Reza2kn's picture
Reduce MLX cache-rebase allocation and report sustained GPU streaming
fddfb10 verified
Raw History Blame Contribute Delete
38.8 kB
{
"status": "complete_verified_regression",
"cases": [
{
"id": "ami-ES2004a",
"exact_tokens_and_window_schedules": true,
"counts": {
"hits": 327,
"substitutions": 32,
"deletions": 73,
"insertions": 10,
"reference_units": 432,
"hypothesis_units": 369,
"errors": 115,
"error_rate": 0.2662037037037037
},
"baseline": {
"memory": {
"os_peak_process_rss_bytes": 2338947072,
"os_peak_memory_footprint_bytes": 3111176664,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 317.58,
"user_seconds": 56.33,
"system_seconds": 43.3,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 1944959424,
"mlx_cache_bytes": 67374128,
"mlx_peak_bytes": 2750158063,
"process_peak_rss_bytes": 2338947072
},
"mature_windows": 718,
"mature_work_rtf": 0.8615857967648826
},
"candidate": {
"memory": {
"os_peak_process_rss_bytes": 2339045376,
"os_peak_memory_footprint_bytes": 2633222568,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 317.25,
"user_seconds": 62.88,
"system_seconds": 44.23,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 1944959424,
"mlx_cache_bytes": 67098655,
"mlx_peak_bytes": 2395437740,
"process_peak_rss_bytes": 2339045376
},
"mature_windows": 718,
"mature_work_rtf": 0.8642073193592007
},
"original_alignment_timing": {
"measured": true,
"actual_arrival_paced": true,
"timing_basis": "Independent host receipt of flushed JSONL since paced20ms producer start; all source words retained",
"correctly_matched_units": 327,
"ambiguous_matched_units": 1,
"ambiguity_method": "all minimum unit-cost Levenshtein paths, without timing-based selection",
"unambiguous_matched_units": 326,
"unambiguous_maximum_seconds": 0.8196784589937884,
"unmatched_reference_units": 105,
"matched_reference_fraction": 0.7569444444444444,
"min_seconds": -1.6159387909821135,
"p50_seconds": 0.572301375000734,
"p95_seconds": 0.7263229999723819,
"maximum_seconds": 0.8196784589937884,
"matched_units_above_one_second": 0,
"all_matched_units_within_one_second": null,
"whole_release_pass": false,
"limitation": "Lag summaries use one deterministic text alignment and are descriptive when alternative optimal alignments exist. Only unambiguous matches can support per-word timing claims. Omissions/substitutions remain errors, never zero-latency successes. Published reference timing uncertainty and overlapping-speaker word order are not independently corrected."
},
"supplementary_causal_timing": {
"same_session_unambiguous_matches": 326,
"cross_reset_matches": [],
"maximum_seconds": 0.8196784589937884,
"p95_seconds": 0.7263229999723819,
"over_one_second": 0,
"omitted_and_substituted_words_remain_errors": true
}
},
{
"id": "meanwhile-748OyesQy84",
"exact_tokens_and_window_schedules": true,
"counts": {
"hits": 164,
"substitutions": 23,
"deletions": 5,
"insertions": 9,
"reference_units": 192,
"hypothesis_units": 196,
"errors": 37,
"error_rate": 0.19270833333333334
},
"baseline": {
"memory": {
"os_peak_process_rss_bytes": 2339684352,
"os_peak_memory_footprint_bytes": 3103066560,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 70.68,
"user_seconds": 22.78,
"system_seconds": 17.47,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 2092941760,
"mlx_cache_bytes": 67692770,
"mlx_peak_bytes": 2750192485,
"process_peak_rss_bytes": 2339684352
},
"mature_windows": 387,
"mature_work_rtf": 0.8532934121375281
},
"candidate": {
"memory": {
"os_peak_process_rss_bytes": 2338701312,
"os_peak_memory_footprint_bytes": 2975418792,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 70.53,
"user_seconds": 24.06,
"system_seconds": 17.87,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 2092941760,
"mlx_cache_bytes": 66486713,
"mlx_peak_bytes": 2395422972,
"process_peak_rss_bytes": 2338701312
},
"mature_windows": 387,
"mature_work_rtf": 0.8633251343607103
},
"original_alignment_timing": {
"measured": false,
"actual_arrival_paced": true,
"timing_basis": "Independent host receipt of flushed JSONL since paced20ms producer start; all source words retained",
"correctly_matched_units": 0,
"ambiguous_matched_units": 0,
"ambiguity_method": "all minimum unit-cost Levenshtein paths, without timing-based selection",
"unambiguous_matched_units": 0,
"unambiguous_maximum_seconds": null,
"unmatched_reference_units": 28,
"matched_reference_fraction": null,
"min_seconds": null,
"p50_seconds": null,
"p95_seconds": null,
"maximum_seconds": null,
"matched_units_above_one_second": null,
"all_matched_units_within_one_second": null,
"whole_release_pass": false,
"limitation": "Lag summaries use one deterministic text alignment and are descriptive when alternative optimal alignments exist. Only unambiguous matches can support per-word timing claims. Omissions/substitutions remain errors, never zero-latency successes. Published reference timing uncertainty and overlapping-speaker word order are not independently corrected."
},
"supplementary_causal_timing": {
"same_session_unambiguous_matches": 0,
"cross_reset_matches": [],
"maximum_seconds": null,
"p95_seconds": null,
"over_one_second": 0,
"omitted_and_substituted_words_remain_errors": true
}
},
{
"id": "meanwhile-8prs9Pq5Xhk",
"exact_tokens_and_window_schedules": true,
"counts": {
"hits": 165,
"substitutions": 7,
"deletions": 4,
"insertions": 2,
"reference_units": 176,
"hypothesis_units": 174,
"errors": 13,
"error_rate": 0.07386363636363637
},
"baseline": {
"memory": {
"os_peak_process_rss_bytes": 2342289408,
"os_peak_memory_footprint_bytes": 3074181544,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 68.86,
"user_seconds": 22.52,
"system_seconds": 17.03,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 2092936640,
"mlx_cache_bytes": 67492791,
"mlx_peak_bytes": 2744042892,
"process_peak_rss_bytes": 2342289408
},
"mature_windows": 362,
"mature_work_rtf": 0.864825141212169
},
"candidate": {
"memory": {
"os_peak_process_rss_bytes": 2338832384,
"os_peak_memory_footprint_bytes": 2907736488,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 68.55,
"user_seconds": 23.84,
"system_seconds": 17.29,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 2092940736,
"mlx_cache_bytes": 66617812,
"mlx_peak_bytes": 2395418736,
"process_peak_rss_bytes": 2338832384
},
"mature_windows": 362,
"mature_work_rtf": 0.8641732087892188
},
"original_alignment_timing": {
"measured": false,
"actual_arrival_paced": true,
"timing_basis": "Independent host receipt of flushed JSONL since paced20ms producer start; all source words retained",
"correctly_matched_units": 0,
"ambiguous_matched_units": 0,
"ambiguity_method": "all minimum unit-cost Levenshtein paths, without timing-based selection",
"unambiguous_matched_units": 0,
"unambiguous_maximum_seconds": null,
"unmatched_reference_units": 11,
"matched_reference_fraction": null,
"min_seconds": null,
"p50_seconds": null,
"p95_seconds": null,
"maximum_seconds": null,
"matched_units_above_one_second": null,
"all_matched_units_within_one_second": null,
"whole_release_pass": false,
"limitation": "Lag summaries use one deterministic text alignment and are descriptive when alternative optimal alignments exist. Only unambiguous matches can support per-word timing claims. Omissions/substitutions remain errors, never zero-latency successes. Published reference timing uncertainty and overlapping-speaker word order are not independently corrected."
},
"supplementary_causal_timing": {
"same_session_unambiguous_matches": 0,
"cross_reset_matches": [],
"maximum_seconds": null,
"p95_seconds": null,
"over_one_second": 0,
"omitted_and_substituted_words_remain_errors": true
}
},
{
"id": "BAC009S0906W0121",
"exact_tokens_and_window_schedules": true,
"counts": {
"hits": 14,
"substitutions": 2,
"deletions": 0,
"insertions": 1,
"reference_units": 16,
"hypothesis_units": 17,
"errors": 3,
"error_rate": 0.1875
},
"baseline": {
"memory": {
"os_peak_process_rss_bytes": 2338603008,
"os_peak_memory_footprint_bytes": 2724383072,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 15.8,
"user_seconds": 8.93,
"system_seconds": 2.46,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 1998690752,
"mlx_cache_bytes": 66400996,
"mlx_peak_bytes": 2110160394,
"process_peak_rss_bytes": 2338603008
},
"mature_windows": 0,
"mature_work_rtf": null
},
"candidate": {
"memory": {
"os_peak_process_rss_bytes": 2338635776,
"os_peak_memory_footprint_bytes": 2648623456,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 15.49,
"user_seconds": 9.01,
"system_seconds": 2.45,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 1998698944,
"mlx_cache_bytes": 66471501,
"mlx_peak_bytes": 2110150922,
"process_peak_rss_bytes": 2338635776
},
"mature_windows": 0,
"mature_work_rtf": null
},
"original_alignment_timing": {
"measured": false,
"actual_arrival_paced": true,
"timing_basis": "Independent host receipt of flushed JSONL since paced20ms producer start; all source words retained",
"correctly_matched_units": 0,
"ambiguous_matched_units": 0,
"ambiguity_method": "all minimum unit-cost Levenshtein paths, without timing-based selection",
"unambiguous_matched_units": 0,
"unambiguous_maximum_seconds": null,
"unmatched_reference_units": 2,
"matched_reference_fraction": null,
"min_seconds": null,
"p50_seconds": null,
"p95_seconds": null,
"maximum_seconds": null,
"matched_units_above_one_second": null,
"all_matched_units_within_one_second": null,
"whole_release_pass": false,
"limitation": "Lag summaries use one deterministic text alignment and are descriptive when alternative optimal alignments exist. Only unambiguous matches can support per-word timing claims. Omissions/substitutions remain errors, never zero-latency successes. Published reference timing uncertainty and overlapping-speaker word order are not independently corrected."
},
"supplementary_causal_timing": {
"same_session_unambiguous_matches": 0,
"cross_reset_matches": [],
"maximum_seconds": null,
"p95_seconds": null,
"over_one_second": 0,
"omitted_and_substituted_words_remain_errors": true
}
},
{
"id": "BAC009S0906W0122",
"exact_tokens_and_window_schedules": true,
"counts": {
"hits": 13,
"substitutions": 0,
"deletions": 0,
"insertions": 0,
"reference_units": 13,
"hypothesis_units": 13,
"errors": 0,
"error_rate": 0.0
},
"baseline": {
"memory": {
"os_peak_process_rss_bytes": 2338701312,
"os_peak_memory_footprint_bytes": 2606221664,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 13.89,
"user_seconds": 8.39,
"system_seconds": 2.03,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 1973400000,
"mlx_cache_bytes": 58040427,
"mlx_peak_bytes": 2071830104,
"process_peak_rss_bytes": 2338701312
},
"mature_windows": 0,
"mature_work_rtf": null
},
"candidate": {
"memory": {
"os_peak_process_rss_bytes": 2338488320,
"os_peak_memory_footprint_bytes": 2604059024,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 13.53,
"user_seconds": 8.45,
"system_seconds": 2.0,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 1973403072,
"mlx_cache_bytes": 67025185,
"mlx_peak_bytes": 2070028168,
"process_peak_rss_bytes": 2338488320
},
"mature_windows": 0,
"mature_work_rtf": null
},
"original_alignment_timing": {
"measured": false,
"actual_arrival_paced": true,
"timing_basis": "Independent host receipt of flushed JSONL since paced20ms producer start; all source words retained",
"correctly_matched_units": 0,
"ambiguous_matched_units": 0,
"ambiguity_method": "all minimum unit-cost Levenshtein paths, without timing-based selection",
"unambiguous_matched_units": 0,
"unambiguous_maximum_seconds": null,
"unmatched_reference_units": 0,
"matched_reference_fraction": null,
"min_seconds": null,
"p50_seconds": null,
"p95_seconds": null,
"maximum_seconds": null,
"matched_units_above_one_second": null,
"all_matched_units_within_one_second": null,
"whole_release_pass": false,
"limitation": "Lag summaries use one deterministic text alignment and are descriptive when alternative optimal alignments exist. Only unambiguous matches can support per-word timing claims. Omissions/substitutions remain errors, never zero-latency successes. Published reference timing uncertainty and overlapping-speaker word order are not independently corrected."
},
"supplementary_causal_timing": {
"same_session_unambiguous_matches": 0,
"cross_reset_matches": [],
"maximum_seconds": null,
"p95_seconds": null,
"over_one_second": 0,
"omitted_and_substituted_words_remain_errors": true
}
},
{
"id": "BAC009S0907W0121",
"exact_tokens_and_window_schedules": true,
"counts": {
"hits": 16,
"substitutions": 0,
"deletions": 0,
"insertions": 0,
"reference_units": 16,
"hypothesis_units": 16,
"errors": 0,
"error_rate": 0.0
},
"baseline": {
"memory": {
"os_peak_process_rss_bytes": 2339979264,
"os_peak_memory_footprint_bytes": 2628127072,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 15.24,
"user_seconds": 8.77,
"system_seconds": 2.32,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 1987032512,
"mlx_cache_bytes": 67527451,
"mlx_peak_bytes": 2093671884,
"process_peak_rss_bytes": 2339979264
},
"mature_windows": 0,
"mature_work_rtf": null
},
"candidate": {
"memory": {
"os_peak_process_rss_bytes": 2338668544,
"os_peak_memory_footprint_bytes": 2792606048,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 14.63,
"user_seconds": 8.79,
"system_seconds": 2.24,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 1987032512,
"mlx_cache_bytes": 66452362,
"mlx_peak_bytes": 2092803544,
"process_peak_rss_bytes": 2338668544
},
"mature_windows": 0,
"mature_work_rtf": null
},
"original_alignment_timing": {
"measured": false,
"actual_arrival_paced": true,
"timing_basis": "Independent host receipt of flushed JSONL since paced20ms producer start; all source words retained",
"correctly_matched_units": 0,
"ambiguous_matched_units": 0,
"ambiguity_method": "all minimum unit-cost Levenshtein paths, without timing-based selection",
"unambiguous_matched_units": 0,
"unambiguous_maximum_seconds": null,
"unmatched_reference_units": 0,
"matched_reference_fraction": null,
"min_seconds": null,
"p50_seconds": null,
"p95_seconds": null,
"maximum_seconds": null,
"matched_units_above_one_second": null,
"all_matched_units_within_one_second": null,
"whole_release_pass": false,
"limitation": "Lag summaries use one deterministic text alignment and are descriptive when alternative optimal alignments exist. Only unambiguous matches can support per-word timing claims. Omissions/substitutions remain errors, never zero-latency successes. Published reference timing uncertainty and overlapping-speaker word order are not independently corrected."
},
"supplementary_causal_timing": {
"same_session_unambiguous_matches": 0,
"cross_reset_matches": [],
"maximum_seconds": null,
"p95_seconds": null,
"over_one_second": 0,
"omitted_and_substituted_words_remain_errors": true
}
},
{
"id": "BAC009S0907W0122",
"exact_tokens_and_window_schedules": true,
"counts": {
"hits": 15,
"substitutions": 0,
"deletions": 0,
"insertions": 0,
"reference_units": 15,
"hypothesis_units": 15,
"errors": 0,
"error_rate": 0.0
},
"baseline": {
"memory": {
"os_peak_process_rss_bytes": 2339602432,
"os_peak_memory_footprint_bytes": 2610104672,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 14.05,
"user_seconds": 8.49,
"system_seconds": 2.1,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 1977595328,
"mlx_cache_bytes": 66439737,
"mlx_peak_bytes": 2078649656,
"process_peak_rss_bytes": 2339602432
},
"mature_windows": 0,
"mature_work_rtf": null
},
"candidate": {
"memory": {
"os_peak_process_rss_bytes": 2339995648,
"os_peak_memory_footprint_bytes": 2609006944,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 13.87,
"user_seconds": 8.5,
"system_seconds": 2.06,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 1977598400,
"mlx_cache_bytes": 59412062,
"mlx_peak_bytes": 2076519752,
"process_peak_rss_bytes": 2339995648
},
"mature_windows": 0,
"mature_work_rtf": null
},
"original_alignment_timing": {
"measured": false,
"actual_arrival_paced": true,
"timing_basis": "Independent host receipt of flushed JSONL since paced20ms producer start; all source words retained",
"correctly_matched_units": 0,
"ambiguous_matched_units": 0,
"ambiguity_method": "all minimum unit-cost Levenshtein paths, without timing-based selection",
"unambiguous_matched_units": 0,
"unambiguous_maximum_seconds": null,
"unmatched_reference_units": 0,
"matched_reference_fraction": null,
"min_seconds": null,
"p50_seconds": null,
"p95_seconds": null,
"maximum_seconds": null,
"matched_units_above_one_second": null,
"all_matched_units_within_one_second": null,
"whole_release_pass": false,
"limitation": "Lag summaries use one deterministic text alignment and are descriptive when alternative optimal alignments exist. Only unambiguous matches can support per-word timing claims. Omissions/substitutions remain errors, never zero-latency successes. Published reference timing uncertainty and overlapping-speaker word order are not independently corrected."
},
"supplementary_causal_timing": {
"same_session_unambiguous_matches": 0,
"cross_reset_matches": [],
"maximum_seconds": null,
"p95_seconds": null,
"over_one_second": 0,
"omitted_and_substituted_words_remain_errors": true
}
},
{
"id": "BAC009S0908W0121",
"exact_tokens_and_window_schedules": true,
"counts": {
"hits": 14,
"substitutions": 1,
"deletions": 0,
"insertions": 0,
"reference_units": 15,
"hypothesis_units": 15,
"errors": 1,
"error_rate": 0.06666666666666667
},
"baseline": {
"memory": {
"os_peak_process_rss_bytes": 2339831808,
"os_peak_memory_footprint_bytes": 2688649544,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 17.25,
"user_seconds": 9.25,
"system_seconds": 2.81,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 2019670464,
"mlx_cache_bytes": 66145068,
"mlx_peak_bytes": 2149484042,
"process_peak_rss_bytes": 2339831808
},
"mature_windows": 0,
"mature_work_rtf": null
},
"candidate": {
"memory": {
"os_peak_process_rss_bytes": 2338619392,
"os_peak_memory_footprint_bytes": 2689091936,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 17.08,
"user_seconds": 9.44,
"system_seconds": 2.88,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 2019671488,
"mlx_cache_bytes": 66889599,
"mlx_peak_bytes": 2149483530,
"process_peak_rss_bytes": 2338619392
},
"mature_windows": 0,
"mature_work_rtf": null
},
"original_alignment_timing": {
"measured": false,
"actual_arrival_paced": true,
"timing_basis": "Independent host receipt of flushed JSONL since paced20ms producer start; all source words retained",
"correctly_matched_units": 0,
"ambiguous_matched_units": 0,
"ambiguity_method": "all minimum unit-cost Levenshtein paths, without timing-based selection",
"unambiguous_matched_units": 0,
"unambiguous_maximum_seconds": null,
"unmatched_reference_units": 1,
"matched_reference_fraction": null,
"min_seconds": null,
"p50_seconds": null,
"p95_seconds": null,
"maximum_seconds": null,
"matched_units_above_one_second": null,
"all_matched_units_within_one_second": null,
"whole_release_pass": false,
"limitation": "Lag summaries use one deterministic text alignment and are descriptive when alternative optimal alignments exist. Only unambiguous matches can support per-word timing claims. Omissions/substitutions remain errors, never zero-latency successes. Published reference timing uncertainty and overlapping-speaker word order are not independently corrected."
},
"supplementary_causal_timing": {
"same_session_unambiguous_matches": 0,
"cross_reset_matches": [],
"maximum_seconds": null,
"p95_seconds": null,
"over_one_second": 0,
"omitted_and_substituted_words_remain_errors": true
}
},
{
"id": "BAC009S0908W0122",
"exact_tokens_and_window_schedules": true,
"counts": {
"hits": 12,
"substitutions": 0,
"deletions": 0,
"insertions": 0,
"reference_units": 12,
"hypothesis_units": 12,
"errors": 0,
"error_rate": 0.0
},
"baseline": {
"memory": {
"os_peak_process_rss_bytes": 2338963456,
"os_peak_memory_footprint_bytes": 2624457056,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 14.37,
"user_seconds": 8.63,
"system_seconds": 2.2,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 1981789632,
"mlx_cache_bytes": 67165393,
"mlx_peak_bytes": 2085353832,
"process_peak_rss_bytes": 2338963456
},
"mature_windows": 0,
"mature_work_rtf": null
},
"candidate": {
"memory": {
"os_peak_process_rss_bytes": 2339487744,
"os_peak_memory_footprint_bytes": 2618198344,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 14.2,
"user_seconds": 8.62,
"system_seconds": 2.14,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 1981790656,
"mlx_cache_bytes": 57989172,
"mlx_peak_bytes": 2085305720,
"process_peak_rss_bytes": 2339487744
},
"mature_windows": 0,
"mature_work_rtf": null
},
"original_alignment_timing": {
"measured": false,
"actual_arrival_paced": true,
"timing_basis": "Independent host receipt of flushed JSONL since paced20ms producer start; all source words retained",
"correctly_matched_units": 0,
"ambiguous_matched_units": 0,
"ambiguity_method": "all minimum unit-cost Levenshtein paths, without timing-based selection",
"unambiguous_matched_units": 0,
"unambiguous_maximum_seconds": null,
"unmatched_reference_units": 0,
"matched_reference_fraction": null,
"min_seconds": null,
"p50_seconds": null,
"p95_seconds": null,
"maximum_seconds": null,
"matched_units_above_one_second": null,
"all_matched_units_within_one_second": null,
"whole_release_pass": false,
"limitation": "Lag summaries use one deterministic text alignment and are descriptive when alternative optimal alignments exist. Only unambiguous matches can support per-word timing claims. Omissions/substitutions remain errors, never zero-latency successes. Published reference timing uncertainty and overlapping-speaker word order are not independently corrected."
},
"supplementary_causal_timing": {
"same_session_unambiguous_matches": 0,
"cross_reset_matches": [],
"maximum_seconds": null,
"p95_seconds": null,
"over_one_second": 0,
"omitted_and_substituted_words_remain_errors": true
}
},
{
"id": "BAC009S0912W0121",
"exact_tokens_and_window_schedules": true,
"counts": {
"hits": 11,
"substitutions": 0,
"deletions": 0,
"insertions": 0,
"reference_units": 11,
"hypothesis_units": 11,
"errors": 0,
"error_rate": 0.0
},
"baseline": {
"memory": {
"os_peak_process_rss_bytes": 2338783232,
"os_peak_memory_footprint_bytes": 2597341536,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 13.54,
"user_seconds": 8.33,
"system_seconds": 1.96,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 1968161216,
"mlx_cache_bytes": 66544912,
"mlx_peak_bytes": 2064062920,
"process_peak_rss_bytes": 2338783232
},
"mature_windows": 0,
"mature_work_rtf": null
},
"candidate": {
"memory": {
"os_peak_process_rss_bytes": 2339536896,
"os_peak_memory_footprint_bytes": 2599192904,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 13.24,
"user_seconds": 8.33,
"system_seconds": 1.9,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 1968160192,
"mlx_cache_bytes": 67304368,
"mlx_peak_bytes": 2063504840,
"process_peak_rss_bytes": 2339536896
},
"mature_windows": 0,
"mature_work_rtf": null
},
"original_alignment_timing": {
"measured": false,
"actual_arrival_paced": true,
"timing_basis": "Independent host receipt of flushed JSONL since paced20ms producer start; all source words retained",
"correctly_matched_units": 0,
"ambiguous_matched_units": 0,
"ambiguity_method": "all minimum unit-cost Levenshtein paths, without timing-based selection",
"unambiguous_matched_units": 0,
"unambiguous_maximum_seconds": null,
"unmatched_reference_units": 0,
"matched_reference_fraction": null,
"min_seconds": null,
"p50_seconds": null,
"p95_seconds": null,
"maximum_seconds": null,
"matched_units_above_one_second": null,
"all_matched_units_within_one_second": null,
"whole_release_pass": false,
"limitation": "Lag summaries use one deterministic text alignment and are descriptive when alternative optimal alignments exist. Only unambiguous matches can support per-word timing claims. Omissions/substitutions remain errors, never zero-latency successes. Published reference timing uncertainty and overlapping-speaker word order are not independently corrected."
},
"supplementary_causal_timing": {
"same_session_unambiguous_matches": 0,
"cross_reset_matches": [],
"maximum_seconds": null,
"p95_seconds": null,
"over_one_second": 0,
"omitted_and_substituted_words_remain_errors": true
}
},
{
"id": "BAC009S0912W0122",
"exact_tokens_and_window_schedules": true,
"counts": {
"hits": 19,
"substitutions": 0,
"deletions": 0,
"insertions": 0,
"reference_units": 19,
"hypothesis_units": 19,
"errors": 0,
"error_rate": 0.0
},
"baseline": {
"memory": {
"os_peak_process_rss_bytes": 2338914304,
"os_peak_memory_footprint_bytes": 2754447688,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 15.96,
"user_seconds": 9.01,
"system_seconds": 2.55,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 2002895296,
"mlx_cache_bytes": 66088725,
"mlx_peak_bytes": 2118020298,
"process_peak_rss_bytes": 2338914304
},
"mature_windows": 0,
"mature_work_rtf": null
},
"candidate": {
"memory": {
"os_peak_process_rss_bytes": 2338701312,
"os_peak_memory_footprint_bytes": 2656913736,
"footprint_available": true,
"rss_source_unit": "bytes",
"elapsed_seconds": 15.81,
"user_seconds": 9.08,
"system_seconds": 2.54,
"scope": "Fresh native child process including loading and inference; excludes Python harness. RSS and footprint are separate measurements and must not be summed."
},
"mlx_memory": {
"mlx_active_bytes": 2002893248,
"mlx_cache_bytes": 67643508,
"mlx_peak_bytes": 2118033930,
"process_peak_rss_bytes": 2338701312
},
"mature_windows": 0,
"mature_work_rtf": null
},
"original_alignment_timing": {
"measured": false,
"actual_arrival_paced": true,
"timing_basis": "Independent host receipt of flushed JSONL since paced20ms producer start; all source words retained",
"correctly_matched_units": 0,
"ambiguous_matched_units": 0,
"ambiguity_method": "all minimum unit-cost Levenshtein paths, without timing-based selection",
"unambiguous_matched_units": 0,
"unambiguous_maximum_seconds": null,
"unmatched_reference_units": 0,
"matched_reference_fraction": null,
"min_seconds": null,
"p50_seconds": null,
"p95_seconds": null,
"maximum_seconds": null,
"matched_units_above_one_second": null,
"all_matched_units_within_one_second": null,
"whole_release_pass": false,
"limitation": "Lag summaries use one deterministic text alignment and are descriptive when alternative optimal alignments exist. Only unambiguous matches can support per-word timing claims. Omissions/substitutions remain errors, never zero-latency successes. Published reference timing uncertainty and overlapping-speaker word order are not independently corrected."
},
"supplementary_causal_timing": {
"same_session_unambiguous_matches": 0,
"cross_reset_matches": [],
"maximum_seconds": null,
"p95_seconds": null,
"over_one_second": 0,
"omitted_and_substituted_words_remain_errors": true
}
}
],
"peaks": {
"baseline": {
"rss_bytes": 2342289408,
"footprint_bytes": 3111176664,
"mlx_allocation_bytes": 2750192485
},
"candidate": {
"rss_bytes": 2339995648,
"footprint_bytes": 2975418792,
"mlx_allocation_bytes": 2395437740
}
},
"all_eleven_exact_token_and_schedule_parity": true,
"measured_mature_rtf_range": [
0.8633251343607103,
0.8642073193592007
],
"all_measured_mature_rtf_under_one": true,
"observed_same_session_unambiguous_over_one_second": 0,
"already_evaluated_inputs": true,
"fresh_heldout": false,
"production_ready": false,
"comparison_limit": "Serial runs on the same host, not an interleaved controlled speed experiment. OS footprint and RSS are distinct from MLX allocation. Errors and alignment ambiguity are retained.",
"plan_sha256": "05ec870469c6dc383f841e299650a412f1ca608a69a928a9a2b477ba19fd467c",
"campaign_sha256": "cc95c1e22acf1f0949bc7648dbf141a279ebb4e9f9598ca00bad25acf7d08908",
"script_sha256": "368b63a07853603f4eb8dd7bcaba05f73764d2c35ab220a3eb3f8d6c6e3aca37"
}