Lower Mac streaming batch size after separate paired qualification

52ecfff verified about 1 month ago

14.4 kB

{
  "format": "audio8-silero-paired-qualification-v2",
  "status": "frozen",
  "created_unix": 1790330470.812763,
  "input_manifest_sha256": {
    "data/continuous-v3/manifest.json": "7ad682b2ff707832a91af12dd3550d0e45532d4bb41b56d27cb78eac07e92cbd",
    "data/heldout-zh-v3/manifest.json": "52f63ea126debffc78ba0df29f904436943c070dd74d31b778993c566ab21df9"
  },
  "cases": [
    {
      "id": "ami-ES2004a",
      "role": "heldout",
      "language": "en",
      "sha256": "e97522bb8277e087f8afeefa607b9a7468bc8c0921b95d3ed89e64d505b307b3",
      "samples": 4928641,
      "reference": "Hmm hmm hmm Are we we're not allowed to dim the lights so people can see that a bit better Yeah Okay that's fine Am I supposed to be standing up there So Okay we've got both of these clipped on She gonna answer me Yeah or not I've got Right both of them okay Yes God Jesus it's gonna fall off Okay Yep yep Okay Okay Tu tu tu tu Hello everybody Hi good morning Um I'm Sarah the Project Manager and this is our first meeting surprisingly enough Okay this is our agenda um we will do some stuff get to know each other a bit better to feel more comfortable with each other Um then we'll go do tool training talk about the project plan discuss our own ideas and everything um and we've got twenty five minutes to do that as far as I can understand Now we're developing a remote control which you probably already know Um we want it to be original something that's uh people haven't thought of that's not out in the shops um trendy appealing to a wide market but you know not a hunk of metal and user-friendly grannies to kids maybe even pooches should be able to use it Okay um first is the functional design um this is where we all go off and do our individual work um what needs need to be fulfilled by the product um what effects the product has to have and how it's actually going to do that Um conceptual design what we're thinking how it's gonna go and then the detailed design how we're actually gonna put it into practice and make it work 'Kay Okay right We're gonna practice with the pens and draw our favourite animal on the white board I'll go first and um sum up the characteristics of that animal So Oops Okay I'll leave space for everyone else Um What's missing We're running out of blue Okay I'm not gonna ask you to guess I'm going to tell you that's supposed to be a tiger Mm And I see them Oh as sorry majestic Mm-hmm and independent and proud Now who would like to go next Yeah me 'Kay Cat Where did this come from Is that your lapel then Uh There you go yep Thank you Uh maybe you can guess what I'm trying to make A kind of dog"
    },
    {
      "id": "meanwhile-748OyesQy84",
      "role": "heldout",
      "language": "en",
      "sha256": "b64f2f62c1b8759eb7711b92500dd4cf6d32fcd84ac489ae4c8d8b3d76efc8f5",
      "samples": 976000,
      "reference": "FOLKS, IF YOU WATCH THE SHOW, YOU KNOW, I SPEND MOST OF MY TIME, RIGHT OVER THERE, CAREFULLY BLENDING FOR YOU THE DAY'S NEWSIEST, MOST TOPICAL FLOUR, EGGS, MILK, AND BUTTER, AND STRAINING IT INTO A FINE BATTER TO MAKE DELICATE, YET INFORMATIVE COMEDY PANCAKES. THEN I GLAZE THEM IN THE JUICE AND ZEST OF THE MOST RELEVANT MIDNIGHT VALENCIA ORANGES, AND DOUSE IT ALL IN A FINE DELAMAIN DE VOYAGE COGNAC, BEFORE FLAMBEING AND BASTING THEM TABLE SIDE, TO SERVE FOR YOU THE JAMES BEARD AWARD-WORTHY CREPES SUZETTE THAT IS MY NIGHTLY, MONOLOGUE. BUT SOMETIMES, JUST SOMETIMES FOLKS, I WAKE UP IN THE BAGGAGE HOLD OF A GREYHOUND BUS AS ITS BEING HOISTED BY THE SCRAPYARD CLAW TOWARD THE BURN PIT, ESCAPE TO A NEARBY ABANDONED PRICE CHOPPER, WHERE I SCROUNGE FOR OLD BREAD SCRAPS, BUSTED OPEN BAGS OF STAR FRUIT CANDIES, AND EXPIRED EGGS, CHUCK IT ALL IN A DIRTY HUBCAP AND SLAP IT OVER A TIRE FIRE BEFORE USING THE LEGS OF A STAINED PAIR OF SWEATPANTS AS OVEN MITTS TO EXTRACT AND SERVE THE DEMENTED TRANSIENT'S POUND CAKE OF NEWS THAT IS MY SEGMENT: \"MEANWHILE.\""
    },
    {
      "id": "meanwhile-8prs9Pq5Xhk",
      "role": "heldout",
      "language": "en",
      "sha256": "200fa0e6241e661951b58b244af44f11fffdfc6cfa31c4c8e7278ec714aaa36f",
      "samples": 944000,
      "reference": "FOLKS, IF YOU WATCH THE SHOW, AND I HOPE YOU DO, I SPEND A LOT OF TIME RIGHT OVER THERE, TIRELESSLY STUDYING THE LINEAGE OF THE DAY'S MOST IMPORTANT THOROUGHBRED STORIES AND HOLSTEINER HEADLINES, WORKING WITH THE BEST TRAINERS MONEY CAN BUY TO REAR THEIR COMEDY OFFSPRING WITH A HAND THAT IS STERN, YET GENTLE, INTO THE TRIPLE-CROWN-WINNING EQUINE SPECIMEN THAT IS MY NIGHTLY MONOLOGUE. BUT SOMETIMES, SOMETIMES, FOLKS I BREAK INTO AN UNINCORPORATED VETERINARY GENETICS LAB AND GRAB WHATEVER TEST TUBES I CAN FIND. AND THEN, UNDER A GROW LIGHT I GOT FROM A DISCARDED CHIA PET, I MIX THE PILFERED D.N.A. OF A HORSE AND WHATEVER WAS IN A TUBE LABELED \"KEITH-COLON-EXTRA,\" SLURRYING THE CONCOCTION WITH CAFFEINE PILLS AND A MICROWAVED RED BULL, I SCREAM-SING A PRAYER TO JANUS, INITIATOR OF HUMAN LIFE AND GOD OF TRANSFORMATION AS A HALF-HORSE, HALF-MAN FREAK, SEIZES TO LIFE BEFORE ME IN THE HIDEOUS COLLECTION OF LOOSE ANIMAL PARTS AND CORRUPTED MAN TISSUE THAT IS MY SEGMENT: MEANWHILE!"
    },
    {
      "id": "BAC009S0906W0121",
      "role": "heldout",
      "language": "zh",
      "sha256": "3007311d3481745e1d811df826d78e6bd1a9b29984725a578299b5b13955631f",
      "samples": 98992,
      "reference": "    双降  会  令  市场  看  贬  人民  币  的  情绪  持续"
    },
    {
      "id": "BAC009S0906W0122",
      "role": "heldout",
      "language": "zh",
      "sha256": "da8e43b1877e150055bd1e3382a30346b9013fe143fd91a6a9e6bf10d72ae8af",
      "samples": 68784,
      "reference": "    人民  币  未来  贬值  压力  依然  较大"
    },
    {
      "id": "BAC009S0907W0121",
      "role": "heldout",
      "language": "zh",
      "sha256": "f08f40aef11df244a750827b9685ed8a3958c2112a19cd9d6b7b9b4f7de27afb",
      "samples": 85728,
      "reference": "    也  不  代表  开发  商  资金  面  已经  不  再  紧张"
    },
    {
      "id": "BAC009S0907W0122",
      "role": "heldout",
      "language": "zh",
      "sha256": "fcfc001b953445d17b305531afa3217b09e5c2abc1ee57155c827b3a1d742b62",
      "samples": 74176,
      "reference": "    背后  可能  蕴含  着  开发  商  更多  的  窘境"
    },
    {
      "id": "BAC009S0908W0121",
      "role": "heldout",
      "language": "zh",
      "sha256": "f44d9dd8f8f45c595b45db59cd35af85836dc7df201b660074c2f64609ba5564",
      "samples": 123888,
      "reference": "    将  进一步  提振  刚  需  购房  者  入市  信心"
    },
    {
      "id": "BAC009S0908W0122",
      "role": "heldout",
      "language": "zh",
      "sha256": "f4d06fcc44df08e4d0c638971103a4d0c57c0519cebf8febc6020fa29668bdd1",
      "samples": 78432,
      "reference": "    加速  今年  楼市  成交  复苏  回暖"
    },
    {
      "id": "BAC009S0912W0121",
      "role": "heldout",
      "language": "zh",
      "sha256": "93c27170f5de04b1f4952cb1b3fde554f1fbc972547dd546cf6fa36935a88a84",
      "samples": 63439,
      "reference": "    房地  产  相关  领域  问题  频发"
    },
    {
      "id": "BAC009S0912W0122",
      "role": "heldout",
      "language": "zh",
      "sha256": "d0e900a32b91a9184f15b24c076264cc127b5a0c77fff0e07a9b25a67a290d5c",
      "samples": 104000,
      "reference": "    东  地产  财经  周刊  新  一年  度  审计  工作  报告  出炉"
    }
  ],
  "silero_model_sha256": "1a153a22f4509e292a94e67d6f9b85e8deb25b4988682b7e174c65279d8788e3",
  "silence_policy": "silero_v5_threshold0.5_silence2000ms_preroll400ms_v1",
  "vad_runtime_identity_sha256": "12a3e66090507f9f6f62ed09c0c0f76407e643f121d4e0a45e655d6704ca3800",
  "original_environment_lock_sha256": "bf1698431487b1415fb624d1f12e88bb9e4c2bdfaaa7cef88347bd24945420e5",
  "original_profile": {
    "weights": "original_bf16",
    "device": "cuda",
    "dtype": "bfloat16",
    "startup": "realtime18",
    "delay_ms": 240,
    "batch_windows": 1,
    "attention": "sdpa_math_only",
    "tf32": false
  },
  "source_sha256": {
    "baseline_probe.py": "d9fcfb9d4c99868b6f8d3a365411135238959e8833ed85a14f23e95fc05a2f34",
    "mlx_runtime/__init__.py": "687ad18a08c3dd374e55113a43b33ac1b27e208b01ce99a827187af4244ccd47",
    "mlx_runtime/silero_input.py": "8d36f9cfc8aa92ce34afd1702113cec17f688a5bb0e0cd5ca73f387821cc620d",
    "mlx_runtime/stream.py": "60bc5d7b5c6946ad63df284feb8d04bc99163bf15eab2f75d23f5a4ca7a97a55",
    "run_native_probe.py": "55c7a1c6147448af8f7a2ef950944b3f891ee7b3bbfc400dc44e07b833affc60",
    "scripts/audit_caption_cross_reset_matches_v1.py": "96ee32f197dadf768ae6da7846e3e1a8d78807b2b92ebee249acf9b032e95874",
    "scripts/freeze_mlx_q3_b2_qualification_v1.py": "c5fca0edcffc4a5d29a96fcb95779b8a10970d809e83b6a6317f7d7efd827bd6",
    "scripts/native_candidate_prefix.py": "9f9c1f1197d8660bb72c97b59a1fcb9398b7940b36c45ab88f8292a38c7d6e18",
    "scripts/prepare_silero_qualification_v3.py": "e9b81d28549fbdff5a8573ea13eb8280213dce210fa2f1b2b1dbcde9860d8444",
    "scripts/probe_native_candidate.py": "4e2e61f51bd3198539fe4487c1219b95d4fe7f6f1c889c2bdb534319231c84d8",
    "scripts/probe_native_stream.py": "c2743868c4769764fc93488497624cbce82ecddffc660b2c91cb1a24695a8961",
    "scripts/probe_torch_cuda_silero_case_v3.py": "a1dc997def15c882996d9eb3f3dafb20acfad61f7fc04e477f63614b0e2fd3b4",
    "scripts/probe_torch_cuda_silero_segments.py": "23c1a118a1db84413b814ba2d39caace06a9a30cdd05a03dcf4d3d82ba17d8e9",
    "scripts/probe_torch_stream_cuda.py": "bb046caa35155aae19d8a953c3dd522341e24c29cfef1599eb941ba47f163f5c",
    "scripts/probe_torch_stream_reference.py": "f41947e18533463a729ca15000f72ab6012133131f08aada6a74b443c2995c45",
    "scripts/run_mlx_q3_b2_qualification_v1.py": "6c370dfb7a8c982ca8d29de77cb809a190993bee044c03fdf53a5d29096c96b1",
    "scripts/run_original_q3_b2_qualification_v1.py": "6763972622e3cb7dbb9195425c826153898ec55b05e3c2a018fb9dbb9d2a924c",
    "scripts/score_stream_latency.py": "566ec98d6f6a7331d916bb104fd1be848c5d5c006a2b95997f16fce06c713eda",
    "scripts/silero_qualification_v3.py": "eca4c6bbf7fe9fd2c996aeac964a8656821cd01642f1d0b5a1dd28eafea55131",
    "scripts/validate_caption_run.py": "57b9264af3e6b16f0e6c15fd63f48e71e2182843e4a976f60b729d7cf1df2b3a",
    "scripts/validate_caption_run_v2.py": "251011d52fa3d2ff6d85dceed0fadb8730a671034e2c2debda11fcbe21036d21",
    "work/mlx-native-q3-v1/mlx_runtime/__init__.py": "687ad18a08c3dd374e55113a43b33ac1b27e208b01ce99a827187af4244ccd47",
    "work/mlx-native-q3-v1/mlx_runtime/__main__.py": "cf7390a2508a0fcad4777a4a629fc5212ab655cb45845975995f835676a834e3",
    "work/mlx-native-q3-v1/mlx_runtime/cli.py": "c8d7029a9e77436ba50739499b2132ce02d45e16c148640f3d3824bf61883158",
    "work/mlx-native-q3-v1/mlx_runtime/frontend.py": "fcfa93161c7c431fe5918fb559c52faa587c9eb0e790fae4057caed278f7f383",
    "work/mlx-native-q3-v1/mlx_runtime/gptq_q4_provenance.py": "461163cd7c68fcf5c4ff0061be1d895636f91ff31cfb12ba6c89f218aec0c418",
    "work/mlx-native-q3-v1/mlx_runtime/math.py": "839517e331f76bb3b000998e8368761c18575e67c07cc6be49cdd5029e634851",
    "work/mlx-native-q3-v1/mlx_runtime/model.py": "d9cd49025be5263a24ad86bdc29b817aaac348bd2a85059aa4f05cfe8b409b83",
    "work/mlx-native-q3-v1/mlx_runtime/native_model_format.py": "600b957bb9f6ed38b1078d2d2d870d201b120774415d4b5d8f4121784f4be258",
    "work/mlx-native-q3-v1/mlx_runtime/native_weights.py": "4e518da4e765a90a3e419ad0c9827651f2a5829c55052bcef9ec855cf2ea2ffd",
    "work/mlx-native-q3-v1/mlx_runtime/silero_input.py": "8d36f9cfc8aa92ce34afd1702113cec17f688a5bb0e0cd5ca73f387821cc620d",
    "work/mlx-native-q3-v1/mlx_runtime/silero_stream.py": "1406e2e8972c638dedfd0dcf9b8e72d06296aed503f1bef28d86a1b9d8aa595e",
    "work/mlx-native-q3-v1/mlx_runtime/stream.py": "60bc5d7b5c6946ad63df284feb8d04bc99163bf15eab2f75d23f5a4ca7a97a55",
    "work/mlx-native-q3-v1/mlx_runtime/weights.py": "4c34bb6e77e41cdaa268498d80a65dc595a8c46e273d2414e2f122e3f0a04c1d"
  },
  "selected_candidate": {
    "mlx_bundle": "artifacts/native-q3-mlp-rt240-v1",
    "mlx_source_root": "work/mlx-native-q3-v1",
    "assets": {
      "artifacts/native-q3-mlp-rt240-v1/manifest.json": "f132d8c4dd0af2eaf49cef779635e83678e3169f002015e9a0a35c9f6f685656",
      "artifacts/native-q3-mlp-rt240-v1/model.a8m": "e38b5a14c8636e1d9a586cd725538d17164554c8ef52f60f7fd293791acaa6c4",
      "models/original/config.json": "744baed356a8c86a9791fc76cecb73219ed4aff795147c6cdc082337a13f019c",
      "artifacts/frontend-primitive.a8m": "ab3a69e95c9a7fd54adeec09c9cdfe4c79fa81f85908f2154d074bde9ee43102",
      "artifacts/native-tokenizer-v1.a8tok": "7fdc1e903297ea1fa438b77503db993c50a67f8b325b4e55754f08b8579c5410",
      "vendor/Audio8-ASR-Infinite/client/vendor/silero_vad.onnx": "1a153a22f4509e292a94e67d6f9b85e8deb25b4988682b7e174c65279d8788e3",
      ".venv-mlx/bin/python": "d6b64f766d3b08326aa10cdb37c9d922e3af38b57ec07caa28b894e5fccf6e69"
    },
    "profile": {
      "mlx": {
        "startup": "realtime18",
        "delay_ms": 240,
        "dtype": "float16",
        "cache_dtype": "bfloat16",
        "math_mode": "compiled",
        "fuse_projections": true,
        "warmup_projections": true,
        "batch_windows": 2
      }
    }
  },
  "selected_on_exposed_data_only": {
    "path": "reports/mlx-q3-batch-selection-v1/selection.json",
    "sha256": "598beb2702d42c0028bcf7e18f6e4465d9167e384904ccddde7b68bf330683c9"
  },
  "evaluation_contract": {
    "arms": [
      "mlx",
      "original_cuda"
    ],
    "paired_original_required": true,
    "all_reference_words_including_vad_omissions": true,
    "no_fitting_or_candidate_selection_on_these_outputs": true,
    "report_all_errors_and_blank_regressions": true,
    "report_all_matched_word_lags_and_alignment_uncertainty": true,
    "real_time_requires_mature_cache_rtf_below_one": true,
    "one_second_is_after_word_end": true,
    "memory_target_soft": true,
    "automatic_production_promotion": false,
    "primary_scorer": "unchanged frozen global WER and text-only temporal alignment",
    "supplementary_causal_audit": "Reject cross-reset matches for temporal identity; retain them as unresolved recognition evidence, never latency successes. Report all same-session unambiguous matches, maximum, and over-one-second count alongside the unchanged original score."
  },
  "production_ready": false
}

	{
	"format": "audio8-silero-paired-qualification-v2",
	"status": "frozen",
	"created_unix": 1790330470.812763,
	"input_manifest_sha256": {
	"data/continuous-v3/manifest.json": "7ad682b2ff707832a91af12dd3550d0e45532d4bb41b56d27cb78eac07e92cbd",
	"data/heldout-zh-v3/manifest.json": "52f63ea126debffc78ba0df29f904436943c070dd74d31b778993c566ab21df9"
	},
	"cases": [
	{
	"id": "ami-ES2004a",
	"role": "heldout",
	"language": "en",
	"sha256": "e97522bb8277e087f8afeefa607b9a7468bc8c0921b95d3ed89e64d505b307b3",
	"samples": 4928641,
	"reference": "Hmm hmm hmm Are we we're not allowed to dim the lights so people can see that a bit better Yeah Okay that's fine Am I supposed to be standing up there So Okay we've got both of these clipped on She gonna answer me Yeah or not I've got Right both of them okay Yes God Jesus it's gonna fall off Okay Yep yep Okay Okay Tu tu tu tu Hello everybody Hi good morning Um I'm Sarah the Project Manager and this is our first meeting surprisingly enough Okay this is our agenda um we will do some stuff get to know each other a bit better to feel more comfortable with each other Um then we'll go do tool training talk about the project plan discuss our own ideas and everything um and we've got twenty five minutes to do that as far as I can understand Now we're developing a remote control which you probably already know Um we want it to be original something that's uh people haven't thought of that's not out in the shops um trendy appealing to a wide market but you know not a hunk of metal and user-friendly grannies to kids maybe even pooches should be able to use it Okay um first is the functional design um this is where we all go off and do our individual work um what needs need to be fulfilled by the product um what effects the product has to have and how it's actually going to do that Um conceptual design what we're thinking how it's gonna go and then the detailed design how we're actually gonna put it into practice and make it work 'Kay Okay right We're gonna practice with the pens and draw our favourite animal on the white board I'll go first and um sum up the characteristics of that animal So Oops Okay I'll leave space for everyone else Um What's missing We're running out of blue Okay I'm not gonna ask you to guess I'm going to tell you that's supposed to be a tiger Mm And I see them Oh as sorry majestic Mm-hmm and independent and proud Now who would like to go next Yeah me 'Kay Cat Where did this come from Is that your lapel then Uh There you go yep Thank you Uh maybe you can guess what I'm trying to make A kind of dog"
	},
	{
	"id": "meanwhile-748OyesQy84",
	"role": "heldout",
	"language": "en",
	"sha256": "b64f2f62c1b8759eb7711b92500dd4cf6d32fcd84ac489ae4c8d8b3d76efc8f5",
	"samples": 976000,
	"reference": "FOLKS, IF YOU WATCH THE SHOW, YOU KNOW, I SPEND MOST OF MY TIME, RIGHT OVER THERE, CAREFULLY BLENDING FOR YOU THE DAY'S NEWSIEST, MOST TOPICAL FLOUR, EGGS, MILK, AND BUTTER, AND STRAINING IT INTO A FINE BATTER TO MAKE DELICATE, YET INFORMATIVE COMEDY PANCAKES. THEN I GLAZE THEM IN THE JUICE AND ZEST OF THE MOST RELEVANT MIDNIGHT VALENCIA ORANGES, AND DOUSE IT ALL IN A FINE DELAMAIN DE VOYAGE COGNAC, BEFORE FLAMBEING AND BASTING THEM TABLE SIDE, TO SERVE FOR YOU THE JAMES BEARD AWARD-WORTHY CREPES SUZETTE THAT IS MY NIGHTLY, MONOLOGUE. BUT SOMETIMES, JUST SOMETIMES FOLKS, I WAKE UP IN THE BAGGAGE HOLD OF A GREYHOUND BUS AS ITS BEING HOISTED BY THE SCRAPYARD CLAW TOWARD THE BURN PIT, ESCAPE TO A NEARBY ABANDONED PRICE CHOPPER, WHERE I SCROUNGE FOR OLD BREAD SCRAPS, BUSTED OPEN BAGS OF STAR FRUIT CANDIES, AND EXPIRED EGGS, CHUCK IT ALL IN A DIRTY HUBCAP AND SLAP IT OVER A TIRE FIRE BEFORE USING THE LEGS OF A STAINED PAIR OF SWEATPANTS AS OVEN MITTS TO EXTRACT AND SERVE THE DEMENTED TRANSIENT'S POUND CAKE OF NEWS THAT IS MY SEGMENT: \"MEANWHILE.\""
	},
	{
	"id": "meanwhile-8prs9Pq5Xhk",
	"role": "heldout",
	"language": "en",
	"sha256": "200fa0e6241e661951b58b244af44f11fffdfc6cfa31c4c8e7278ec714aaa36f",
	"samples": 944000,
	"reference": "FOLKS, IF YOU WATCH THE SHOW, AND I HOPE YOU DO, I SPEND A LOT OF TIME RIGHT OVER THERE, TIRELESSLY STUDYING THE LINEAGE OF THE DAY'S MOST IMPORTANT THOROUGHBRED STORIES AND HOLSTEINER HEADLINES, WORKING WITH THE BEST TRAINERS MONEY CAN BUY TO REAR THEIR COMEDY OFFSPRING WITH A HAND THAT IS STERN, YET GENTLE, INTO THE TRIPLE-CROWN-WINNING EQUINE SPECIMEN THAT IS MY NIGHTLY MONOLOGUE. BUT SOMETIMES, SOMETIMES, FOLKS I BREAK INTO AN UNINCORPORATED VETERINARY GENETICS LAB AND GRAB WHATEVER TEST TUBES I CAN FIND. AND THEN, UNDER A GROW LIGHT I GOT FROM A DISCARDED CHIA PET, I MIX THE PILFERED D.N.A. OF A HORSE AND WHATEVER WAS IN A TUBE LABELED \"KEITH-COLON-EXTRA,\" SLURRYING THE CONCOCTION WITH CAFFEINE PILLS AND A MICROWAVED RED BULL, I SCREAM-SING A PRAYER TO JANUS, INITIATOR OF HUMAN LIFE AND GOD OF TRANSFORMATION AS A HALF-HORSE, HALF-MAN FREAK, SEIZES TO LIFE BEFORE ME IN THE HIDEOUS COLLECTION OF LOOSE ANIMAL PARTS AND CORRUPTED MAN TISSUE THAT IS MY SEGMENT: MEANWHILE!"
	},
	{
	"id": "BAC009S0906W0121",
	"role": "heldout",
	"language": "zh",
	"sha256": "3007311d3481745e1d811df826d78e6bd1a9b29984725a578299b5b13955631f",
	"samples": 98992,
	"reference": " 双降会令市场看贬人民币的情绪持续"
	},
	{
	"id": "BAC009S0906W0122",
	"role": "heldout",
	"language": "zh",
	"sha256": "da8e43b1877e150055bd1e3382a30346b9013fe143fd91a6a9e6bf10d72ae8af",
	"samples": 68784,
	"reference": " 人民币未来贬值压力依然较大"
	},
	{
	"id": "BAC009S0907W0121",
	"role": "heldout",
	"language": "zh",
	"sha256": "f08f40aef11df244a750827b9685ed8a3958c2112a19cd9d6b7b9b4f7de27afb",
	"samples": 85728,
	"reference": " 也不代表开发商资金面已经不再紧张"
	},
	{
	"id": "BAC009S0907W0122",
	"role": "heldout",
	"language": "zh",
	"sha256": "fcfc001b953445d17b305531afa3217b09e5c2abc1ee57155c827b3a1d742b62",
	"samples": 74176,
	"reference": " 背后可能蕴含着开发商更多的窘境"
	},
	{
	"id": "BAC009S0908W0121",
	"role": "heldout",
	"language": "zh",
	"sha256": "f44d9dd8f8f45c595b45db59cd35af85836dc7df201b660074c2f64609ba5564",
	"samples": 123888,
	"reference": " 将进一步提振刚需购房者入市信心"
	},
	{
	"id": "BAC009S0908W0122",
	"role": "heldout",
	"language": "zh",
	"sha256": "f4d06fcc44df08e4d0c638971103a4d0c57c0519cebf8febc6020fa29668bdd1",
	"samples": 78432,
	"reference": " 加速今年楼市成交复苏回暖"
	},
	{
	"id": "BAC009S0912W0121",
	"role": "heldout",
	"language": "zh",
	"sha256": "93c27170f5de04b1f4952cb1b3fde554f1fbc972547dd546cf6fa36935a88a84",
	"samples": 63439,
	"reference": " 房地产相关领域问题频发"
	},
	{
	"id": "BAC009S0912W0122",
	"role": "heldout",
	"language": "zh",
	"sha256": "d0e900a32b91a9184f15b24c076264cc127b5a0c77fff0e07a9b25a67a290d5c",
	"samples": 104000,
	"reference": " 东地产财经周刊新一年度审计工作报告出炉"
	}
	],
	"silero_model_sha256": "1a153a22f4509e292a94e67d6f9b85e8deb25b4988682b7e174c65279d8788e3",
	"silence_policy": "silero_v5_threshold0.5_silence2000ms_preroll400ms_v1",
	"vad_runtime_identity_sha256": "12a3e66090507f9f6f62ed09c0c0f76407e643f121d4e0a45e655d6704ca3800",
	"original_environment_lock_sha256": "bf1698431487b1415fb624d1f12e88bb9e4c2bdfaaa7cef88347bd24945420e5",
	"original_profile": {
	"weights": "original_bf16",
	"device": "cuda",
	"dtype": "bfloat16",
	"startup": "realtime18",
	"delay_ms": 240,
	"batch_windows": 1,
	"attention": "sdpa_math_only",
	"tf32": false
	},
	"source_sha256": {
	"baseline_probe.py": "d9fcfb9d4c99868b6f8d3a365411135238959e8833ed85a14f23e95fc05a2f34",
	"mlx_runtime/__init__.py": "687ad18a08c3dd374e55113a43b33ac1b27e208b01ce99a827187af4244ccd47",
	"mlx_runtime/silero_input.py": "8d36f9cfc8aa92ce34afd1702113cec17f688a5bb0e0cd5ca73f387821cc620d",
	"mlx_runtime/stream.py": "60bc5d7b5c6946ad63df284feb8d04bc99163bf15eab2f75d23f5a4ca7a97a55",
	"run_native_probe.py": "55c7a1c6147448af8f7a2ef950944b3f891ee7b3bbfc400dc44e07b833affc60",
	"scripts/audit_caption_cross_reset_matches_v1.py": "96ee32f197dadf768ae6da7846e3e1a8d78807b2b92ebee249acf9b032e95874",
	"scripts/freeze_mlx_q3_b2_qualification_v1.py": "c5fca0edcffc4a5d29a96fcb95779b8a10970d809e83b6a6317f7d7efd827bd6",
	"scripts/native_candidate_prefix.py": "9f9c1f1197d8660bb72c97b59a1fcb9398b7940b36c45ab88f8292a38c7d6e18",
	"scripts/prepare_silero_qualification_v3.py": "e9b81d28549fbdff5a8573ea13eb8280213dce210fa2f1b2b1dbcde9860d8444",
	"scripts/probe_native_candidate.py": "4e2e61f51bd3198539fe4487c1219b95d4fe7f6f1c889c2bdb534319231c84d8",
	"scripts/probe_native_stream.py": "c2743868c4769764fc93488497624cbce82ecddffc660b2c91cb1a24695a8961",
	"scripts/probe_torch_cuda_silero_case_v3.py": "a1dc997def15c882996d9eb3f3dafb20acfad61f7fc04e477f63614b0e2fd3b4",
	"scripts/probe_torch_cuda_silero_segments.py": "23c1a118a1db84413b814ba2d39caace06a9a30cdd05a03dcf4d3d82ba17d8e9",
	"scripts/probe_torch_stream_cuda.py": "bb046caa35155aae19d8a953c3dd522341e24c29cfef1599eb941ba47f163f5c",
	"scripts/probe_torch_stream_reference.py": "f41947e18533463a729ca15000f72ab6012133131f08aada6a74b443c2995c45",
	"scripts/run_mlx_q3_b2_qualification_v1.py": "6c370dfb7a8c982ca8d29de77cb809a190993bee044c03fdf53a5d29096c96b1",
	"scripts/run_original_q3_b2_qualification_v1.py": "6763972622e3cb7dbb9195425c826153898ec55b05e3c2a018fb9dbb9d2a924c",
	"scripts/score_stream_latency.py": "566ec98d6f6a7331d916bb104fd1be848c5d5c006a2b95997f16fce06c713eda",
	"scripts/silero_qualification_v3.py": "eca4c6bbf7fe9fd2c996aeac964a8656821cd01642f1d0b5a1dd28eafea55131",
	"scripts/validate_caption_run.py": "57b9264af3e6b16f0e6c15fd63f48e71e2182843e4a976f60b729d7cf1df2b3a",
	"scripts/validate_caption_run_v2.py": "251011d52fa3d2ff6d85dceed0fadb8730a671034e2c2debda11fcbe21036d21",
	"work/mlx-native-q3-v1/mlx_runtime/__init__.py": "687ad18a08c3dd374e55113a43b33ac1b27e208b01ce99a827187af4244ccd47",
	"work/mlx-native-q3-v1/mlx_runtime/__main__.py": "cf7390a2508a0fcad4777a4a629fc5212ab655cb45845975995f835676a834e3",
	"work/mlx-native-q3-v1/mlx_runtime/cli.py": "c8d7029a9e77436ba50739499b2132ce02d45e16c148640f3d3824bf61883158",
	"work/mlx-native-q3-v1/mlx_runtime/frontend.py": "fcfa93161c7c431fe5918fb559c52faa587c9eb0e790fae4057caed278f7f383",
	"work/mlx-native-q3-v1/mlx_runtime/gptq_q4_provenance.py": "461163cd7c68fcf5c4ff0061be1d895636f91ff31cfb12ba6c89f218aec0c418",
	"work/mlx-native-q3-v1/mlx_runtime/math.py": "839517e331f76bb3b000998e8368761c18575e67c07cc6be49cdd5029e634851",
	"work/mlx-native-q3-v1/mlx_runtime/model.py": "d9cd49025be5263a24ad86bdc29b817aaac348bd2a85059aa4f05cfe8b409b83",
	"work/mlx-native-q3-v1/mlx_runtime/native_model_format.py": "600b957bb9f6ed38b1078d2d2d870d201b120774415d4b5d8f4121784f4be258",
	"work/mlx-native-q3-v1/mlx_runtime/native_weights.py": "4e518da4e765a90a3e419ad0c9827651f2a5829c55052bcef9ec855cf2ea2ffd",
	"work/mlx-native-q3-v1/mlx_runtime/silero_input.py": "8d36f9cfc8aa92ce34afd1702113cec17f688a5bb0e0cd5ca73f387821cc620d",
	"work/mlx-native-q3-v1/mlx_runtime/silero_stream.py": "1406e2e8972c638dedfd0dcf9b8e72d06296aed503f1bef28d86a1b9d8aa595e",
	"work/mlx-native-q3-v1/mlx_runtime/stream.py": "60bc5d7b5c6946ad63df284feb8d04bc99163bf15eab2f75d23f5a4ca7a97a55",
	"work/mlx-native-q3-v1/mlx_runtime/weights.py": "4c34bb6e77e41cdaa268498d80a65dc595a8c46e273d2414e2f122e3f0a04c1d"
	},
	"selected_candidate": {
	"mlx_bundle": "artifacts/native-q3-mlp-rt240-v1",
	"mlx_source_root": "work/mlx-native-q3-v1",
	"assets": {
	"artifacts/native-q3-mlp-rt240-v1/manifest.json": "f132d8c4dd0af2eaf49cef779635e83678e3169f002015e9a0a35c9f6f685656",
	"artifacts/native-q3-mlp-rt240-v1/model.a8m": "e38b5a14c8636e1d9a586cd725538d17164554c8ef52f60f7fd293791acaa6c4",
	"models/original/config.json": "744baed356a8c86a9791fc76cecb73219ed4aff795147c6cdc082337a13f019c",
	"artifacts/frontend-primitive.a8m": "ab3a69e95c9a7fd54adeec09c9cdfe4c79fa81f85908f2154d074bde9ee43102",
	"artifacts/native-tokenizer-v1.a8tok": "7fdc1e903297ea1fa438b77503db993c50a67f8b325b4e55754f08b8579c5410",
	"vendor/Audio8-ASR-Infinite/client/vendor/silero_vad.onnx": "1a153a22f4509e292a94e67d6f9b85e8deb25b4988682b7e174c65279d8788e3",
	".venv-mlx/bin/python": "d6b64f766d3b08326aa10cdb37c9d922e3af38b57ec07caa28b894e5fccf6e69"
	},
	"profile": {
	"mlx": {
	"startup": "realtime18",
	"delay_ms": 240,
	"dtype": "float16",
	"cache_dtype": "bfloat16",
	"math_mode": "compiled",
	"fuse_projections": true,
	"warmup_projections": true,
	"batch_windows": 2
	}
	}
	},
	"selected_on_exposed_data_only": {
	"path": "reports/mlx-q3-batch-selection-v1/selection.json",
	"sha256": "598beb2702d42c0028bcf7e18f6e4465d9167e384904ccddde7b68bf330683c9"
	},
	"evaluation_contract": {
	"arms": [
	"mlx",
	"original_cuda"
	],
	"paired_original_required": true,
	"all_reference_words_including_vad_omissions": true,
	"no_fitting_or_candidate_selection_on_these_outputs": true,
	"report_all_errors_and_blank_regressions": true,
	"report_all_matched_word_lags_and_alignment_uncertainty": true,
	"real_time_requires_mature_cache_rtf_below_one": true,
	"one_second_is_after_word_end": true,
	"memory_target_soft": true,
	"automatic_production_promotion": false,
	"primary_scorer": "unchanged frozen global WER and text-only temporal alignment",
	"supplementary_causal_audit": "Reject cross-reset matches for temporal identity; retain them as unresolved recognition evidence, never latency successes. Report all same-session unambiguous matches, maximum, and over-one-second count alongside the unchanged original score."
	},
	"production_ready": false
	}