0
- package/src/services/voice/speaker-preset-cache.d.ts +77 -0
- package/src/services/voice/speaker-preset-cache.d.ts.map +1 -0
- package/src/services/voice/speaker-preset-cache.test.ts +154 -0
- package/src/services/voice/speaker-preset-cache.ts +195 -0
- package/src/services/voice/streaming-asr/streaming-pipeline-adapter.ts +292 -0
- package/src/services/voice/system-audio-sink.d.ts +73 -0
- package/src/services/voice/system-audio-sink.d.ts.map +1 -0
- package/src/services/voice/system-audio-sink.test.ts +29 -0
- package/src/services/voice/system-audio-sink.ts +366 -0
- package/src/services/voice/transcriber.d.ts +244 -0
- package/src/services/voice/transcriber.d.ts.map +1 -0
- package/src/services/voice/transcriber.test.ts +392 -0
- package/src/services/voice/transcriber.ts +704 -0
- package/src/services/voice/turn-controller.d.ts +183 -0
- package/src/services/voice/turn-controller.d.ts.map +1 -0
- package/src/services/voice/turn-controller.test.ts +575 -0
- package/src/services/voice/turn-controller.ts +596 -0
- package/src/services/voice/types.d.ts +643 -0
- package/src/services/voice/types.d.ts.map +1 -0
- package/src/services/voice/types.ts +699 -0
- package/src/services/voice/vad.d.ts +282 -0
- package/src/services/voice/vad.d.ts.map +1 -0
- package/src/services/voice/vad.test.ts +480 -0
- package/src/services/voice/vad.ts +827 -0
- package/src/services/voice/vad.v1-v4.test.ts +222 -0
- package/src/services/voice/voice-budget.d.ts +241 -0
- package/src/services/voice/voice-budget.d.ts.map +1 -0
- package/src/services/voice/voice-budget.test.ts +420 -0
- package/src/services/voice/voice-budget.ts +656 -0
- package/src/services/voice/voice-duet.test.ts +375 -0
- package/src/services/voice/voice-emotion-classifier.d.ts +95 -0
- package/src/services/voice/voice-emotion-classifier.d.ts.map +1 -0
- package/src/services/voice/voice-emotion-classifier.test.ts +210 -0
- package/src/services/voice/voice-emotion-classifier.ts +273 -0
- package/src/services/voice/voice-preset-format.d.ts +158 -0
- package/src/services/voice/voice-preset-format.d.ts.map +1 -0
- package/src/services/voice/voice-preset-format.ts +700 -0
- package/src/services/voice/voice-preset-generator.test.ts +89 -0
- package/src/services/voice/voice-profile-artifact.d.ts +116 -0
- package/src/services/voice/voice-profile-artifact.d.ts.map +1 -0
- package/src/services/voice/voice-profile-artifact.test.ts +138 -0
- package/src/services/voice/voice-profile-artifact.ts +518 -0
- package/src/services/voice/voice-profile-routes.d.ts +83 -0
- package/src/services/voice/voice-profile-routes.d.ts.map +1 -0
- package/src/services/voice/voice-profile-routes.test.ts +429 -0
- package/src/services/voice/voice-profile-routes.ts +425 -0
- package/src/services/voice/voice-scenario.ts +154 -0
- package/src/services/voice/voice-settings.d.ts +82 -0
- package/src/services/voice/voice-settings.d.ts.map +1 -0
- package/src/services/voice/voice-settings.ts +172 -0
- package/src/services/voice/voice-state-machine.d.ts +364 -0
- package/src/services/voice/voice-state-machine.d.ts.map +1 -0
- package/src/services/voice/voice-state-machine.ts +727 -0
- package/src/services/voice/voice-workbench-report.test.ts +168 -0
- package/src/services/voice/voice-workbench-report.ts +326 -0
- package/src/services/voice/voice-workbench.test.ts +158 -0
- package/src/services/voice/voice.test.ts +1070 -0
- package/src/services/voice/wake-word-ggml.d.ts +101 -0
- package/src/services/voice/wake-word-ggml.d.ts.map +1 -0
- package/src/services/voice/wake-word-ggml.ts +320 -0
- package/src/services/voice/wake-word.d.ts +255 -0
- package/src/services/voice/wake-word.d.ts.map +1 -0
- package/src/services/voice/wake-word.test.ts +298 -0
- package/src/services/voice/wake-word.ts +554 -0
- package/src/services/voice/wrap-with-first-line-cache.d.ts +70 -0
- package/src/services/voice/wrap-with-first-line-cache.d.ts.map +1 -0
- package/src/services/voice/wrap-with-first-line-cache.ts +267 -0
- package/src/services/voice-model-updater.d.ts +240 -0
- package/src/services/voice-model-updater.d.ts.map +1 -0
- package/src/services/voice-model-updater.ts +724 -0
- package/src/services/voice-prewarm.d.ts +3 -0
- package/src/services/voice-prewarm.d.ts.map +1 -0
- package/src/services/voice-prewarm.ts +51 -0
- package/dist/index.d.ts +0 -37
- package/dist/index.js +0 -1098
|
@@ -0,0 +1,149 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* DeviceResourceMetrics — on-device resource profiling accumulator.
|
|
3
|
+
*
|
|
4
|
+
* Sibling to `VoiceRunMetrics` (`latency-trace.ts`), generalised for the Mobile
|
|
5
|
+
* Resource Workbench (issue #8800). Where `VoiceRunMetrics` accumulates the
|
|
6
|
+
* non-latency signals of a *host-side* voice run (server RSS, MTP accept-rate),
|
|
7
|
+
* this accumulates the signals of an *on-device* agent run:
|
|
8
|
+
*
|
|
9
|
+
* - per-generation **prefill / decode / combined tokens-per-second** + TTFT
|
|
10
|
+
* (fed from the device-bridge `computeGenerationThroughput` differencing),
|
|
11
|
+
* - **peak + steady resident memory (RSS)** with an RSS-leak flag,
|
|
12
|
+
* - **battery drain** (% and, where the OS exposes a charge counter, µAh),
|
|
13
|
+
* - a **thermal-state timeline** + transition count,
|
|
14
|
+
* - **low-power-mode transitions**.
|
|
15
|
+
*
|
|
16
|
+
* Every field is optional on input and every output is `null` when nothing
|
|
17
|
+
* could be measured — a quantity the device could not report is recorded as
|
|
18
|
+
* missing, never as a fabricated `0` (AGENTS.md §3 / §7). The workbench writes
|
|
19
|
+
* `summary()` to `results/<workload>/latest.json`.
|
|
20
|
+
*/
|
|
21
|
+
import { type HistogramSummary } from "./latency-trace";
|
|
22
|
+
export type DeviceThermalState = "nominal" | "fair" | "serious" | "critical" | "unknown";
|
|
23
|
+
/** A single generation's differenced throughput (see `throughput.ts`). */
|
|
24
|
+
export interface GenerationObservation {
|
|
25
|
+
prefillTokensPerSecond?: number | null;
|
|
26
|
+
decodeTokensPerSecond?: number | null;
|
|
27
|
+
combinedTokensPerSecond?: number | null;
|
|
28
|
+
ttftMs?: number | null;
|
|
29
|
+
}
|
|
30
|
+
/**
|
|
31
|
+
* A sampled device-resource snapshot, taken on an interval across a workload.
|
|
32
|
+
* Sourced from the native `getResourceSnapshot` bridge (iOS/Android) and/or
|
|
33
|
+
* host-side OS probes (`adb dumpsys`, `/proc`).
|
|
34
|
+
*/
|
|
35
|
+
export interface ResourceSample {
|
|
36
|
+
/** Timestamp in ms (epoch or run-relative); orders the timeline + drain rate. */
|
|
37
|
+
atMs: number;
|
|
38
|
+
/** Resident set size in MB (peak / steady / leak tracking). */
|
|
39
|
+
residentMemoryMb?: number | null;
|
|
40
|
+
/** Available device RAM in MB at sample time. */
|
|
41
|
+
availableRamMb?: number | null;
|
|
42
|
+
/** Battery level 0..100. */
|
|
43
|
+
batteryLevelPct?: number | null;
|
|
44
|
+
/** Cumulative charge counter in µAh (Android `BATTERY_PROPERTY_CHARGE_COUNTER`). */
|
|
45
|
+
batteryChargeMicroAmpHours?: number | null;
|
|
46
|
+
/** Whether the device was charging at sample time. */
|
|
47
|
+
isCharging?: boolean | null;
|
|
48
|
+
thermalState?: DeviceThermalState | null;
|
|
49
|
+
lowPowerMode?: boolean | null;
|
|
50
|
+
/** Cumulative process CPU time in ms, when the platform exposes it. */
|
|
51
|
+
cpuTimeMs?: number | null;
|
|
52
|
+
}
|
|
53
|
+
export interface RssSummary {
|
|
54
|
+
firstMb: number | null;
|
|
55
|
+
lastMb: number | null;
|
|
56
|
+
peakMb: number | null;
|
|
57
|
+
/** Median RSS over the back half of the run (warm/steady state). */
|
|
58
|
+
steadyMb: number | null;
|
|
59
|
+
samples: number;
|
|
60
|
+
/** lastMb − firstMb, or null. */
|
|
61
|
+
growthMb: number | null;
|
|
62
|
+
/** True when RSS is monotone non-decreasing across ≥4 samples and grew past the threshold. */
|
|
63
|
+
leakSuspected: boolean;
|
|
64
|
+
}
|
|
65
|
+
export interface BatterySummary {
|
|
66
|
+
firstPct: number | null;
|
|
67
|
+
lastPct: number | null;
|
|
68
|
+
/** firstPct − lastPct (positive = battery consumed), or null. */
|
|
69
|
+
drainPct: number | null;
|
|
70
|
+
samples: number;
|
|
71
|
+
/** True if any sample reported charging (drain numbers are then unreliable). */
|
|
72
|
+
chargingObserved: boolean;
|
|
73
|
+
/** firstCharge − lastCharge in µAh (positive = consumed), when the counter is present. */
|
|
74
|
+
energyMicroAmpHoursDelta: number | null;
|
|
75
|
+
/** Wall-clock span across battery samples in ms, or null. */
|
|
76
|
+
durationMs: number | null;
|
|
77
|
+
}
|
|
78
|
+
export interface ThermalTransition {
|
|
79
|
+
atMs: number;
|
|
80
|
+
state: DeviceThermalState;
|
|
81
|
+
}
|
|
82
|
+
export interface ThermalSummary {
|
|
83
|
+
samples: number;
|
|
84
|
+
initialState: DeviceThermalState | null;
|
|
85
|
+
/** Worst (highest-severity) known state observed. */
|
|
86
|
+
maxState: DeviceThermalState | null;
|
|
87
|
+
/** State-change timeline (first known state + every change after). */
|
|
88
|
+
transitions: ThermalTransition[];
|
|
89
|
+
transitionCount: number;
|
|
90
|
+
/** Fraction of known-state samples at serious|critical, or null when no known state. */
|
|
91
|
+
fractionThrottled: number | null;
|
|
92
|
+
}
|
|
93
|
+
export interface LowPowerSummary {
|
|
94
|
+
samples: number;
|
|
95
|
+
everEnabled: boolean;
|
|
96
|
+
transitionCount: number;
|
|
97
|
+
}
|
|
98
|
+
export interface DeviceResourceSummary {
|
|
99
|
+
generations: number;
|
|
100
|
+
resourceSamples: number;
|
|
101
|
+
prefillTokensPerSecond: HistogramSummary;
|
|
102
|
+
decodeTokensPerSecond: HistogramSummary;
|
|
103
|
+
combinedTokensPerSecond: HistogramSummary;
|
|
104
|
+
ttftMs: HistogramSummary;
|
|
105
|
+
rss: RssSummary;
|
|
106
|
+
battery: BatterySummary;
|
|
107
|
+
thermal: ThermalSummary;
|
|
108
|
+
lowPowerMode: LowPowerSummary;
|
|
109
|
+
}
|
|
110
|
+
export declare class DeviceResourceMetrics {
|
|
111
|
+
private readonly opts;
|
|
112
|
+
private generations;
|
|
113
|
+
private readonly prefillHist;
|
|
114
|
+
private readonly decodeHist;
|
|
115
|
+
private readonly combinedHist;
|
|
116
|
+
private readonly ttftHist;
|
|
117
|
+
private readonly rssSamples;
|
|
118
|
+
private resourceSamples;
|
|
119
|
+
private firstBattery;
|
|
120
|
+
private lastBattery;
|
|
121
|
+
private firstChargeUah;
|
|
122
|
+
private lastChargeUah;
|
|
123
|
+
private batterySampleCount;
|
|
124
|
+
private chargingObserved;
|
|
125
|
+
private thermalSampleCount;
|
|
126
|
+
private thermalInitial;
|
|
127
|
+
private thermalLast;
|
|
128
|
+
private thermalMaxRank;
|
|
129
|
+
private thermalMaxState;
|
|
130
|
+
private thermalThrottledCount;
|
|
131
|
+
private thermalKnownCount;
|
|
132
|
+
private readonly thermalTransitions;
|
|
133
|
+
private lowPowerSampleCount;
|
|
134
|
+
private lowPowerEver;
|
|
135
|
+
private lowPowerLast;
|
|
136
|
+
private lowPowerTransitions;
|
|
137
|
+
constructor(opts?: {
|
|
138
|
+
leakGrowthMbThreshold?: number;
|
|
139
|
+
});
|
|
140
|
+
/** Record one generation's differenced throughput. */
|
|
141
|
+
recordGeneration(obs: GenerationObservation): void;
|
|
142
|
+
/** Record one sampled device-resource snapshot. */
|
|
143
|
+
recordResourceSample(sample: ResourceSample): void;
|
|
144
|
+
private rssSummary;
|
|
145
|
+
private batterySummary;
|
|
146
|
+
private thermalSummary;
|
|
147
|
+
summary(): DeviceResourceSummary;
|
|
148
|
+
}
|
|
149
|
+
//# sourceMappingURL=device-resource-metrics.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"device-resource-metrics.d.ts","sourceRoot":"","sources":["device-resource-metrics.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AAEH,OAAO,EAAoB,KAAK,gBAAgB,EAAE,MAAM,iBAAiB,CAAC;AAE1E,MAAM,MAAM,kBAAkB,GAC3B,SAAS,GACT,MAAM,GACN,SAAS,GACT,UAAU,GACV,SAAS,CAAC;AAUb,0EAA0E;AAC1E,MAAM,WAAW,qBAAqB;IACrC,sBAAsB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACvC,qBAAqB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACtC,uBAAuB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACxC,MAAM,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CACvB;AAED;;;;GAIG;AACH,MAAM,WAAW,cAAc;IAC9B,iFAAiF;IACjF,IAAI,EAAE,MAAM,CAAC;IACb,+DAA+D;IAC/D,gBAAgB,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IACjC,iDAAiD;IACjD,cAAc,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC/B,4BAA4B;IAC5B,eAAe,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAChC,oFAAoF;IACpF,0BAA0B,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3C,sDAAsD;IACtD,UAAU,CAAC,EAAE,OAAO,GAAG,IAAI,CAAC;IAC5B,YAAY,CAAC,EAAE,kBAAkB,GAAG,IAAI,CAAC;IACzC,YAAY,CAAC,EAAE,OAAO,GAAG,IAAI,CAAC;IAC9B,uEAAuE;IACvE,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CAC1B;AAED,MAAM,WAAW,UAAU;IAC1B,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,oEAAoE;IACpE,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,OAAO,EAAE,MAAM,CAAC;IAChB,iCAAiC;IACjC,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,8FAA8F;IAC9F,aAAa,EAAE,OAAO,CAAC;CACvB;AAED,MAAM,WAAW,cAAc;IAC9B,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB,iEAAiE;IACjE,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,OAAO,EAAE,MAAM,CAAC;IAChB,gFAAgF;IAChF,gBAAgB,EAAE,OAAO,CAAC;IAC1B,0FAA0F;IAC1F,wBAAwB,EAAE,MAAM,GAAG,IAAI,CAAC;IACxC,6DAA6D;IAC7D,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;CAC1B;AAED,MAAM,WAAW,iBAAiB;IACjC,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,kBAAkB,CAAC;CAC1B;AAED,MAAM,WAAW,cAAc;IAC9B,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,kBAAkB,GAAG,IAAI,CAAC;IACxC,qDAAqD;IACrD,QAAQ,EAAE,kBAAkB,GAAG,IAAI,CAAC;IACpC,sEAAsE;IACtE,WAAW,EAAE,iBAAiB,EAAE,CAAC;IACjC,eAAe,EAAE,MAAM,CAAC;IACxB,wFAAwF;IACxF,iBAAiB,EAAE,MAAM,GAAG,IAAI,CAAC;CACjC;AAED,MAAM,WAAW,eAAe;IAC/B,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,EAAE,OAAO,CAAC;IACrB,eAAe,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,qBAAqB;IACrC,WAAW,EAAE,MAAM,CAAC;IACpB,eAAe,EAAE,MAAM,CAAC;IACxB,sBAAsB,EAAE,gBAAgB,CAAC;IACzC,qBAAqB,EAAE,gBAAgB,CAAC;IACxC,uBAAuB,EAAE,gBAAgB,CAAC;IAC1C,MAAM,EAAE,gBAAgB,CAAC;IACzB,GAAG,EAAE,UAAU,CAAC;IAChB,OAAO,EAAE,cAAc,CAAC;IACxB,OAAO,EAAE,cAAc,CAAC;IACxB,YAAY,EAAE,eAAe,CAAC;CAC9B;AAoBD,qBAAa,qBAAqB;IA+BrB,OAAO,CAAC,QAAQ,CAAC,IAAI;IA9BjC,OAAO,CAAC,WAAW,CAAK;IACxB,OAAO,CAAC,QAAQ,CAAC,WAAW,CAA4C;IACxE,OAAO,CAAC,QAAQ,CAAC,UAAU,CAA4C;IACvE,OAAO,CAAC,QAAQ,CAAC,YAAY,CAA4C;IACzE,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAA4C;IAErE,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAgB;IAE3C,OAAO,CAAC,eAAe,CAAK;IAC5B,OAAO,CAAC,YAAY,CAA8C;IAClE,OAAO,CAAC,WAAW,CAA8C;IACjE,OAAO,CAAC,cAAc,CAAuB;IAC7C,OAAO,CAAC,aAAa,CAAuB;IAC5C,OAAO,CAAC,kBAAkB,CAAK;IAC/B,OAAO,CAAC,gBAAgB,CAAS;IAEjC,OAAO,CAAC,kBAAkB,CAAK;IAC/B,OAAO,CAAC,cAAc,CAAmC;IACzD,OAAO,CAAC,WAAW,CAAmC;IACtD,OAAO,CAAC,cAAc,CAAM;IAC5B,OAAO,CAAC,eAAe,CAAmC;IAC1D,OAAO,CAAC,qBAAqB,CAAK;IAClC,OAAO,CAAC,iBAAiB,CAAK;IAC9B,OAAO,CAAC,QAAQ,CAAC,kBAAkB,CAA2B;IAE9D,OAAO,CAAC,mBAAmB,CAAK;IAChC,OAAO,CAAC,YAAY,CAAS;IAC7B,OAAO,CAAC,YAAY,CAAwB;IAC5C,OAAO,CAAC,mBAAmB,CAAK;gBAEH,IAAI,GAAE;QAAE,qBAAqB,CAAC,EAAE,MAAM,CAAA;KAAO;IAE1E,sDAAsD;IACtD,gBAAgB,CAAC,GAAG,EAAE,qBAAqB,GAAG,IAAI;IAWlD,mDAAmD;IACnD,oBAAoB,CAAC,MAAM,EAAE,cAAc,GAAG,IAAI;IA4DlD,OAAO,CAAC,UAAU;IA+BlB,OAAO,CAAC,cAAc;IAwBtB,OAAO,CAAC,cAAc;IAetB,OAAO,IAAI,qBAAqB;CAkBhC"}
|
|
@@ -0,0 +1,98 @@
|
|
|
1
|
+
import { describe, expect, it } from "vitest";
|
|
2
|
+
import { DeviceResourceMetrics } from "./device-resource-metrics";
|
|
3
|
+
|
|
4
|
+
describe("DeviceResourceMetrics", () => {
|
|
5
|
+
it("summarizes generation throughput and sampled device resources", () => {
|
|
6
|
+
const metrics = new DeviceResourceMetrics({ leakGrowthMbThreshold: 100 });
|
|
7
|
+
|
|
8
|
+
metrics.recordGeneration({
|
|
9
|
+
prefillTokensPerSecond: 320,
|
|
10
|
+
decodeTokensPerSecond: 160,
|
|
11
|
+
combinedTokensPerSecond: 128,
|
|
12
|
+
ttftMs: 200,
|
|
13
|
+
});
|
|
14
|
+
metrics.recordResourceSample({
|
|
15
|
+
atMs: 0,
|
|
16
|
+
residentMemoryMb: 1000,
|
|
17
|
+
batteryLevelPct: 80,
|
|
18
|
+
batteryChargeMicroAmpHours: 10_000,
|
|
19
|
+
isCharging: false,
|
|
20
|
+
thermalState: "nominal",
|
|
21
|
+
lowPowerMode: false,
|
|
22
|
+
});
|
|
23
|
+
metrics.recordResourceSample({
|
|
24
|
+
atMs: 1000,
|
|
25
|
+
residentMemoryMb: 1050,
|
|
26
|
+
batteryLevelPct: 79,
|
|
27
|
+
batteryChargeMicroAmpHours: 9900,
|
|
28
|
+
thermalState: "serious",
|
|
29
|
+
lowPowerMode: true,
|
|
30
|
+
});
|
|
31
|
+
metrics.recordResourceSample({
|
|
32
|
+
atMs: 2000,
|
|
33
|
+
residentMemoryMb: 1125,
|
|
34
|
+
batteryLevelPct: 78,
|
|
35
|
+
batteryChargeMicroAmpHours: 9800,
|
|
36
|
+
thermalState: "critical",
|
|
37
|
+
lowPowerMode: false,
|
|
38
|
+
});
|
|
39
|
+
metrics.recordResourceSample({
|
|
40
|
+
atMs: 3000,
|
|
41
|
+
residentMemoryMb: 1205,
|
|
42
|
+
batteryLevelPct: 77,
|
|
43
|
+
batteryChargeMicroAmpHours: 9700,
|
|
44
|
+
thermalState: "critical",
|
|
45
|
+
lowPowerMode: false,
|
|
46
|
+
});
|
|
47
|
+
|
|
48
|
+
const summary = metrics.summary();
|
|
49
|
+
|
|
50
|
+
expect(summary.generations).toBe(1);
|
|
51
|
+
expect(summary.prefillTokensPerSecond.p50).toBe(320);
|
|
52
|
+
expect(summary.decodeTokensPerSecond.p50).toBe(160);
|
|
53
|
+
expect(summary.combinedTokensPerSecond.p50).toBe(128);
|
|
54
|
+
expect(summary.ttftMs.p50).toBe(200);
|
|
55
|
+
expect(summary.rss).toMatchObject({
|
|
56
|
+
firstMb: 1000,
|
|
57
|
+
lastMb: 1205,
|
|
58
|
+
peakMb: 1205,
|
|
59
|
+
steadyMb: 1165,
|
|
60
|
+
growthMb: 205,
|
|
61
|
+
leakSuspected: true,
|
|
62
|
+
});
|
|
63
|
+
expect(summary.battery).toMatchObject({
|
|
64
|
+
firstPct: 80,
|
|
65
|
+
lastPct: 77,
|
|
66
|
+
drainPct: 3,
|
|
67
|
+
energyMicroAmpHoursDelta: 300,
|
|
68
|
+
durationMs: 3000,
|
|
69
|
+
chargingObserved: false,
|
|
70
|
+
});
|
|
71
|
+
expect(summary.thermal).toMatchObject({
|
|
72
|
+
initialState: "nominal",
|
|
73
|
+
maxState: "critical",
|
|
74
|
+
transitionCount: 2,
|
|
75
|
+
fractionThrottled: 0.75,
|
|
76
|
+
});
|
|
77
|
+
expect(summary.lowPowerMode).toMatchObject({
|
|
78
|
+
everEnabled: true,
|
|
79
|
+
transitionCount: 2,
|
|
80
|
+
});
|
|
81
|
+
});
|
|
82
|
+
|
|
83
|
+
it("keeps unmeasured quantities null instead of fabricating zeros", () => {
|
|
84
|
+
const metrics = new DeviceResourceMetrics();
|
|
85
|
+
|
|
86
|
+
metrics.recordGeneration({});
|
|
87
|
+
metrics.recordResourceSample({ atMs: 0 });
|
|
88
|
+
|
|
89
|
+
const summary = metrics.summary();
|
|
90
|
+
|
|
91
|
+
expect(summary.generations).toBe(1);
|
|
92
|
+
expect(summary.prefillTokensPerSecond.count).toBe(0);
|
|
93
|
+
expect(summary.prefillTokensPerSecond.p50).toBeNull();
|
|
94
|
+
expect(summary.rss.firstMb).toBeNull();
|
|
95
|
+
expect(summary.battery.drainPct).toBeNull();
|
|
96
|
+
expect(summary.thermal.fractionThrottled).toBeNull();
|
|
97
|
+
});
|
|
98
|
+
});
|
|
@@ -0,0 +1,346 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* DeviceResourceMetrics — on-device resource profiling accumulator.
|
|
3
|
+
*
|
|
4
|
+
* Sibling to `VoiceRunMetrics` (`latency-trace.ts`), generalised for the Mobile
|
|
5
|
+
* Resource Workbench (issue #8800). Where `VoiceRunMetrics` accumulates the
|
|
6
|
+
* non-latency signals of a *host-side* voice run (server RSS, MTP accept-rate),
|
|
7
|
+
* this accumulates the signals of an *on-device* agent run:
|
|
8
|
+
*
|
|
9
|
+
* - per-generation **prefill / decode / combined tokens-per-second** + TTFT
|
|
10
|
+
* (fed from the device-bridge `computeGenerationThroughput` differencing),
|
|
11
|
+
* - **peak + steady resident memory (RSS)** with an RSS-leak flag,
|
|
12
|
+
* - **battery drain** (% and, where the OS exposes a charge counter, µAh),
|
|
13
|
+
* - a **thermal-state timeline** + transition count,
|
|
14
|
+
* - **low-power-mode transitions**.
|
|
15
|
+
*
|
|
16
|
+
* Every field is optional on input and every output is `null` when nothing
|
|
17
|
+
* could be measured — a quantity the device could not report is recorded as
|
|
18
|
+
* missing, never as a fabricated `0` (AGENTS.md §3 / §7). The workbench writes
|
|
19
|
+
* `summary()` to `results/<workload>/latest.json`.
|
|
20
|
+
*/
|
|
21
|
+
|
|
22
|
+
import { BoundedHistogram, type HistogramSummary } from "./latency-trace";
|
|
23
|
+
|
|
24
|
+
export type DeviceThermalState =
|
|
25
|
+
| "nominal"
|
|
26
|
+
| "fair"
|
|
27
|
+
| "serious"
|
|
28
|
+
| "critical"
|
|
29
|
+
| "unknown";
|
|
30
|
+
|
|
31
|
+
/** Severity rank for thermal states. `unknown` is unranked (excluded). */
|
|
32
|
+
const THERMAL_RANK: Record<Exclude<DeviceThermalState, "unknown">, number> = {
|
|
33
|
+
nominal: 0,
|
|
34
|
+
fair: 1,
|
|
35
|
+
serious: 2,
|
|
36
|
+
critical: 3,
|
|
37
|
+
};
|
|
38
|
+
|
|
39
|
+
/** A single generation's differenced throughput (see `throughput.ts`). */
|
|
40
|
+
export interface GenerationObservation {
|
|
41
|
+
prefillTokensPerSecond?: number | null;
|
|
42
|
+
decodeTokensPerSecond?: number | null;
|
|
43
|
+
combinedTokensPerSecond?: number | null;
|
|
44
|
+
ttftMs?: number | null;
|
|
45
|
+
}
|
|
46
|
+
|
|
47
|
+
/**
|
|
48
|
+
* A sampled device-resource snapshot, taken on an interval across a workload.
|
|
49
|
+
* Sourced from the native `getResourceSnapshot` bridge (iOS/Android) and/or
|
|
50
|
+
* host-side OS probes (`adb dumpsys`, `/proc`).
|
|
51
|
+
*/
|
|
52
|
+
export interface ResourceSample {
|
|
53
|
+
/** Timestamp in ms (epoch or run-relative); orders the timeline + drain rate. */
|
|
54
|
+
atMs: number;
|
|
55
|
+
/** Resident set size in MB (peak / steady / leak tracking). */
|
|
56
|
+
residentMemoryMb?: number | null;
|
|
57
|
+
/** Available device RAM in MB at sample time. */
|
|
58
|
+
availableRamMb?: number | null;
|
|
59
|
+
/** Battery level 0..100. */
|
|
60
|
+
batteryLevelPct?: number | null;
|
|
61
|
+
/** Cumulative charge counter in µAh (Android `BATTERY_PROPERTY_CHARGE_COUNTER`). */
|
|
62
|
+
batteryChargeMicroAmpHours?: number | null;
|
|
63
|
+
/** Whether the device was charging at sample time. */
|
|
64
|
+
isCharging?: boolean | null;
|
|
65
|
+
thermalState?: DeviceThermalState | null;
|
|
66
|
+
lowPowerMode?: boolean | null;
|
|
67
|
+
/** Cumulative process CPU time in ms, when the platform exposes it. */
|
|
68
|
+
cpuTimeMs?: number | null;
|
|
69
|
+
}
|
|
70
|
+
|
|
71
|
+
export interface RssSummary {
|
|
72
|
+
firstMb: number | null;
|
|
73
|
+
lastMb: number | null;
|
|
74
|
+
peakMb: number | null;
|
|
75
|
+
/** Median RSS over the back half of the run (warm/steady state). */
|
|
76
|
+
steadyMb: number | null;
|
|
77
|
+
samples: number;
|
|
78
|
+
/** lastMb − firstMb, or null. */
|
|
79
|
+
growthMb: number | null;
|
|
80
|
+
/** True when RSS is monotone non-decreasing across ≥4 samples and grew past the threshold. */
|
|
81
|
+
leakSuspected: boolean;
|
|
82
|
+
}
|
|
83
|
+
|
|
84
|
+
export interface BatterySummary {
|
|
85
|
+
firstPct: number | null;
|
|
86
|
+
lastPct: number | null;
|
|
87
|
+
/** firstPct − lastPct (positive = battery consumed), or null. */
|
|
88
|
+
drainPct: number | null;
|
|
89
|
+
samples: number;
|
|
90
|
+
/** True if any sample reported charging (drain numbers are then unreliable). */
|
|
91
|
+
chargingObserved: boolean;
|
|
92
|
+
/** firstCharge − lastCharge in µAh (positive = consumed), when the counter is present. */
|
|
93
|
+
energyMicroAmpHoursDelta: number | null;
|
|
94
|
+
/** Wall-clock span across battery samples in ms, or null. */
|
|
95
|
+
durationMs: number | null;
|
|
96
|
+
}
|
|
97
|
+
|
|
98
|
+
export interface ThermalTransition {
|
|
99
|
+
atMs: number;
|
|
100
|
+
state: DeviceThermalState;
|
|
101
|
+
}
|
|
102
|
+
|
|
103
|
+
export interface ThermalSummary {
|
|
104
|
+
samples: number;
|
|
105
|
+
initialState: DeviceThermalState | null;
|
|
106
|
+
/** Worst (highest-severity) known state observed. */
|
|
107
|
+
maxState: DeviceThermalState | null;
|
|
108
|
+
/** State-change timeline (first known state + every change after). */
|
|
109
|
+
transitions: ThermalTransition[];
|
|
110
|
+
transitionCount: number;
|
|
111
|
+
/** Fraction of known-state samples at serious|critical, or null when no known state. */
|
|
112
|
+
fractionThrottled: number | null;
|
|
113
|
+
}
|
|
114
|
+
|
|
115
|
+
export interface LowPowerSummary {
|
|
116
|
+
samples: number;
|
|
117
|
+
everEnabled: boolean;
|
|
118
|
+
transitionCount: number;
|
|
119
|
+
}
|
|
120
|
+
|
|
121
|
+
export interface DeviceResourceSummary {
|
|
122
|
+
generations: number;
|
|
123
|
+
resourceSamples: number;
|
|
124
|
+
prefillTokensPerSecond: HistogramSummary;
|
|
125
|
+
decodeTokensPerSecond: HistogramSummary;
|
|
126
|
+
combinedTokensPerSecond: HistogramSummary;
|
|
127
|
+
ttftMs: HistogramSummary;
|
|
128
|
+
rss: RssSummary;
|
|
129
|
+
battery: BatterySummary;
|
|
130
|
+
thermal: ThermalSummary;
|
|
131
|
+
lowPowerMode: LowPowerSummary;
|
|
132
|
+
}
|
|
133
|
+
|
|
134
|
+
const HISTOGRAM_CAPACITY = 1024;
|
|
135
|
+
const DEFAULT_LEAK_GROWTH_MB = 256;
|
|
136
|
+
/** Cap the thermal timeline so a long run cannot grow it without bound. */
|
|
137
|
+
const MAX_THERMAL_TRANSITIONS = 256;
|
|
138
|
+
|
|
139
|
+
function isFiniteNumber(v: number | null | undefined): v is number {
|
|
140
|
+
return typeof v === "number" && Number.isFinite(v);
|
|
141
|
+
}
|
|
142
|
+
|
|
143
|
+
function median(values: number[]): number | null {
|
|
144
|
+
if (values.length === 0) return null;
|
|
145
|
+
const s = [...values].sort((a, b) => a - b);
|
|
146
|
+
const mid = Math.floor(s.length / 2);
|
|
147
|
+
return s.length % 2
|
|
148
|
+
? (s[mid] as number)
|
|
149
|
+
: ((s[mid - 1] as number) + (s[mid] as number)) / 2;
|
|
150
|
+
}
|
|
151
|
+
|
|
152
|
+
export class DeviceResourceMetrics {
|
|
153
|
+
private generations = 0;
|
|
154
|
+
private readonly prefillHist = new BoundedHistogram(HISTOGRAM_CAPACITY);
|
|
155
|
+
private readonly decodeHist = new BoundedHistogram(HISTOGRAM_CAPACITY);
|
|
156
|
+
private readonly combinedHist = new BoundedHistogram(HISTOGRAM_CAPACITY);
|
|
157
|
+
private readonly ttftHist = new BoundedHistogram(HISTOGRAM_CAPACITY);
|
|
158
|
+
|
|
159
|
+
private readonly rssSamples: number[] = [];
|
|
160
|
+
|
|
161
|
+
private resourceSamples = 0;
|
|
162
|
+
private firstBattery: { pct: number; atMs: number } | null = null;
|
|
163
|
+
private lastBattery: { pct: number; atMs: number } | null = null;
|
|
164
|
+
private firstChargeUah: number | null = null;
|
|
165
|
+
private lastChargeUah: number | null = null;
|
|
166
|
+
private batterySampleCount = 0;
|
|
167
|
+
private chargingObserved = false;
|
|
168
|
+
|
|
169
|
+
private thermalSampleCount = 0;
|
|
170
|
+
private thermalInitial: DeviceThermalState | null = null;
|
|
171
|
+
private thermalLast: DeviceThermalState | null = null;
|
|
172
|
+
private thermalMaxRank = -1;
|
|
173
|
+
private thermalMaxState: DeviceThermalState | null = null;
|
|
174
|
+
private thermalThrottledCount = 0;
|
|
175
|
+
private thermalKnownCount = 0;
|
|
176
|
+
private readonly thermalTransitions: ThermalTransition[] = [];
|
|
177
|
+
|
|
178
|
+
private lowPowerSampleCount = 0;
|
|
179
|
+
private lowPowerEver = false;
|
|
180
|
+
private lowPowerLast: boolean | null = null;
|
|
181
|
+
private lowPowerTransitions = 0;
|
|
182
|
+
|
|
183
|
+
constructor(private readonly opts: { leakGrowthMbThreshold?: number } = {}) {}
|
|
184
|
+
|
|
185
|
+
/** Record one generation's differenced throughput. */
|
|
186
|
+
recordGeneration(obs: GenerationObservation): void {
|
|
187
|
+
this.generations += 1;
|
|
188
|
+
if (isFiniteNumber(obs.prefillTokensPerSecond))
|
|
189
|
+
this.prefillHist.add(obs.prefillTokensPerSecond);
|
|
190
|
+
if (isFiniteNumber(obs.decodeTokensPerSecond))
|
|
191
|
+
this.decodeHist.add(obs.decodeTokensPerSecond);
|
|
192
|
+
if (isFiniteNumber(obs.combinedTokensPerSecond))
|
|
193
|
+
this.combinedHist.add(obs.combinedTokensPerSecond);
|
|
194
|
+
if (isFiniteNumber(obs.ttftMs)) this.ttftHist.add(obs.ttftMs);
|
|
195
|
+
}
|
|
196
|
+
|
|
197
|
+
/** Record one sampled device-resource snapshot. */
|
|
198
|
+
recordResourceSample(sample: ResourceSample): void {
|
|
199
|
+
this.resourceSamples += 1;
|
|
200
|
+
|
|
201
|
+
if (isFiniteNumber(sample.residentMemoryMb))
|
|
202
|
+
this.rssSamples.push(sample.residentMemoryMb);
|
|
203
|
+
|
|
204
|
+
if (isFiniteNumber(sample.batteryLevelPct)) {
|
|
205
|
+
const entry = { pct: sample.batteryLevelPct, atMs: sample.atMs };
|
|
206
|
+
if (this.firstBattery === null) this.firstBattery = entry;
|
|
207
|
+
this.lastBattery = entry;
|
|
208
|
+
this.batterySampleCount += 1;
|
|
209
|
+
}
|
|
210
|
+
if (isFiniteNumber(sample.batteryChargeMicroAmpHours)) {
|
|
211
|
+
if (this.firstChargeUah === null)
|
|
212
|
+
this.firstChargeUah = sample.batteryChargeMicroAmpHours;
|
|
213
|
+
this.lastChargeUah = sample.batteryChargeMicroAmpHours;
|
|
214
|
+
}
|
|
215
|
+
if (sample.isCharging === true) this.chargingObserved = true;
|
|
216
|
+
|
|
217
|
+
if (sample.thermalState != null) {
|
|
218
|
+
this.thermalSampleCount += 1;
|
|
219
|
+
if (this.thermalInitial === null)
|
|
220
|
+
this.thermalInitial = sample.thermalState;
|
|
221
|
+
const known = sample.thermalState !== "unknown";
|
|
222
|
+
if (known) {
|
|
223
|
+
this.thermalKnownCount += 1;
|
|
224
|
+
const rank =
|
|
225
|
+
THERMAL_RANK[
|
|
226
|
+
sample.thermalState as Exclude<DeviceThermalState, "unknown">
|
|
227
|
+
];
|
|
228
|
+
if (rank > this.thermalMaxRank) {
|
|
229
|
+
this.thermalMaxRank = rank;
|
|
230
|
+
this.thermalMaxState = sample.thermalState;
|
|
231
|
+
}
|
|
232
|
+
if (rank >= THERMAL_RANK.serious) this.thermalThrottledCount += 1;
|
|
233
|
+
}
|
|
234
|
+
if (
|
|
235
|
+
sample.thermalState !== this.thermalLast &&
|
|
236
|
+
this.thermalTransitions.length < MAX_THERMAL_TRANSITIONS
|
|
237
|
+
) {
|
|
238
|
+
this.thermalTransitions.push({
|
|
239
|
+
atMs: sample.atMs,
|
|
240
|
+
state: sample.thermalState,
|
|
241
|
+
});
|
|
242
|
+
}
|
|
243
|
+
this.thermalLast = sample.thermalState;
|
|
244
|
+
}
|
|
245
|
+
|
|
246
|
+
if (typeof sample.lowPowerMode === "boolean") {
|
|
247
|
+
this.lowPowerSampleCount += 1;
|
|
248
|
+
if (sample.lowPowerMode) this.lowPowerEver = true;
|
|
249
|
+
if (
|
|
250
|
+
this.lowPowerLast !== null &&
|
|
251
|
+
this.lowPowerLast !== sample.lowPowerMode
|
|
252
|
+
)
|
|
253
|
+
this.lowPowerTransitions += 1;
|
|
254
|
+
this.lowPowerLast = sample.lowPowerMode;
|
|
255
|
+
}
|
|
256
|
+
}
|
|
257
|
+
|
|
258
|
+
private rssSummary(): RssSummary {
|
|
259
|
+
const n = this.rssSamples.length;
|
|
260
|
+
const firstMb = n > 0 ? (this.rssSamples[0] as number) : null;
|
|
261
|
+
const lastMb = n > 0 ? (this.rssSamples[n - 1] as number) : null;
|
|
262
|
+
const peakMb = n > 0 ? Math.max(...this.rssSamples) : null;
|
|
263
|
+
// Steady state = median of the back half (excludes cold-load ramp).
|
|
264
|
+
const backHalf =
|
|
265
|
+
n >= 2 ? this.rssSamples.slice(Math.floor(n / 2)) : this.rssSamples;
|
|
266
|
+
const steadyMb = median(backHalf);
|
|
267
|
+
const growthMb =
|
|
268
|
+
firstMb !== null && lastMb !== null ? lastMb - firstMb : null;
|
|
269
|
+
const threshold = this.opts.leakGrowthMbThreshold ?? DEFAULT_LEAK_GROWTH_MB;
|
|
270
|
+
let monotone = n >= 4;
|
|
271
|
+
for (let i = 1; i < n; i++) {
|
|
272
|
+
if ((this.rssSamples[i] as number) < (this.rssSamples[i - 1] as number)) {
|
|
273
|
+
monotone = false;
|
|
274
|
+
break;
|
|
275
|
+
}
|
|
276
|
+
}
|
|
277
|
+
const leakSuspected = monotone && growthMb !== null && growthMb > threshold;
|
|
278
|
+
return {
|
|
279
|
+
firstMb,
|
|
280
|
+
lastMb,
|
|
281
|
+
peakMb,
|
|
282
|
+
steadyMb,
|
|
283
|
+
samples: n,
|
|
284
|
+
growthMb,
|
|
285
|
+
leakSuspected,
|
|
286
|
+
};
|
|
287
|
+
}
|
|
288
|
+
|
|
289
|
+
private batterySummary(): BatterySummary {
|
|
290
|
+
const firstPct = this.firstBattery?.pct ?? null;
|
|
291
|
+
const lastPct = this.lastBattery?.pct ?? null;
|
|
292
|
+
const drainPct =
|
|
293
|
+
firstPct !== null && lastPct !== null ? firstPct - lastPct : null;
|
|
294
|
+
const energyMicroAmpHoursDelta =
|
|
295
|
+
this.firstChargeUah !== null && this.lastChargeUah !== null
|
|
296
|
+
? this.firstChargeUah - this.lastChargeUah
|
|
297
|
+
: null;
|
|
298
|
+
const durationMs =
|
|
299
|
+
this.firstBattery !== null && this.lastBattery !== null
|
|
300
|
+
? this.lastBattery.atMs - this.firstBattery.atMs
|
|
301
|
+
: null;
|
|
302
|
+
return {
|
|
303
|
+
firstPct,
|
|
304
|
+
lastPct,
|
|
305
|
+
drainPct,
|
|
306
|
+
samples: this.batterySampleCount,
|
|
307
|
+
chargingObserved: this.chargingObserved,
|
|
308
|
+
energyMicroAmpHoursDelta,
|
|
309
|
+
durationMs,
|
|
310
|
+
};
|
|
311
|
+
}
|
|
312
|
+
|
|
313
|
+
private thermalSummary(): ThermalSummary {
|
|
314
|
+
const fractionThrottled =
|
|
315
|
+
this.thermalKnownCount > 0
|
|
316
|
+
? this.thermalThrottledCount / this.thermalKnownCount
|
|
317
|
+
: null;
|
|
318
|
+
return {
|
|
319
|
+
samples: this.thermalSampleCount,
|
|
320
|
+
initialState: this.thermalInitial,
|
|
321
|
+
maxState: this.thermalMaxState,
|
|
322
|
+
transitions: [...this.thermalTransitions],
|
|
323
|
+
transitionCount: Math.max(0, this.thermalTransitions.length - 1),
|
|
324
|
+
fractionThrottled,
|
|
325
|
+
};
|
|
326
|
+
}
|
|
327
|
+
|
|
328
|
+
summary(): DeviceResourceSummary {
|
|
329
|
+
return {
|
|
330
|
+
generations: this.generations,
|
|
331
|
+
resourceSamples: this.resourceSamples,
|
|
332
|
+
prefillTokensPerSecond: this.prefillHist.summary(),
|
|
333
|
+
decodeTokensPerSecond: this.decodeHist.summary(),
|
|
334
|
+
combinedTokensPerSecond: this.combinedHist.summary(),
|
|
335
|
+
ttftMs: this.ttftHist.summary(),
|
|
336
|
+
rss: this.rssSummary(),
|
|
337
|
+
battery: this.batterySummary(),
|
|
338
|
+
thermal: this.thermalSummary(),
|
|
339
|
+
lowPowerMode: {
|
|
340
|
+
samples: this.lowPowerSampleCount,
|
|
341
|
+
everEnabled: this.lowPowerEver,
|
|
342
|
+
transitionCount: this.lowPowerTransitions,
|
|
343
|
+
},
|
|
344
|
+
};
|
|
345
|
+
}
|
|
346
|
+
}
|