nightly_run

Nightly benchmark run

Generated. Do not edit. Produced by .github/workflows/bench-nightly.yml; every edit is overwritten by the next run. The curated figures, measured on a named machine, are in performance. The last known reading for a method quiet tonight is in benchmark_latest.

Read the ratios, not the means. These run on a GitHub hosted runner: a shared VM whose hardware differs from night to night and whose neighbours are unknown. An absolute figure here is not comparable to the performance page, and not reliably comparable to yesterday's. A ratio against a baseline measured in the same run, on the same VM, in the same minute, is.

This run

• Commit: 2483f1a00691271083f00baf3835de96bf0a4076
• Previous run: 2483f1a00691271083f00baf3835de96bf0a4076
• Runner: Linux / X64 (GitHub hosted)

Classes re-run

Selected by tools/select_benchmarks.py from the sources that changed since the previous run:

• BatchEmbeddingBenchmarks
• BlockedTableBenchmarks
• BpeBenchmarks
• BpeScalingBenchmarks
• BucketRouteDiagnostics
• FuzzBenchmarks
• IndelBenchmarks
• LcsGateBenchmarks
• LevenshteinBenchmarks
• LevenshteinCodePointBenchmarks
• MetricsBenchmarks
• MyersGateBenchmarks
• PersistenceBenchmarks
• StopWordBenchmarks
• VectorMathBenchmarks
• VectorizerBenchmarks

Lodestar.Text.Benchmarks.BatchEmbeddingBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3

Method	CorpusSize	Mean	Error	StdDev	Ratio	Gen0	Gen1	Allocated	Alloc Ratio
UnitLoop	1	5.762 μs	0.4998 μs	0.0274 μs	1.00	0.1678	-	2.76 KB	1.00
EmbedBatch	1	5.862 μs	0.2816 μs	0.0154 μs	1.02	0.1907	-	3.16 KB	1.14
EmbedBatchBucketed	1	5.953 μs	1.1615 μs	0.0637 μs	1.03	0.1907	-	3.16 KB	1.14
UnitLoop	8	104.244 μs	1.3103 μs	0.0718 μs	1.00	8.1787	0.2441	134.29 KB	1.00
EmbedBatch	8	67.069 μs	2.3774 μs	0.1303 μs	0.64	7.6904	0.3662	127.31 KB	0.95
EmbedBatchBucketed	8	65.684 μs	2.0940 μs	0.1148 μs	0.63	7.6904	0.3662	127.31 KB	0.95
UnitLoop	32	385.679 μs	9.9524 μs	0.5455 μs	1.00	28.3203	0.9766	468.71 KB	1.00
EmbedBatch	32	237.153 μs	27.3526 μs	1.4993 μs	0.61	26.8555	1.7090	441.32 KB	0.94
EmbedBatchBucketed	32	224.372 μs	44.3849 μs	2.4329 μs	0.58	26.1230	1.7090	427.8 KB	0.91
UnitLoop	128	1,474.941 μs	17.4570 μs	0.9569 μs	1.00	113.2813	5.8594	1874.78 KB	1.00
EmbedBatch	128	922.552 μs	30.2488 μs	1.6580 μs	0.63	107.4219	15.6250	1764.42 KB	0.94
EmbedBatchBucketed	128	856.390 μs	39.9231 μs	2.1883 μs	0.58	103.5156	15.6250	1696.9 KB	0.91

Lodestar.Text.Benchmarks.BlockedTableBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3

Method	length	Mean	Error	StdDev	Allocated
Latin	1000	55.84 μs	5.850 μs	0.321 μs	-
Cjk	1000	63.61 μs	0.579 μs	0.032 μs	-
Latin	10000	5,508.30 μs	610.059 μs	33.439 μs	-
Cjk	10000	6,967.70 μs	377.859 μs	20.712 μs	-
Latin	65536	228,427.27 μs	7,904.409 μs	433.267 μs	-

Lodestar.Text.Benchmarks.BpeBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3

Method	Mean	Error	StdDev	Ratio	Gen0	Allocated	Alloc Ratio
Unigram	594.3 ms	6.59 ms	0.36 ms	1.00	32000.0000	519.51 MB	1.00
Bpe	570.9 ms	32.10 ms	1.76 ms	0.96	7000.0000	112.18 MB	0.22

Lodestar.Text.Benchmarks.BpeScalingBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3

Method	Length	Mean	Error	StdDev	Gen0	Allocated
BpeOnOnePathologicalToken	512	103.1 μs	12.26 μs	0.67 μs	1.2207	20.38 KB
BpeOnOnePathologicalToken	1024	210.1 μs	12.75 μs	0.70 μs	2.4414	39.93 KB
BpeOnOnePathologicalToken	2048	499.6 μs	34.15 μs	1.87 μs	3.9063	78.98 KB
BpeOnOnePathologicalToken	4096	1,015.4 μs	63.85 μs	3.50 μs	7.8125	157.03 KB

Lodestar.Text.Benchmarks.BucketRouteDiagnostics-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3

Method	Alphabet	Mean	Error	StdDev	Allocated
DpGroup	cjk	19.04 μs	1.690 μs	0.093 μs	-
MyersGroup	cjk	171.84 μs	0.489 μs	0.027 μs	-
DpGroup	latin	10.52 μs	0.065 μs	0.004 μs	-
MyersGroup	latin	114.12 μs	22.211 μs	1.217 μs	-

Lodestar.Text.Benchmarks.FuzzBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3

Method	Mean	Error	StdDev	Ratio	RatioSD	Gen0	Allocated	Alloc Ratio
Ratio	108.6 ns	1.07 ns	0.06 ns	1.00	0.00	-	-	NA
PartialRatio	13,469.7 ns	252.92 ns	13.86 ns	123.98	0.12	-	-	NA
TokenSortRatio	941.0 ns	93.16 ns	5.11 ns	8.66	0.04	0.0782	1312 B	NA
TokenSetRatio	3,227.4 ns	410.85 ns	22.52 ns	29.71	0.18	0.3433	5760 B	NA
WRatio	4,496.2 ns	490.25 ns	26.87 ns	41.39	0.22	0.4272	7200 B	NA

Lodestar.Text.Benchmarks.IndelBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3

Method	Length	Mean	Error	StdDev	Ratio	RatioSD	Allocated	Alloc Ratio
Distance_Utf16	8	29.43 ns	0.294 ns	0.016 ns	1.00	0.00	-	NA
Distance_CodePoint	8	132.54 ns	3.118 ns	0.171 ns	4.50	0.01	-	NA
NormalizedSimilarity_Utf16	8	30.06 ns	1.463 ns	0.080 ns	1.02	0.00	-	NA
SubsequenceLength_Utf16	8	28.91 ns	0.785 ns	0.043 ns	0.98	0.00	-	NA
Distance_Utf16	12	31.63 ns	3.247 ns	0.178 ns	1.00	0.01	-	NA
Distance_CodePoint	12	139.51 ns	1.259 ns	0.069 ns	4.41	0.02	-	NA
NormalizedSimilarity_Utf16	12	33.76 ns	0.177 ns	0.010 ns	1.07	0.01	-	NA
SubsequenceLength_Utf16	12	32.21 ns	0.943 ns	0.052 ns	1.02	0.01	-	NA
Distance_Utf16	16	34.86 ns	1.354 ns	0.074 ns	1.00	0.00	-	NA
Distance_CodePoint	16	147.95 ns	6.170 ns	0.338 ns	4.24	0.01	-	NA
NormalizedSimilarity_Utf16	16	36.36 ns	1.323 ns	0.073 ns	1.04	0.00	-	NA
SubsequenceLength_Utf16	16	32.31 ns	0.650 ns	0.036 ns	0.93	0.00	-	NA
Distance_Utf16	20	39.04 ns	0.853 ns	0.047 ns	1.00	0.00	-	NA
Distance_CodePoint	20	188.03 ns	14.826 ns	0.813 ns	4.82	0.02	-	NA
NormalizedSimilarity_Utf16	20	38.32 ns	0.319 ns	0.017 ns	0.98	0.00	-	NA
SubsequenceLength_Utf16	20	34.56 ns	2.110 ns	0.116 ns	0.89	0.00	-	NA
Distance_Utf16	24	58.18 ns	5.415 ns	0.297 ns	1.00	0.01	-	NA
Distance_CodePoint	24	852.69 ns	252.558 ns	13.844 ns	14.66	0.22	-	NA
NormalizedSimilarity_Utf16	24	60.75 ns	1.469 ns	0.081 ns	1.04	0.00	-	NA
SubsequenceLength_Utf16	24	61.33 ns	4.978 ns	0.273 ns	1.05	0.01	-	NA
Distance_Utf16	32	69.01 ns	1.429 ns	0.078 ns	1.00	0.00	-	NA
Distance_CodePoint	32	1,030.47 ns	202.566 ns	11.103 ns	14.93	0.14	-	NA
NormalizedSimilarity_Utf16	32	75.85 ns	3.178 ns	0.174 ns	1.10	0.00	-	NA
SubsequenceLength_Utf16	32	66.12 ns	3.105 ns	0.170 ns	0.96	0.00	-	NA
Distance_Utf16	128	868.23 ns	3.086 ns	0.169 ns	1.00	0.00	-	NA
Distance_CodePoint	128	17,840.47 ns	291.377 ns	15.971 ns	20.55	0.02	-	NA
NormalizedSimilarity_Utf16	128	874.09 ns	3.676 ns	0.201 ns	1.01	0.00	-	NA
SubsequenceLength_Utf16	128	874.70 ns	4.497 ns	0.246 ns	1.01	0.00	-	NA
Distance_Utf16	512	8,060.50 ns	67.091 ns	3.677 ns	1.00	0.00	-	NA
Distance_CodePoint	512	307,439.52 ns	109,944.936 ns	6,026.453 ns	38.14	0.65	-	NA
NormalizedSimilarity_Utf16	512	8,133.36 ns	723.181 ns	39.640 ns	1.01	0.00	-	NA
SubsequenceLength_Utf16	512	8,116.36 ns	137.228 ns	7.522 ns	1.01	0.00	-	NA

Lodestar.Text.Benchmarks.LcsGateBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3

Method	Band	Mean	Error	StdDev	Ratio	RatioSD	Allocated	Alloc Ratio
Dp	8	130.10 ns	1.332 ns	0.073 ns	1.00	0.00	-	NA
Kernel	8	58.92 ns	1.086 ns	0.060 ns	0.45	0.00	-	NA
Dp_Cjk	8	130.19 ns	3.683 ns	0.202 ns	1.00	0.00	-	NA
Kernel_Cjk	8	107.32 ns	4.332 ns	0.237 ns	0.82	0.00	-	NA
Dp	12	231.72 ns	239.874 ns	13.148 ns	1.00	0.07	-	NA
Kernel	12	68.80 ns	4.671 ns	0.256 ns	0.30	0.01	-	NA
Dp_Cjk	12	221.20 ns	4.267 ns	0.234 ns	0.96	0.05	-	NA
Kernel_Cjk	12	117.90 ns	3.860 ns	0.212 ns	0.51	0.02	-	NA
Dp	14	275.82 ns	3.556 ns	0.195 ns	1.00	0.00	-	NA
Kernel	14	73.31 ns	1.814 ns	0.099 ns	0.27	0.00	-	NA
Dp_Cjk	14	305.57 ns	404.022 ns	22.146 ns	1.11	0.07	-	NA
Kernel_Cjk	14	124.66 ns	3.272 ns	0.179 ns	0.45	0.00	-	NA
Dp	16	400.36 ns	78.860 ns	4.323 ns	1.00	0.01	-	NA
Kernel	16	77.64 ns	2.030 ns	0.111 ns	0.19	0.00	-	NA
Dp_Cjk	16	369.10 ns	495.391 ns	27.154 ns	0.92	0.06	-	NA
Kernel_Cjk	16	131.18 ns	45.416 ns	2.489 ns	0.33	0.01	-	NA
Dp	18	718.57 ns	291.538 ns	15.980 ns	1.00	0.03	-	NA
Kernel	18	84.33 ns	2.013 ns	0.110 ns	0.12	0.00	-	NA
Dp_Cjk	18	725.55 ns	427.335 ns	23.424 ns	1.01	0.03	-	NA
Kernel_Cjk	18	136.62 ns	5.086 ns	0.279 ns	0.19	0.00	-	NA
Dp	20	818.21 ns	8.643 ns	0.474 ns	1.00	0.00	-	NA
Kernel	20	88.16 ns	3.900 ns	0.214 ns	0.11	0.00	-	NA
Dp_Cjk	20	825.52 ns	21.519 ns	1.180 ns	1.01	0.00	-	NA
Kernel_Cjk	20	141.43 ns	3.195 ns	0.175 ns	0.17	0.00	-	NA
Dp	24	995.70 ns	20.263 ns	1.111 ns	1.00	0.00	-	NA
Kernel	24	96.85 ns	0.892 ns	0.049 ns	0.10	0.00	-	NA
Dp_Cjk	24	995.35 ns	39.796 ns	2.181 ns	1.00	0.00	-	NA
Kernel_Cjk	24	154.12 ns	0.966 ns	0.053 ns	0.15	0.00	-	NA
Dp	32	1,577.38 ns	50.653 ns	2.776 ns	1.00	0.00	-	NA
Kernel	32	117.86 ns	1.799 ns	0.099 ns	0.07	0.00	-	NA
Dp_Cjk	32	1,595.94 ns	482.806 ns	26.464 ns	1.01	0.01	-	NA
Kernel_Cjk	32	187.49 ns	3.924 ns	0.215 ns	0.12	0.00	-	NA
Dp	48	3,277.13 ns	856.685 ns	46.958 ns	1.00	0.02	-	NA
Kernel	48	144.40 ns	2.373 ns	0.130 ns	0.04	0.00	-	NA
Dp_Cjk	48	3,278.82 ns	961.936 ns	52.727 ns	1.00	0.02	-	NA
Kernel_Cjk	48	261.14 ns	11.160 ns	0.612 ns	0.08	0.00	-	NA
Dp	64	5,256.35 ns	2,238.724 ns	122.712 ns	1.00	0.03	-	NA
Kernel	64	171.50 ns	2.435 ns	0.133 ns	0.03	0.00	-	NA
Dp_Cjk	64	5,383.11 ns	964.200 ns	52.851 ns	1.02	0.02	-	NA
Kernel_Cjk	64	314.39 ns	44.990 ns	2.466 ns	0.06	0.00	-	NA
Dp	96	10,993.46 ns	9,055.212 ns	496.347 ns	1.00	0.05	-	NA
Kernel	96	738.90 ns	52.545 ns	2.880 ns	0.07	0.00	-	NA
Dp_Cjk	96	11,318.38 ns	3,704.579 ns	203.060 ns	1.03	0.04	-	NA
Kernel_Cjk	96	1,090.31 ns	32.858 ns	1.801 ns	0.10	0.00	-	NA

Lodestar.Text.Benchmarks.LevenshteinBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3

Method	Length	Mean	Error	StdDev	Ratio	Allocated	Alloc Ratio
Distance_Utf16	8	27.52 ns	0.114 ns	0.006 ns	1.00	-	NA
Distance_CodePoint	8	125.07 ns	0.586 ns	0.032 ns	4.55	-	NA
NormalizedSimilarity_Utf16	8	28.12 ns	0.656 ns	0.036 ns	1.02	-	NA
Distance_Utf16	64	312.38 ns	13.702 ns	0.751 ns	1.00	-	NA
Distance_CodePoint	64	757.50 ns	90.473 ns	4.959 ns	2.42	-	NA
NormalizedSimilarity_Utf16	64	313.51 ns	28.638 ns	1.570 ns	1.00	-	NA
Distance_Utf16	512	15,400.45 ns	144.129 ns	7.900 ns	1.00	-	NA
Distance_CodePoint	512	18,427.47 ns	456.752 ns	25.036 ns	1.20	-	NA
NormalizedSimilarity_Utf16	512	15,510.60 ns	62.237 ns	3.411 ns	1.01	-	NA

Lodestar.Text.Benchmarks.LevenshteinCodePointBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3

Method	Length	Distinct	Mean	Error	StdDev	Ratio	Allocated	Alloc Ratio
Distance_CodePoint	16	32	358.5 ns	30.81 ns	1.69 ns	1.00	-	NA
Distance_Utf16	16	32	257.8 ns	20.08 ns	1.10 ns	0.72	-	NA
Distance_CodePoint	16	512	352.0 ns	21.35 ns	1.17 ns	1.00	-	NA
Distance_Utf16	16	512	256.1 ns	0.38 ns	0.02 ns	0.73	-	NA
Distance_CodePoint	24	32	446.3 ns	11.73 ns	0.64 ns	1.00	-	NA
Distance_Utf16	24	32	352.7 ns	20.41 ns	1.12 ns	0.79	-	NA
Distance_CodePoint	24	512	445.6 ns	2.96 ns	0.16 ns	1.00	-	NA
Distance_Utf16	24	512	348.4 ns	4.96 ns	0.27 ns	0.78	-	NA
Distance_CodePoint	32	32	538.0 ns	27.77 ns	1.52 ns	1.00	-	NA
Distance_Utf16	32	32	445.6 ns	7.84 ns	0.43 ns	0.83	-	NA
Distance_CodePoint	32	512	537.8 ns	5.50 ns	0.30 ns	1.00	-	NA
Distance_Utf16	32	512	444.8 ns	3.69 ns	0.20 ns	0.83	-	NA
Distance_CodePoint	40	32	627.8 ns	22.28 ns	1.22 ns	1.00	-	NA
Distance_Utf16	40	32	1,345.8 ns	12.52 ns	0.69 ns	2.14	-	NA
Distance_CodePoint	40	512	631.5 ns	16.35 ns	0.90 ns	1.00	-	NA
Distance_Utf16	40	512	1,331.5 ns	7.39 ns	0.41 ns	2.11	-	NA
Distance_CodePoint	128	32	2,624.1 ns	7.50 ns	0.41 ns	1.00	-	NA
Distance_Utf16	128	32	5,714.6 ns	26.36 ns	1.44 ns	2.18	-	NA
Distance_CodePoint	128	512	2,574.2 ns	326.71 ns	17.91 ns	1.00	-	NA
Distance_Utf16	128	512	5,762.7 ns	209.88 ns	11.50 ns	2.24	-	NA
Distance_CodePoint	512	32	19,777.9 ns	185.10 ns	10.15 ns	1.00	-	NA
Distance_Utf16	512	32	65,862.4 ns	2,793.35 ns	153.11 ns	3.33	-	NA
Distance_CodePoint	512	512	446,929.8 ns	56,597.44 ns	3,102.30 ns	1.00	-	NA
Distance_Utf16	512	512	71,111.7 ns	2,576.49 ns	141.23 ns	0.16	-	NA

Lodestar.Text.Benchmarks.MetricsBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3

Method	Samples	Classes	Mean	Error	StdDev	Gen0	Allocated
Matrix	1000	2	7.095 μs	5.8493 μs	0.3206 μs	0.0153	312 B
MatrixWeighted	1000	2	7.185 μs	0.9194 μs	0.0504 μs	0.0153	312 B
AccuracyScore	1000	2	1.030 μs	0.0093 μs	0.0005 μs	-	-
F1Macro	1000	2	7.130 μs	0.3066 μs	0.0168 μs	0.0229	472 B
Report	1000	2	9.833 μs	0.3122 μs	0.0171 μs	0.3815	6520 B
Matrix	1000	10	7.576 μs	0.5966 μs	0.0327 μs	0.0687	1248 B
MatrixWeighted	1000	10	7.287 μs	0.2115 μs	0.0116 μs	0.0687	1248 B
AccuracyScore	1000	10	1.032 μs	0.0426 μs	0.0023 μs	-	-
F1Macro	1000	10	7.737 μs	0.4397 μs	0.0241 μs	0.0916	1664 B
Report	1000	10	14.570 μs	0.7803 μs	0.0428 μs	0.9155	15496 B
Matrix	100000	2	797.538 μs	64.1144 μs	3.5143 μs	-	313 B
MatrixWeighted	100000	2	834.942 μs	227.2187 μs	12.4546 μs	-	313 B
AccuracyScore	100000	2	142.996 μs	119.4107 μs	6.5453 μs	-	-
F1Macro	100000	2	796.810 μs	41.6167 μs	2.2812 μs	-	473 B
Report	100000	2	900.869 μs	52.3567 μs	2.8698 μs	-	6544 B
Matrix	100000	10	974.593 μs	18.9619 μs	1.0394 μs	-	1249 B
MatrixWeighted	100000	10	943.110 μs	147.3727 μs	8.0780 μs	-	1248 B
AccuracyScore	100000	10	233.056 μs	417.3221 μs	22.8748 μs	-	-
F1Macro	100000	10	971.798 μs	14.4674 μs	0.7930 μs	-	1665 B
Report	100000	10	952.680 μs	44.2457 μs	2.4253 μs	-	15841 B
Matrix	1000000	2	8,581.748 μs	726.0817 μs	39.7990 μs	-	324 B
MatrixWeighted	1000000	2	8,642.139 μs	962.1497 μs	52.7387 μs	-	324 B
AccuracyScore	1000000	2	1,932.130 μs	12.9613 μs	0.7105 μs	-	-
F1Macro	1000000	2	8,453.837 μs	233.3103 μs	12.7885 μs	-	484 B
Report	1000000	2	8,986.848 μs	291.7913 μs	15.9941 μs	-	6572 B
Matrix	1000000	10	10,441.496 μs	594.7467 μs	32.6001 μs	-	1260 B
MatrixWeighted	1000000	10	10,063.510 μs	1,483.4199 μs	81.3112 μs	-	1260 B
AccuracyScore	1000000	10	3,110.547 μs	77.5317 μs	4.2498 μs	-	-
F1Macro	1000000	10	10,132.870 μs	875.3936 μs	47.9833 μs	-	1676 B
Report	1000000	10	9,981.726 μs	204.1232 μs	11.1887 μs	-	15892 B

Lodestar.Text.Benchmarks.MyersGateBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3

Method	Band	Mean	Error	StdDev	Ratio	Allocated	Alloc Ratio
Dp	4	81.78 ns	1.845 ns	0.101 ns	1.00	-	NA
Kernel	4	81.01 ns	1.083 ns	0.059 ns	0.99	-	NA
Dp_Cjk	4	81.94 ns	2.094 ns	0.115 ns	1.00	-	NA
Kernel_Cjk	4	82.82 ns	0.443 ns	0.024 ns	1.01	-	NA
Dp	6	117.18 ns	3.937 ns	0.216 ns	1.00	-	NA
Kernel	6	87.38 ns	0.980 ns	0.054 ns	0.75	-	NA
Dp_Cjk	6	124.62 ns	12.693 ns	0.696 ns	1.06	-	NA
Kernel_Cjk	6	147.74 ns	1.601 ns	0.088 ns	1.26	-	NA
Dp	8	157.35 ns	5.569 ns	0.305 ns	1.00	-	NA
Kernel	8	100.06 ns	0.691 ns	0.038 ns	0.64	-	NA
Dp_Cjk	8	157.92 ns	3.807 ns	0.209 ns	1.00	-	NA
Kernel_Cjk	8	196.39 ns	26.646 ns	1.461 ns	1.25	-	NA
Dp	10	204.07 ns	37.288 ns	2.044 ns	1.00	-	NA
Kernel	10	110.08 ns	1.136 ns	0.062 ns	0.54	-	NA
Dp_Cjk	10	216.10 ns	0.488 ns	0.027 ns	1.06	-	NA
Kernel_Cjk	10	169.70 ns	4.199 ns	0.230 ns	0.83	-	NA
Dp	12	270.13 ns	31.187 ns	1.709 ns	1.00	-	NA
Kernel	12	118.31 ns	5.541 ns	0.304 ns	0.44	-	NA
Dp_Cjk	12	267.45 ns	15.366 ns	0.842 ns	0.99	-	NA
Kernel_Cjk	12	178.86 ns	7.544 ns	0.414 ns	0.66	-	NA
Dp	16	433.18 ns	14.454 ns	0.792 ns	1.00	-	NA
Kernel	16	139.14 ns	10.085 ns	0.553 ns	0.32	-	NA
Dp_Cjk	16	431.50 ns	6.720 ns	0.368 ns	1.00	-	NA
Kernel_Cjk	16	198.34 ns	0.850 ns	0.047 ns	0.46	-	NA
Dp	24	863.01 ns	24.164 ns	1.325 ns	1.00	-	NA
Kernel	24	180.19 ns	2.073 ns	0.114 ns	0.21	-	NA
Dp_Cjk	24	865.80 ns	23.700 ns	1.299 ns	1.00	-	NA
Kernel_Cjk	24	245.68 ns	19.229 ns	1.054 ns	0.28	-	NA
Dp	32	1,480.35 ns	9.306 ns	0.510 ns	1.00	-	NA
Kernel	32	233.69 ns	5.915 ns	0.324 ns	0.16	-	NA
Dp_Cjk	32	1,483.84 ns	78.067 ns	4.279 ns	1.00	-	NA
Kernel_Cjk	32	294.69 ns	2.142 ns	0.117 ns	0.20	-	NA
Dp	48	3,243.26 ns	118.066 ns	6.472 ns	1.00	-	NA
Kernel	48	300.98 ns	3.421 ns	0.188 ns	0.09	-	NA
Dp_Cjk	48	3,265.45 ns	693.897 ns	38.035 ns	1.01	-	NA
Kernel_Cjk	48	388.54 ns	3.044 ns	0.167 ns	0.12	-	NA
Dp	64	5,655.40 ns	27.010 ns	1.481 ns	1.00	-	NA
Kernel	64	382.46 ns	2.633 ns	0.144 ns	0.07	-	NA
Dp_Cjk	64	5,657.95 ns	610.936 ns	33.487 ns	1.00	-	NA
Kernel_Cjk	64	483.75 ns	1.169 ns	0.064 ns	0.09	-	NA
Dp	96	12,540.86 ns	534.400 ns	29.292 ns	1.00	-	NA
Kernel	96	1,261.79 ns	5.311 ns	0.291 ns	0.10	-	NA
Dp_Cjk	96	12,533.05 ns	439.813 ns	24.108 ns	1.00	-	NA
Kernel_Cjk	96	1,723.19 ns	44.204 ns	2.423 ns	0.14	-	NA

Lodestar.Text.Benchmarks.PersistenceBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3

Method	Mean	Error	StdDev	Gen0	Gen1	Gen2	Allocated
VocabTxt	4.309 ms	2.3881 ms	0.1309 ms	117.1875	109.3750	39.0625	3.62 MB
TokenizerJsonWordPiece	11.236 ms	2.5204 ms	0.1381 ms	187.5000	171.8750	46.8750	5.72 MB
TokenizerJsonUnigram	11.128 ms	0.9725 ms	0.0533 ms	93.7500	78.1250	31.2500	4.64 MB
SpieceModel	4.084 ms	2.4965 ms	0.1368 ms	109.3750	101.5625	31.2500	3.36 MB
TfidfSave	1.978 ms	0.1366 ms	0.0075 ms	27.3438	23.4375	23.4375	2.09 MB
TfidfLoad	4.625 ms	1.5118 ms	0.0829 ms	85.9375	78.1250	23.4375	2.86 MB
EmbeddingIndexSave	5.153 ms	0.7840 ms	0.0430 ms	445.3125	445.3125	445.3125	39.64 MB
EmbeddingIndexLoad	5.114 ms	3.8587 ms	0.2115 ms	500.0000	468.7500	437.5000	35.35 MB

Lodestar.Text.Benchmarks.StopWordBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3

Method	Documents	Mean	Error	StdDev	Ratio	Gen0	Gen1	Gen2	Allocated	Alloc Ratio
Count	200	7.550 ms	0.8547 ms	0.0469 ms	1.00	500.0000	234.3750	62.5000	7.92 MB	1.00
CountWithStopWords	200	6.409 ms	0.2601 ms	0.0143 ms	0.85	390.6250	187.5000	-	6.34 MB	0.80
Hashing	200	7.442 ms	0.9934 ms	0.0545 ms	0.99	507.8125	171.8750	70.3125	8.08 MB	1.02
HashingWithStopWords	200	6.451 ms	0.2164 ms	0.0119 ms	0.85	406.2500	156.2500	-	6.52 MB	0.82
Count	1000	30.961 ms	2.8716 ms	0.1574 ms	1.00	2718.7500	1031.2500	531.2500	38.64 MB	1.00
CountWithStopWords	1000	24.055 ms	3.2810 ms	0.1798 ms	0.78	1968.7500	781.2500	250.0000	30.99 MB	0.80
Hashing	1000	29.047 ms	3.0191 ms	0.1655 ms	0.94	2562.5000	750.0000	593.7500	39.55 MB	1.02
HashingWithStopWords	1000	24.710 ms	2.1486 ms	0.1178 ms	0.80	2031.2500	625.0000	250.0000	31.83 MB	0.82

Lodestar.Text.Benchmarks.VectorMathBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3

Method	Dim	Mean	Error	StdDev	Ratio	Allocated	Alloc Ratio
Dot	384	51.47 ns	2.328 ns	0.128 ns	1.00	-	NA
L2Norm	384	48.74 ns	8.218 ns	0.450 ns	0.95	-	NA
Dot	768	100.10 ns	1.911 ns	0.105 ns	1.00	-	NA
L2Norm	768	90.59 ns	1.355 ns	0.074 ns	0.91	-	NA
Dot	1024	132.84 ns	1.428 ns	0.078 ns	1.00	-	NA
L2Norm	1024	125.01 ns	0.419 ns	0.023 ns	0.94	-	NA

Lodestar.Text.Benchmarks.VectorizerBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
AMD EPYC 9V74, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3

Method	Documents	Mean	Error	StdDev	Ratio	RatioSD	Gen0	Gen1	Gen2	Allocated	Alloc Ratio
Count	200	3.001 ms	1.9437 ms	0.1065 ms	1.00	0.04	93.7500	39.0625	-	1.6 MB	1.00
Tfidf	200	3.243 ms	6.5386 ms	0.3584 ms	1.08	0.11	101.5625	39.0625	-	1.63 MB	1.02
CountBigrams	200	3.859 ms	0.6201 ms	0.0340 ms	1.29	0.04	171.8750	109.3750	-	2.78 MB	1.74
Hashing	200	3.058 ms	1.8891 ms	0.1035 ms	1.02	0.04	93.7500	23.4375	-	1.6 MB	1.00
Count	1000	7.184 ms	1.9917 ms	0.1092 ms	1.00	0.02	484.3750	343.7500	62.5000	7.83 MB	1.00
Tfidf	1000	7.315 ms	0.9306 ms	0.0510 ms	1.02	0.01	484.3750	312.5000	93.7500	7.97 MB	1.02
CountBigrams	1000	11.861 ms	1.7554 ms	0.0962 ms	1.65	0.02	906.2500	375.0000	265.6250	13.42 MB	1.71
Hashing	1000	7.049 ms	1.0836 ms	0.0594 ms	0.98	0.01	492.1875	156.2500	70.3125	7.85 MB	1.00

Against rapidfuzz, in this same run

Both sides on this VM in these minutes, which is what makes the ratio readable where the absolutes are not.

• indel
• levenshtein
• metrics
• persistence

compare-indel

Python: rapidfuzz 3.14.5 (py 3.12.14)
C#:     Lodestar.Text on .NET 10.0.11 (mode Utf16Unit)

alphabet	length	Python ns/pair	C# ns/pair	speedup (py/C#)
latin	8	113.3	23.6	4.79x C# faster
latin	32	159.4	68.2	2.34x C# faster
latin	128	478.2	850.2	1.78x Py faster
latin	512	4862.9	8277.5	1.70x Py faster
cjk	8	128.4	23.7	5.42x C# faster
cjk	32	234.1	144.9	1.62x C# faster
cjk	128	2047.9	1740.6	1.18x C# faster
cjk	512	16797.4	11995.0	1.40x C# faster

Note: Indel is len(a)+len(b)-2*LCS on both sides, so this compares the subsequence kernels. Lodestar's is Hyyro's bit-parallel LLCS above a pattern of 8 and a rolling-row dynamic program below it (#273).

compare-levenshtein

Python: rapidfuzz 3.14.5 (py 3.12.14)
C#:     Lodestar.Text on .NET 10.0.11 (mode Utf16Unit)

alphabet	length	Python ns/pair	C# ns/pair	speedup (py/C#)
latin	8	137.6	18.0	7.63x C# faster
latin	32	256.8	127.9	2.01x C# faster
latin	128	1812.8	1557.1	1.16x C# faster
latin	512	15595.6	16669.3	1.07x Py faster
cjk	8	150.1	18.1	8.31x C# faster
cjk	32	299.7	190.6	1.57x C# faster
cjk	128	3028.6	2511.6	1.21x C# faster
cjk	512	26191.9	20163.4	1.30x C# faster

Note: Python times the realistic per-call loop; rapidfuzz's C core uses the bit-parallel Myers algorithm, so it scales better on long strings.

compare-metrics

Python: {'scikit-learn': '1.9.0', 'numpy': '2.5.1'} (py 3.12.14)
C#:     Lodestar on .NET 10.0.11

operation	C# ms	Py ms	wall	C# cpu	Py cpu	cpu
confusion_matrix_n1000_k2	0.009	0.773	86.83x	0.009	0.773	86.83x
accuracy_n1000_k2	0.001	0.402	382.64x	0.001	0.402	382.67x
precision_recall_f1_macro_n1000_k2	0.007	1.411	193.56x	0.007	1.411	193.53x
classification_report_n1000_k2	0.010	5.328	545.78x	0.010	5.327	545.77x
roc_auc_binary_n1000_k2	0.016	1.562	99.29x	0.016	1.562	99.29x
balanced_accuracy_n1000_k2	0.007	0.837	113.92x	0.007	0.837	113.91x
matthews_n1000_k2	0.007	1.591	221.82x	0.007	1.591	221.80x
cohen_kappa_n1000_k2	0.007	0.890	120.97x	0.007	0.890	120.97x
mse_n1000_k2	0.003	0.219	80.72x	0.003	0.219	80.71x
mae_n1000_k2	0.003	0.220	80.92x	0.003	0.220	80.93x
median_ae_n1000_k2	0.006	0.236	39.92x	0.006	0.235	39.92x
r2_n1000_k2	0.003	0.273	99.97x	0.003	0.273	99.98x
confusion_matrix_n1000_k10	0.009	0.788	85.45x	0.009	0.788	85.45x
accuracy_n1000_k10	0.001	0.413	390.91x	0.001	0.413	390.83x
precision_recall_f1_macro_n1000_k10	0.008	1.453	182.67x	0.008	1.453	182.67x
classification_report_n1000_k10	0.014	5.586	399.87x	0.014	5.586	399.89x
roc_auc_ovr_macro_n1000_k10	0.458	8.239	17.99x	0.458	8.237	17.99x
balanced_accuracy_n1000_k10	0.008	0.854	107.40x	0.008	0.854	107.39x
matthews_n1000_k10	0.008	1.635	202.06x	0.008	1.635	202.06x
cohen_kappa_n1000_k10	0.008	0.897	107.51x	0.008	0.897	107.50x
mse_n1000_k10	0.003	0.220	80.51x	0.003	0.220	80.49x
mae_n1000_k10	0.003	0.218	79.83x	0.003	0.218	79.83x
median_ae_n1000_k10	0.006	0.237	40.34x	0.006	0.237	40.34x
r2_n1000_k10	0.003	0.272	99.66x	0.003	0.272	99.66x
confusion_matrix_n100000_k2	1.002	11.034	11.01x	1.002	11.033	11.01x
accuracy_n100000_k2	0.106	3.795	35.84x	0.106	3.794	35.83x
precision_recall_f1_macro_n100000_k2	0.859	12.616	14.69x	0.859	12.615	14.69x
classification_report_n100000_k2	0.867	27.037	31.18x	0.867	27.034	31.18x
roc_auc_binary_n100000_k2	2.933	28.848	9.84x	2.933	28.844	9.83x
balanced_accuracy_n100000_k2	0.857	11.119	12.97x	0.857	11.118	12.97x
matthews_n100000_k2	0.859	22.264	25.92x	0.859	22.260	25.91x
cohen_kappa_n100000_k2	0.857	11.185	13.05x	0.857	11.183	13.05x
mse_n100000_k2	0.268	0.374	1.39x	0.268	0.374	1.39x
mae_n100000_k2	0.268	0.373	1.39x	0.268	0.373	1.39x
median_ae_n100000_k2	0.676	1.893	2.80x	0.693	1.893	2.73x
r2_n100000_k2	0.260	0.584	2.25x	0.260	0.584	2.25x
confusion_matrix_n100000_k10	0.956	10.978	11.49x	0.956	10.976	11.48x
accuracy_n100000_k10	0.106	3.795	35.85x	0.106	3.794	35.84x
precision_recall_f1_macro_n100000_k10	1.016	13.315	13.10x	1.016	13.314	13.10x
classification_report_n100000_k10	0.979	29.903	30.54x	0.979	29.901	30.54x
roc_auc_ovr_macro_n100000_k10	31.134	235.252	7.56x	31.140	235.244	7.55x
balanced_accuracy_n100000_k10	0.974	11.150	11.44x	0.974	11.148	11.44x
matthews_n100000_k10	0.981	23.079	23.53x	0.981	23.076	23.53x
cohen_kappa_n100000_k10	0.999	11.240	11.25x	0.999	11.240	11.25x
mse_n100000_k10	0.267	0.379	1.42x	0.267	0.378	1.42x
mae_n100000_k10	0.267	0.378	1.42x	0.267	0.378	1.42x
median_ae_n100000_k10	0.674	1.892	2.81x	0.713	1.891	2.65x
r2_n100000_k10	0.260	0.593	2.28x	0.260	0.593	2.28x
confusion_matrix_n1000000_k2	8.627	103.104	11.95x	8.626	103.099	11.95x
accuracy_n1000000_k2	2.296	34.332	14.95x	2.296	34.329	14.95x
precision_recall_f1_macro_n1000000_k2	8.803	112.680	12.80x	8.801	112.675	12.80x
classification_report_n1000000_k2	8.730	219.722	25.17x	8.730	219.713	25.17x
roc_auc_binary_n1000000_k2	42.019	312.188	7.43x	42.016	312.164	7.43x
balanced_accuracy_n1000000_k2	8.758	102.976	11.76x	8.757	102.964	11.76x
matthews_n1000000_k2	8.748	206.897	23.65x	8.747	206.894	23.65x
cohen_kappa_n1000000_k2	8.758	102.634	11.72x	8.757	102.625	11.72x
mse_n1000000_k2	2.671	1.767	0.66x	2.670	1.767	0.66x
mae_n1000000_k2	2.674	1.748	0.65x	2.674	1.748	0.65x
median_ae_n1000000_k2	6.401	15.371	2.40x	6.479	15.368	2.37x
r2_n1000000_k2	2.598	3.422	1.32x	2.598	3.421	1.32x
confusion_matrix_n1000000_k10	9.945	102.266	10.28x	9.944	102.236	10.28x
accuracy_n1000000_k10	3.313	34.128	10.30x	3.313	34.119	10.30x
precision_recall_f1_macro_n1000000_k10	10.441	119.427	11.44x	10.441	119.413	11.44x
classification_report_n1000000_k10	10.252	247.454	24.14x	10.252	247.408	24.13x
balanced_accuracy_n1000000_k10	10.275	102.510	9.98x	10.275	102.500	9.98x
matthews_n1000000_k10	10.251	214.608	20.93x	10.249	214.592	20.94x
cohen_kappa_n1000000_k10	10.317	102.481	9.93x	10.315	102.470	9.93x
mse_n1000000_k10	5.819	1.757	0.30x	5.819	1.756	0.30x
mae_n1000000_k10	5.807	1.739	0.30x	5.806	1.739	0.30x
median_ae_n1000000_k10	6.281	15.346	2.44x	6.353	15.344	2.42x
r2_n1000000_k10	2.951	3.328	1.13x	2.950	3.328	1.13x

ratio > 1 means Lodestar is faster. cpu is the merge gate for this branch (docs/guides/performance.md): every operation, every size, must be >= 1x.

BELOW GATE on processor time: mse_n1000000_k2 0.66x mae_n1000000_k2 0.65x mse_n1000000_k10 0.30x mae_n1000000_k10 0.30x

compare-persistence

Python: {'tokenizers': '0.23.1', 'sentencepiece': '0.2.2', 'scikit-learn': '1.9.0', 'numpy': '2.5.1'} (py 3.12.14)
C#:     Lodestar on .NET 10.0.11

operation	C# ms	Py ms	wall	C# cpu	Py cpu	cpu	C# bytes	Py bytes
vocab_txt	5.210	9.855	1.89x	5.506	9.855	1.79x	228,891	228,891
tokenizer_json_wordpiece	12.215	17.157	1.40x	12.900	17.156	1.33x	706,526	706,526
tokenizer_json_unigram	11.075	36.994	3.34x	11.360	36.989	3.26x	1,990,038	1,990,038
spiece_model	4.696	29.936	6.38x	4.891	29.935	6.12x	533,084	533,084
tfidf_save	1.657	2.402	1.45x	1.706	2.402	1.41x	581,787	591,922
tfidf_load	4.623	4.201	0.91x	4.819	4.201	0.87x	581,787	591,922
embedding_index_save	4.660	1.342	0.29x	5.214	1.341	0.26x	20,589,007	15,360,128
embedding_index_load	4.621	1.366	0.30x	5.243	1.366	0.26x	20,589,007	15,360,128
embedding_index_load_file	5.735	0.875	0.15x	6.452	0.874	0.14x	20,589,007	15,360,128
embedding_index_load_memory	3.250	1.326	0.41x	3.640	1.326	0.36x	20,589,007	15,360,128
embedding_index_view_floor	0.000	0.001	81.35x	0.000	0.001	81.35x	20,589,007	15,360,128
embedding_index_save_gzip	454.638	639.341	1.41x	455.844	639.266	1.40x	15,251,458	14,022,374
embedding_index_load_gzip	80.959	72.429	0.89x	81.735	72.412	0.89x	15,251,458	14,022,374

ratio > 1 means Lodestar is faster. cpu is the honest one: elapsed time hides work .NET does on background GC threads; CPython is single-threaded. bytes is what the row wrote or read; a results file from before #378 carries none, and the two columns then disappear rather than read zero.