@@ -57,7 +57,10 @@ const modelDefinitions = vi.hoisted(() => [
5757 size : 4 ,
5858 languages : [ "en" , "fr" ] ,
5959 accuracy : 1 ,
60- benchmark : { accuracy : { en : { grade : "A-" , errorPercent : 4 } } , secondsPer30s : 1.2 } ,
60+ benchmark : {
61+ accuracy : { en : { grade : "A-" , errorPercent : 4 } } ,
62+ secondsPer30s : { cpu : 3 , accelerated : 1.2 } ,
63+ } ,
6164 speed : 1 ,
6265 recommended : true ,
6366 supportsStreaming : true ,
@@ -144,31 +147,40 @@ it("rejects empty, misaligned, non-finite, and oversized PCM as invalid audio",
144147 }
145148} ) ;
146149
147- it . effect ( "prepares a batch model before audio arrives and reuses it for transcription" , ( ) =>
148- Effect . gen ( function * ( ) {
149- const speech = yield * SpeechService . SpeechService ;
150- yield * speech . selectModel ( "fallback-model" ) ;
151- yield * speech . prepareModel ;
152- expect ( loadNative ) . toHaveBeenCalledTimes ( 1 ) ;
153- expect ( native . transcribe ) . not . toHaveBeenCalled ( ) ;
154- native . supportsStreaming = false ;
155- yield * Effect . gen ( function * ( ) {
156- const stream = yield * speech . startStream ( {
157- ...DEFAULT_SPEECH_TRANSCRIPTION_OPTIONS ,
158- speechLanguage : "fr" ,
159- speechTranslateToEnglish : true ,
160- speechSilenceFiltering : false ,
150+ it . effect . each ( [ "CPU" , "Vulkan0" ] ) (
151+ "prepares a batch model on %s before audio arrives and reuses it for transcription" ,
152+ ( backend ) =>
153+ Effect . gen ( function * ( ) {
154+ const speech = yield * SpeechService . SpeechService ;
155+ yield * speech . selectModel ( "fallback-model" ) ;
156+ loadNative . mockImplementationOnce ( async ( ) => {
157+ native . backend = backend ;
158+ return native ;
161159 } ) ;
162- expect ( native . begin ) . toHaveBeenLastCalledWith (
163- { timestamps : "none" , language : "fr" , task : "translate" , targetLanguage : "en" } ,
164- false ,
165- "test.gguf" ,
166- ) ;
167- expect ( yield * stream . finish ) . toBe ( "hello" ) ;
168- } ) . pipe ( Effect . scoped ) ;
169- expect ( yield * speech . transcribe ( pcm ( ) , DEFAULT_SPEECH_TRANSCRIPTION_OPTIONS ) ) . toBe ( "hello" ) ;
170- expect ( loadNative ) . toHaveBeenCalledTimes ( 1 ) ;
171- } ) . pipe ( Effect . provide ( layer ) ) ,
160+ yield * speech . prepareModel ;
161+ expect ( yield * speech . status ) . toMatchObject ( {
162+ benchmarkTier : backend === "CPU" ? "cpu" : "accelerated" ,
163+ } ) ;
164+ expect ( loadNative ) . toHaveBeenCalledTimes ( 1 ) ;
165+ expect ( native . transcribe ) . not . toHaveBeenCalled ( ) ;
166+ native . supportsStreaming = false ;
167+ yield * Effect . gen ( function * ( ) {
168+ const stream = yield * speech . startStream ( {
169+ ...DEFAULT_SPEECH_TRANSCRIPTION_OPTIONS ,
170+ speechLanguage : "fr" ,
171+ speechTranslateToEnglish : true ,
172+ speechSilenceFiltering : false ,
173+ } ) ;
174+ expect ( native . begin ) . toHaveBeenLastCalledWith (
175+ { timestamps : "none" , language : "fr" , task : "translate" , targetLanguage : "en" } ,
176+ false ,
177+ "test.gguf" ,
178+ ) ;
179+ expect ( yield * stream . finish ) . toBe ( "hello" ) ;
180+ } ) . pipe ( Effect . scoped ) ;
181+ expect ( yield * speech . transcribe ( pcm ( ) , DEFAULT_SPEECH_TRANSCRIPTION_OPTIONS ) ) . toBe ( "hello" ) ;
182+ expect ( loadNative ) . toHaveBeenCalledTimes ( 1 ) ;
183+ } ) . pipe ( Effect . provide ( layer ) ) ,
172184) ;
173185
174186it . effect . each ( [ "cancellation" , "timeout" ] as const ) (
@@ -234,7 +246,7 @@ it.effect("rejects model mutations outside the offered registry", () =>
234246 expect ( current ) . not . toHaveProperty ( "benchmark" ) ;
235247 expect (
236248 ( yield * speech . models ( true ) ) . models . find ( ( model ) => model . active ) ?. benchmark ?. secondsPer30s ,
237- ) . toBe ( 1.2 ) ;
249+ ) . toEqual ( { cpu : 3 , accelerated : 1.2 } ) ;
238250 } ) . pipe ( Effect . provide ( layer ) ) ,
239251) ;
240252
@@ -609,11 +621,19 @@ it.effect.each([{ devices: [] }, { devices: [{ id: "gpu", name: "Test GPU" }] }]
609621 const speech = yield * SpeechService . SpeechService ;
610622 const discover = vi . mocked ( listNativeSpeechGpuDevices ) . mockReset ( ) ;
611623 discover . mockRejectedValueOnce ( new Error ( "driver not ready" ) ) . mockResolvedValueOnce ( devices ) ;
612- expect ( yield * speech . status ) . toMatchObject ( { gpuDevices : [ ] } ) ;
624+ expect ( yield * speech . status ) . toMatchObject ( { gpuDevices : [ ] , benchmarkTier : "cpu" } ) ;
613625 yield * speech . status ;
614626 expect ( discover ) . toHaveBeenCalledOnce ( ) ;
615627 yield * TestClock . adjust ( "1 minute" ) ;
616- expect ( yield * speech . status ) . toMatchObject ( { gpuDevices : devices } ) ;
628+ expect ( yield * speech . status ) . toMatchObject ( {
629+ gpuDevices : devices ,
630+ benchmarkTier : devices . length ? "accelerated" : "cpu" ,
631+ } ) ;
632+ expect ( yield * speech . updateAcceleration ( "cpu" ) ) . toMatchObject ( { benchmarkTier : "cpu" } ) ;
633+ expect ( yield * speech . updateAcceleration ( 'gpu:["vulkan","gpu-1"]' ) ) . toMatchObject ( {
634+ benchmarkTier : "accelerated" ,
635+ } ) ;
636+ yield * speech . updateAcceleration ( "auto" ) ;
617637 yield * TestClock . adjust ( "1 minute" ) ;
618638 expect ( yield * speech . status ) . toMatchObject ( { gpuDevices : devices } ) ;
619639 expect ( discover ) . toHaveBeenCalledTimes ( 2 ) ;
@@ -644,9 +664,11 @@ it.effect(
644664 ] ,
645665 } ;
646666 const speech = yield * SpeechService . SpeechService ;
667+ expect ( yield * speech . status ) . toMatchObject ( { benchmarkTier : "accelerated" } ) ;
647668 expect ( yield * speech . transcribe ( pcm ( ) , options ) ) . toBe (
648669 cpuSupportsVocabulary ? "git hub Bob" : "GitHub Bob" ,
649670 ) ;
671+ expect ( yield * speech . status ) . toMatchObject ( { benchmarkTier : "cpu" } ) ;
650672 expect ( yield * speech . transcribe ( pcm ( ) , DEFAULT_SPEECH_TRANSCRIPTION_OPTIONS ) ) . toBe (
651673 "hello" ,
652674 ) ;
0 commit comments