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: 5d7a580b6fa8b8891f7c3ddfba22bf3fe791d7a2
• Previous run: 5d7a580b6fa8b8891f7c3ddfba22bf3fe791d7a2
• 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)
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

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

Method	CorpusSize	Mean	Error	StdDev	Ratio	RatioSD	Gen0	Gen1	Allocated	Alloc Ratio
UnitLoop	1	6.785 μs	0.2345 μs	0.0129 μs	1.00	0.00	0.1068	-	2.76 KB	1.00
EmbedBatch	1	6.485 μs	0.1870 μs	0.0103 μs	0.96	0.00	0.1221	-	3.16 KB	1.14
EmbedBatchBucketed	1	6.745 μs	0.1141 μs	0.0063 μs	0.99	0.00	0.1221	-	3.16 KB	1.14
UnitLoop	8	118.275 μs	4.5115 μs	0.2473 μs	1.00	0.00	5.3711	0.1221	134.29 KB	1.00
EmbedBatch	8	76.678 μs	53.5995 μs	2.9380 μs	0.65	0.02	5.1270	0.2441	127.31 KB	0.95
EmbedBatchBucketed	8	76.655 μs	5.3339 μs	0.2924 μs	0.65	0.00	5.1270	0.2441	127.31 KB	0.95
UnitLoop	32	427.317 μs	45.6494 μs	2.5022 μs	1.00	0.01	19.0430	0.4883	468.71 KB	1.00
EmbedBatch	32	257.695 μs	9.7403 μs	0.5339 μs	0.60	0.00	17.5781	0.9766	441.32 KB	0.94
EmbedBatchBucketed	32	248.229 μs	7.1151 μs	0.3900 μs	0.58	0.00	17.0898	0.9766	427.8 KB	0.91
UnitLoop	128	1,720.020 μs	77.0050 μs	4.2209 μs	1.00	0.00	76.1719	3.9063	1874.78 KB	1.00
EmbedBatch	128	1,041.946 μs	32.2891 μs	1.7699 μs	0.61	0.00	70.3125	9.7656	1764.42 KB	0.94
EmbedBatchBucketed	128	987.384 μs	459.1923 μs	25.1699 μs	0.57	0.01	68.3594	9.7656	1696.9 KB	0.91

Lodestar.Text.Benchmarks.BlockedTableBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

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

Method	length	Mean	Error	StdDev	Allocated
Latin	1000	55.01 μs	0.538 μs	0.029 μs	-
Cjk	1000	61.65 μs	0.593 μs	0.032 μs	-
Latin	10000	5,556.50 μs	149.100 μs	8.173 μs	-
Cjk	10000	8,270.07 μs	1,441.151 μs	78.994 μs	-
Latin	65536	197,960.19 μs	10,062.359 μs	551.552 μs	-

Lodestar.Text.Benchmarks.BpeBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

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

Method	Mean	Error	StdDev	Ratio	Gen0	Allocated	Alloc Ratio
Unigram	601.4 ms	40.09 ms	2.20 ms	1.00	21000.0000	519.51 MB	1.00
Bpe	535.1 ms	18.72 ms	1.03 ms	0.89	4000.0000	112.18 MB	0.22

Lodestar.Text.Benchmarks.BpeScalingBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

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

Method	Length	Mean	Error	StdDev	Gen0	Allocated
BpeOnOnePathologicalToken	512	114.1 μs	15.26 μs	0.84 μs	0.7324	20.38 KB
BpeOnOnePathologicalToken	1024	242.5 μs	13.32 μs	0.73 μs	1.4648	39.93 KB
BpeOnOnePathologicalToken	2048	524.4 μs	14.09 μs	0.77 μs	2.9297	78.98 KB
BpeOnOnePathologicalToken	4096	1,115.9 μs	83.80 μs	4.59 μs	5.8594	157.03 KB

Lodestar.Text.Benchmarks.BucketRouteDiagnostics-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

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

Method	Alphabet	Mean	Error	StdDev	Allocated
DpGroup	cjk	17.170 μs	0.0677 μs	0.0037 μs	-
MyersGroup	cjk	240.261 μs	15.1871 μs	0.8325 μs	-
DpGroup	latin	9.462 μs	0.5874 μs	0.0322 μs	-
MyersGroup	latin	133.365 μs	4.8179 μs	0.2641 μs	-

Lodestar.Text.Benchmarks.FuzzBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

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

Method	Mean	Error	StdDev	Ratio	RatioSD	Gen0	Allocated	Alloc Ratio
Ratio	91.02 ns	8.869 ns	0.486 ns	1.00	0.01	-	-	NA
PartialRatio	12,556.91 ns	6,881.303 ns	377.187 ns	137.97	3.65	-	-	NA
TokenSortRatio	1,068.79 ns	180.002 ns	9.867 ns	11.74	0.11	0.0515	1312 B	NA
TokenSetRatio	3,432.67 ns	282.242 ns	15.471 ns	37.72	0.23	0.2289	5760 B	NA
WRatio	4,704.60 ns	170.477 ns	9.344 ns	51.69	0.25	0.2823	7200 B	NA

Lodestar.Text.Benchmarks.IndelBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

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

Method	Length	Mean	Error	StdDev	Ratio	RatioSD	Allocated	Alloc Ratio
Distance_Utf16	8	27.90 ns	0.173 ns	0.009 ns	1.00	0.00	-	NA
Distance_CodePoint	8	137.14 ns	11.119 ns	0.609 ns	4.91	0.02	-	NA
NormalizedSimilarity_Utf16	8	27.66 ns	2.020 ns	0.111 ns	0.99	0.00	-	NA
SubsequenceLength_Utf16	8	26.10 ns	0.112 ns	0.006 ns	0.94	0.00	-	NA
Distance_Utf16	12	27.89 ns	0.958 ns	0.052 ns	1.00	0.00	-	NA
Distance_CodePoint	12	146.99 ns	3.742 ns	0.205 ns	5.27	0.01	-	NA
NormalizedSimilarity_Utf16	12	29.20 ns	4.830 ns	0.265 ns	1.05	0.01	-	NA
SubsequenceLength_Utf16	12	28.44 ns	0.916 ns	0.050 ns	1.02	0.00	-	NA
Distance_Utf16	16	30.13 ns	0.800 ns	0.044 ns	1.00	0.00	-	NA
Distance_CodePoint	16	155.50 ns	7.559 ns	0.414 ns	5.16	0.01	-	NA
NormalizedSimilarity_Utf16	16	31.75 ns	3.332 ns	0.183 ns	1.05	0.01	-	NA
SubsequenceLength_Utf16	16	28.90 ns	1.827 ns	0.100 ns	0.96	0.00	-	NA
Distance_Utf16	20	32.97 ns	0.683 ns	0.037 ns	1.00	0.00	-	NA
Distance_CodePoint	20	216.15 ns	12.905 ns	0.707 ns	6.56	0.02	-	NA
NormalizedSimilarity_Utf16	20	35.93 ns	1.824 ns	0.100 ns	1.09	0.00	-	NA
SubsequenceLength_Utf16	20	31.52 ns	0.842 ns	0.046 ns	0.96	0.00	-	NA
Distance_Utf16	24	50.31 ns	2.776 ns	0.152 ns	1.00	0.00	-	NA
Distance_CodePoint	24	638.42 ns	68.208 ns	3.739 ns	12.69	0.07	-	NA
NormalizedSimilarity_Utf16	24	59.70 ns	2.532 ns	0.139 ns	1.19	0.00	-	NA
SubsequenceLength_Utf16	24	55.17 ns	2.104 ns	0.115 ns	1.10	0.00	-	NA
Distance_Utf16	32	61.57 ns	2.036 ns	0.112 ns	1.00	0.00	-	NA
Distance_CodePoint	32	786.53 ns	726.713 ns	39.834 ns	12.77	0.56	-	NA
NormalizedSimilarity_Utf16	32	65.39 ns	3.344 ns	0.183 ns	1.06	0.00	-	NA
SubsequenceLength_Utf16	32	57.66 ns	1.510 ns	0.083 ns	0.94	0.00	-	NA
Distance_Utf16	128	812.74 ns	13.243 ns	0.726 ns	1.00	0.00	-	NA
Distance_CodePoint	128	18,691.12 ns	813.769 ns	44.605 ns	23.00	0.05	-	NA
NormalizedSimilarity_Utf16	128	888.45 ns	3.381 ns	0.185 ns	1.09	0.00	-	NA
SubsequenceLength_Utf16	128	881.52 ns	9.358 ns	0.513 ns	1.08	0.00	-	NA
Distance_Utf16	512	7,505.99 ns	110.673 ns	6.066 ns	1.00	0.00	-	NA
Distance_CodePoint	512	323,596.48 ns	40,418.776 ns	2,215.489 ns	43.11	0.26	-	NA
NormalizedSimilarity_Utf16	512	7,477.75 ns	289.251 ns	15.855 ns	1.00	0.00	-	NA
SubsequenceLength_Utf16	512	7,492.70 ns	75.206 ns	4.122 ns	1.00	0.00	-	NA

Lodestar.Text.Benchmarks.LcsGateBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

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

Method	Band	Mean	Error	StdDev	Ratio	RatioSD	Allocated	Alloc Ratio
Dp	8	122.44 ns	1.046 ns	0.057 ns	1.00	0.00	-	NA
Kernel	8	55.85 ns	0.511 ns	0.028 ns	0.46	0.00	-	NA
Dp_Cjk	8	126.88 ns	34.047 ns	1.866 ns	1.04	0.01	-	NA
Kernel_Cjk	8	93.30 ns	2.313 ns	0.127 ns	0.76	0.00	-	NA
Dp	12	210.14 ns	8.159 ns	0.447 ns	1.00	0.00	-	NA
Kernel	12	63.33 ns	1.346 ns	0.074 ns	0.30	0.00	-	NA
Dp_Cjk	12	210.89 ns	18.113 ns	0.993 ns	1.00	0.00	-	NA
Kernel_Cjk	12	110.69 ns	28.597 ns	1.567 ns	0.53	0.01	-	NA
Dp	14	298.07 ns	1.962 ns	0.108 ns	1.00	0.00	-	NA
Kernel	14	67.30 ns	0.756 ns	0.041 ns	0.23	0.00	-	NA
Dp_Cjk	14	267.06 ns	5.632 ns	0.309 ns	0.90	0.00	-	NA
Kernel_Cjk	14	111.09 ns	1.353 ns	0.074 ns	0.37	0.00	-	NA
Dp	16	379.20 ns	2.701 ns	0.148 ns	1.00	0.00	-	NA
Kernel	16	71.16 ns	1.994 ns	0.109 ns	0.19	0.00	-	NA
Dp_Cjk	16	380.45 ns	47.112 ns	2.582 ns	1.00	0.01	-	NA
Kernel_Cjk	16	119.34 ns	22.366 ns	1.226 ns	0.31	0.00	-	NA
Dp	18	462.84 ns	22.217 ns	1.218 ns	1.00	0.00	-	NA
Kernel	18	75.17 ns	1.532 ns	0.084 ns	0.16	0.00	-	NA
Dp_Cjk	18	461.27 ns	3.995 ns	0.219 ns	1.00	0.00	-	NA
Kernel_Cjk	18	122.75 ns	6.551 ns	0.359 ns	0.27	0.00	-	NA
Dp	20	558.71 ns	3.894 ns	0.213 ns	1.00	0.00	-	NA
Kernel	20	78.55 ns	2.524 ns	0.138 ns	0.14	0.00	-	NA
Dp_Cjk	20	558.36 ns	11.941 ns	0.655 ns	1.00	0.00	-	NA
Kernel_Cjk	20	128.72 ns	2.487 ns	0.136 ns	0.23	0.00	-	NA
Dp	24	740.12 ns	161.509 ns	8.853 ns	1.00	0.01	-	NA
Kernel	24	86.94 ns	0.654 ns	0.036 ns	0.12	0.00	-	NA
Dp_Cjk	24	799.58 ns	699.153 ns	38.323 ns	1.08	0.05	-	NA
Kernel_Cjk	24	143.10 ns	6.612 ns	0.362 ns	0.19	0.00	-	NA
Dp	32	1,193.21 ns	46.273 ns	2.536 ns	1.00	0.00	-	NA
Kernel	32	103.55 ns	0.137 ns	0.008 ns	0.09	0.00	-	NA
Dp_Cjk	32	1,213.40 ns	524.337 ns	28.741 ns	1.02	0.02	-	NA
Kernel_Cjk	32	167.81 ns	21.522 ns	1.180 ns	0.14	0.00	-	NA
Dp	48	2,794.28 ns	9.694 ns	0.531 ns	1.00	0.00	-	NA
Kernel	48	125.46 ns	3.115 ns	0.171 ns	0.04	0.00	-	NA
Dp_Cjk	48	2,794.16 ns	102.148 ns	5.599 ns	1.00	0.00	-	NA
Kernel_Cjk	48	236.90 ns	2.389 ns	0.131 ns	0.08	0.00	-	NA
Dp	64	4,880.58 ns	7,458.806 ns	408.842 ns	1.00	0.11	-	NA
Kernel	64	153.00 ns	2.188 ns	0.120 ns	0.03	0.00	-	NA
Dp_Cjk	64	5,010.41 ns	2,493.261 ns	136.664 ns	1.03	0.08	-	NA
Kernel_Cjk	64	286.52 ns	18.506 ns	1.014 ns	0.06	0.00	-	NA
Dp	96	11,067.26 ns	4,159.950 ns	228.021 ns	1.00	0.03	-	NA
Kernel	96	682.98 ns	34.199 ns	1.875 ns	0.06	0.00	-	NA
Dp_Cjk	96	10,080.22 ns	1,810.605 ns	99.245 ns	0.91	0.02	-	NA
Kernel_Cjk	96	985.35 ns	2.356 ns	0.129 ns	0.09	0.00	-	NA

Lodestar.Text.Benchmarks.LevenshteinBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

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

Method	Length	Mean	Error	StdDev	Ratio	RatioSD	Allocated	Alloc Ratio
Distance_Utf16	8	26.93 ns	0.732 ns	0.040 ns	1.00	0.00	-	NA
Distance_CodePoint	8	137.34 ns	10.263 ns	0.563 ns	5.10	0.02	-	NA
NormalizedSimilarity_Utf16	8	25.74 ns	1.289 ns	0.071 ns	0.96	0.00	-	NA
Distance_Utf16	64	263.86 ns	11.486 ns	0.630 ns	1.00	0.00	-	NA
Distance_CodePoint	64	717.67 ns	12.554 ns	0.688 ns	2.72	0.01	-	NA
NormalizedSimilarity_Utf16	64	269.50 ns	31.653 ns	1.735 ns	1.02	0.01	-	NA
Distance_Utf16	512	15,527.28 ns	1,929.034 ns	105.737 ns	1.00	0.01	-	NA
Distance_CodePoint	512	18,165.53 ns	369.510 ns	20.254 ns	1.17	0.01	-	NA
NormalizedSimilarity_Utf16	512	15,139.77 ns	464.794 ns	25.477 ns	0.98	0.01	-	NA

Lodestar.Text.Benchmarks.LevenshteinCodePointBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

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

Method	Length	Distinct	Mean	Error	StdDev	Ratio	RatioSD	Allocated	Alloc Ratio
Distance_CodePoint	16	32	363.7 ns	9.02 ns	0.49 ns	1.00	0.00	-	NA
Distance_Utf16	16	32	221.7 ns	4.37 ns	0.24 ns	0.61	0.00	-	NA
Distance_CodePoint	16	512	355.1 ns	6.77 ns	0.37 ns	1.00	0.00	-	NA
Distance_Utf16	16	512	223.9 ns	2.10 ns	0.11 ns	0.63	0.00	-	NA
Distance_CodePoint	24	32	465.9 ns	2.13 ns	0.12 ns	1.00	0.00	-	NA
Distance_Utf16	24	32	306.5 ns	2.34 ns	0.13 ns	0.66	0.00	-	NA
Distance_CodePoint	24	512	453.4 ns	33.37 ns	1.83 ns	1.00	0.00	-	NA
Distance_Utf16	24	512	313.5 ns	6.73 ns	0.37 ns	0.69	0.00	-	NA
Distance_CodePoint	32	32	542.8 ns	10.42 ns	0.57 ns	1.00	0.00	-	NA
Distance_Utf16	32	32	411.3 ns	2.68 ns	0.15 ns	0.76	0.00	-	NA
Distance_CodePoint	32	512	550.5 ns	9.28 ns	0.51 ns	1.00	0.00	-	NA
Distance_Utf16	32	512	397.3 ns	17.25 ns	0.95 ns	0.72	0.00	-	NA
Distance_CodePoint	40	32	633.8 ns	20.12 ns	1.10 ns	1.00	0.00	-	NA
Distance_Utf16	40	32	1,383.6 ns	14.61 ns	0.80 ns	2.18	0.00	-	NA
Distance_CodePoint	40	512	659.2 ns	1.69 ns	0.09 ns	1.00	0.00	-	NA
Distance_Utf16	40	512	1,324.3 ns	141.98 ns	7.78 ns	2.01	0.01	-	NA
Distance_CodePoint	128	32	2,742.3 ns	45.41 ns	2.49 ns	1.00	0.00	-	NA
Distance_Utf16	128	32	5,561.9 ns	150.82 ns	8.27 ns	2.03	0.00	-	NA
Distance_CodePoint	128	512	2,809.6 ns	35.30 ns	1.93 ns	1.00	0.00	-	NA
Distance_Utf16	128	512	5,557.6 ns	277.61 ns	15.22 ns	1.98	0.00	-	NA
Distance_CodePoint	512	32	19,919.4 ns	738.02 ns	40.45 ns	1.00	0.00	-	NA
Distance_Utf16	512	32	68,020.9 ns	14,726.80 ns	807.23 ns	3.41	0.04	-	NA
Distance_CodePoint	512	512	507,750.7 ns	12,811.42 ns	702.24 ns	1.00	0.00	-	NA
Distance_Utf16	512	512	62,553.0 ns	105.90 ns	5.80 ns	0.12	0.00	-	NA

Lodestar.Text.Benchmarks.MetricsBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

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

Method	Samples	Classes	Mean	Error	StdDev	Gen0	Allocated
Matrix	1000	2	6.635 μs	0.1985 μs	0.0109 μs	0.0076	312 B
MatrixWeighted	1000	2	6.455 μs	0.2572 μs	0.0141 μs	0.0076	312 B
AccuracyScore	1000	2	1.133 μs	0.0045 μs	0.0002 μs	-	-
F1Macro	1000	2	6.567 μs	0.2006 μs	0.0110 μs	0.0153	472 B
Report	1000	2	9.555 μs	0.0922 μs	0.0051 μs	0.2594	6520 B
Matrix	1000	10	6.803 μs	0.0796 μs	0.0044 μs	0.0458	1248 B
MatrixWeighted	1000	10	6.620 μs	0.3001 μs	0.0164 μs	0.0458	1248 B
AccuracyScore	1000	10	1.133 μs	0.0341 μs	0.0019 μs	-	-
F1Macro	1000	10	6.821 μs	0.1305 μs	0.0072 μs	0.0610	1664 B
Report	1000	10	14.396 μs	0.2260 μs	0.0124 μs	0.6104	15496 B
Matrix	100000	2	760.847 μs	36.1390 μs	1.9809 μs	-	313 B
MatrixWeighted	100000	2	728.818 μs	21.4795 μs	1.1774 μs	-	313 B
AccuracyScore	100000	2	170.353 μs	25.3851 μs	1.3914 μs	-	-
F1Macro	100000	2	739.163 μs	37.6811 μs	2.0654 μs	-	473 B
Report	100000	2	771.007 μs	91.1608 μs	4.9968 μs	-	6545 B
Matrix	100000	10	889.305 μs	37.0550 μs	2.0311 μs	-	1249 B
MatrixWeighted	100000	10	885.894 μs	44.8458 μs	2.4582 μs	-	1249 B
AccuracyScore	100000	10	274.694 μs	10.6281 μs	0.5826 μs	-	-
F1Macro	100000	10	881.760 μs	30.6264 μs	1.6787 μs	-	1665 B
Report	100000	10	873.074 μs	25.4110 μs	1.3929 μs	-	15841 B
Matrix	1000000	2	7,345.797 μs	281.1026 μs	15.4082 μs	-	318 B
MatrixWeighted	1000000	2	7,577.554 μs	478.6193 μs	26.2347 μs	-	318 B
AccuracyScore	1000000	2	1,973.377 μs	72.4291 μs	3.9701 μs	-	-
F1Macro	1000000	2	7,724.142 μs	249.1775 μs	13.6583 μs	-	478 B
Report	1000000	2	7,524.268 μs	153.5010 μs	8.4139 μs	-	6572 B
Matrix	1000000	10	8,792.776 μs	293.8876 μs	16.1090 μs	-	1260 B
MatrixWeighted	1000000	10	8,719.573 μs	389.7039 μs	21.3610 μs	-	1260 B
AccuracyScore	1000000	10	2,996.003 μs	60.8285 μs	3.3342 μs	-	-
F1Macro	1000000	10	9,144.508 μs	656.9726 μs	36.0109 μs	-	1676 B
Report	1000000	10	9,313.685 μs	198.9309 μs	10.9041 μs	-	15892 B

Lodestar.Text.Benchmarks.MyersGateBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

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

Method	Band	Mean	Error	StdDev	Ratio	RatioSD	Allocated	Alloc Ratio
Dp	4	78.67 ns	1.035 ns	0.057 ns	1.00	0.00	-	NA
Kernel	4	75.93 ns	3.649 ns	0.200 ns	0.97	0.00	-	NA
Dp_Cjk	4	76.73 ns	0.405 ns	0.022 ns	0.98	0.00	-	NA
Kernel_Cjk	4	75.69 ns	2.904 ns	0.159 ns	0.96	0.00	-	NA
Dp	6	105.78 ns	1.510 ns	0.083 ns	1.00	0.00	-	NA
Kernel	6	76.71 ns	0.478 ns	0.026 ns	0.73	0.00	-	NA
Dp_Cjk	6	105.44 ns	1.396 ns	0.077 ns	1.00	0.00	-	NA
Kernel_Cjk	6	128.27 ns	0.892 ns	0.049 ns	1.21	0.00	-	NA
Dp	8	148.73 ns	11.171 ns	0.612 ns	1.00	0.01	-	NA
Kernel	8	85.72 ns	1.868 ns	0.102 ns	0.58	0.00	-	NA
Dp_Cjk	8	149.89 ns	0.243 ns	0.013 ns	1.01	0.00	-	NA
Kernel_Cjk	8	179.12 ns	17.290 ns	0.948 ns	1.20	0.01	-	NA
Dp	10	196.20 ns	5.054 ns	0.277 ns	1.00	0.00	-	NA
Kernel	10	93.69 ns	1.333 ns	0.073 ns	0.48	0.00	-	NA
Dp_Cjk	10	198.02 ns	3.495 ns	0.192 ns	1.01	0.00	-	NA
Kernel_Cjk	10	138.30 ns	2.517 ns	0.138 ns	0.70	0.00	-	NA
Dp	12	269.71 ns	10.151 ns	0.556 ns	1.00	0.00	-	NA
Kernel	12	104.59 ns	28.954 ns	1.587 ns	0.39	0.01	-	NA
Dp_Cjk	12	268.74 ns	1.760 ns	0.096 ns	1.00	0.00	-	NA
Kernel_Cjk	12	150.53 ns	11.847 ns	0.649 ns	0.56	0.00	-	NA
Dp	16	412.21 ns	5.332 ns	0.292 ns	1.00	0.00	-	NA
Kernel	16	119.33 ns	0.604 ns	0.033 ns	0.29	0.00	-	NA
Dp_Cjk	16	412.19 ns	2.669 ns	0.146 ns	1.00	0.00	-	NA
Kernel_Cjk	16	168.94 ns	9.077 ns	0.498 ns	0.41	0.00	-	NA
Dp	24	810.02 ns	17.839 ns	0.978 ns	1.00	0.00	-	NA
Kernel	24	155.44 ns	20.922 ns	1.147 ns	0.19	0.00	-	NA
Dp_Cjk	24	820.97 ns	302.173 ns	16.563 ns	1.01	0.02	-	NA
Kernel_Cjk	24	216.07 ns	5.992 ns	0.328 ns	0.27	0.00	-	NA
Dp	32	1,419.52 ns	8.162 ns	0.447 ns	1.00	0.00	-	NA
Kernel	32	192.19 ns	1.759 ns	0.096 ns	0.14	0.00	-	NA
Dp_Cjk	32	1,425.37 ns	35.521 ns	1.947 ns	1.00	0.00	-	NA
Kernel_Cjk	32	257.55 ns	6.287 ns	0.345 ns	0.18	0.00	-	NA
Dp	48	3,118.79 ns	172.915 ns	9.478 ns	1.00	0.00	-	NA
Kernel	48	261.39 ns	3.823 ns	0.210 ns	0.08	0.00	-	NA
Dp_Cjk	48	3,109.40 ns	46.539 ns	2.551 ns	1.00	0.00	-	NA
Kernel_Cjk	48	339.61 ns	1.996 ns	0.109 ns	0.11	0.00	-	NA
Dp	64	5,470.97 ns	145.119 ns	7.954 ns	1.00	0.00	-	NA
Kernel	64	332.09 ns	10.999 ns	0.603 ns	0.06	0.00	-	NA
Dp_Cjk	64	5,460.83 ns	46.868 ns	2.569 ns	1.00	0.00	-	NA
Kernel_Cjk	64	435.38 ns	3.491 ns	0.191 ns	0.08	0.00	-	NA
Dp	96	12,212.16 ns	276.297 ns	15.145 ns	1.00	0.00	-	NA
Kernel	96	1,310.41 ns	16.653 ns	0.913 ns	0.11	0.00	-	NA
Dp_Cjk	96	12,237.11 ns	427.949 ns	23.457 ns	1.00	0.00	-	NA
Kernel_Cjk	96	1,699.20 ns	43.354 ns	2.376 ns	0.14	0.00	-	NA

Lodestar.Text.Benchmarks.PersistenceBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

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

Method	Mean	Error	StdDev	Gen0	Gen1	Gen2	Allocated
VocabTxt	3.842 ms	0.0966 ms	0.0053 ms	85.9375	78.1250	31.2500	3.62 MB
TokenizerJsonWordPiece	11.472 ms	2.9466 ms	0.1615 ms	125.0000	109.3750	31.2500	5.72 MB
TokenizerJsonUnigram	11.884 ms	0.3498 ms	0.0192 ms	78.1250	62.5000	31.2500	4.64 MB
SpieceModel	4.073 ms	0.9082 ms	0.0498 ms	78.1250	70.3125	23.4375	3.36 MB
TfidfSave	1.944 ms	0.4206 ms	0.0231 ms	33.2031	29.2969	29.2969	2.09 MB
TfidfLoad	4.300 ms	0.3515 ms	0.0193 ms	62.5000	54.6875	23.4375	2.86 MB
EmbeddingIndexSave	5.517 ms	0.6180 ms	0.0339 ms	273.4375	273.4375	273.4375	19.87 MB
EmbeddingIndexLoad	10.096 ms	5.1618 ms	0.2829 ms	484.3750	453.1250	453.1250	35.35 MB

Lodestar.Text.Benchmarks.StopWordBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

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

Method	Documents	Mean	Error	StdDev	Ratio	Gen0	Gen1	Gen2	Allocated	Alloc Ratio
Count	200	7.700 ms	0.4437 ms	0.0243 ms	1.00	343.7500	156.2500	62.5000	7.92 MB	1.00
CountWithStopWords	200	6.297 ms	0.5794 ms	0.0318 ms	0.82	257.8125	125.0000	-	6.34 MB	0.80
Hashing	200	7.852 ms	1.3158 ms	0.0721 ms	1.02	359.3750	140.6250	62.5000	8.08 MB	1.02
HashingWithStopWords	200	6.528 ms	0.0510 ms	0.0028 ms	0.85	265.6250	93.7500	-	6.52 MB	0.82
Count	1000	31.984 ms	0.9332 ms	0.0512 ms	1.00	1687.5000	812.5000	562.5000	38.64 MB	1.00
CountWithStopWords	1000	25.089 ms	0.7729 ms	0.0424 ms	0.78	1406.2500	562.5000	250.0000	30.99 MB	0.80
Hashing	1000	30.456 ms	2.7674 ms	0.1517 ms	0.95	1968.7500	593.7500	593.7500	39.55 MB	1.02
HashingWithStopWords	1000	25.722 ms	0.6029 ms	0.0330 ms	0.80	1343.7500	343.7500	250.0000	31.83 MB	0.82

Lodestar.Text.Benchmarks.VectorMathBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

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

Method	Dim	Mean	Error	StdDev	Ratio	Allocated	Alloc Ratio
Dot	384	54.97 ns	0.270 ns	0.015 ns	1.00	-	NA
L2Norm	384	52.03 ns	0.859 ns	0.047 ns	0.95	-	NA
Dot	768	103.43 ns	1.053 ns	0.058 ns	1.00	-	NA
L2Norm	768	107.37 ns	1.088 ns	0.060 ns	1.04	-	NA
Dot	1024	139.75 ns	2.932 ns	0.161 ns	1.00	-	NA
L2Norm	1024	136.69 ns	8.288 ns	0.454 ns	0.98	-	NA

Lodestar.Text.Benchmarks.VectorizerBenchmarks-report-github

BenchmarkDotNet v0.14.0, Ubuntu 24.04.4 LTS (Noble Numbat)
Intel Xeon Platinum 8370C CPU 2.80GHz, 1 CPU, 4 logical and 2 physical cores
.NET SDK 10.0.400
  [Host]   : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI
  ShortRun : .NET 10.0.11 (10.0.1126.37416), X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

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

Method	Documents	Mean	Error	StdDev	Ratio	RatioSD	Gen0	Gen1	Gen2	Allocated	Alloc Ratio
Count	200	3.022 ms	4.5751 ms	0.2508 ms	1.00	0.10	62.5000	23.4375	-	1.6 MB	1.00
Tfidf	200	3.246 ms	5.3065 ms	0.2909 ms	1.08	0.11	62.5000	23.4375	-	1.63 MB	1.02
CountBigrams	200	3.786 ms	0.1958 ms	0.0107 ms	1.26	0.09	109.3750	62.5000	-	2.78 MB	1.74
Hashing	200	2.974 ms	2.0463 ms	0.1122 ms	0.99	0.08	62.5000	23.4375	-	1.6 MB	1.00
Count	1000	7.167 ms	0.7291 ms	0.0400 ms	1.00	0.01	351.5625	210.9375	70.3125	7.83 MB	1.00
Tfidf	1000	7.453 ms	1.3872 ms	0.0760 ms	1.04	0.01	343.7500	218.7500	93.7500	7.97 MB	1.02
CountBigrams	1000	12.086 ms	1.1205 ms	0.0614 ms	1.69	0.01	640.6250	265.6250	265.6250	13.42 MB	1.71
Hashing	1000	7.209 ms	0.1759 ms	0.0096 ms	1.01	0.00	351.5625	140.6250	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	102.4	21.8	4.69x C# faster
latin	32	147.9	86.7	1.71x C# faster
latin	128	467.2	801.7	1.72x Py faster
latin	512	4731.8	7440.6	1.57x Py faster
cjk	8	113.3	21.8	5.18x C# faster
cjk	32	263.1	214.2	1.23x C# faster
cjk	128	1792.0	1585.3	1.13x C# faster
cjk	512	16300.6	10808.4	1.51x 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	131.6	19.2	6.86x C# faster
latin	32	229.6	154.9	1.48x C# faster
latin	128	1809.0	1543.2	1.17x C# faster
latin	512	15214.5	16134.8	1.06x Py faster
cjk	8	130.9	19.2	6.83x C# faster
cjk	32	339.1	290.0	1.17x C# faster
cjk	128	2978.7	2531.9	1.18x C# faster
cjk	512	25920.2	19808.3	1.31x 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	0.007	0.811	111.52x	0.007	0.811	111.51x
accuracy_n1000_k2	0.001	0.425	372.56x	0.001	0.425	372.52x
precision_recall_f1_macro_n1000_k2	0.007	1.472	220.87x	0.007	1.472	220.88x
classification_report_n1000_k2	0.010	5.619	582.96x	0.010	5.619	582.93x
roc_auc_binary_n1000_k2	0.015	1.627	107.49x	0.015	1.626	107.48x
balanced_accuracy_n1000_k2	0.007	0.871	133.02x	0.007	0.871	133.02x
matthews_n1000_k2	0.007	1.646	250.37x	0.007	1.646	250.36x
cohen_kappa_n1000_k2	0.007	0.905	137.57x	0.007	0.905	137.59x
mse_n1000_k2	0.007	0.232	32.96x	0.007	0.232	32.95x
mae_n1000_k2	0.007	0.231	32.76x	0.007	0.231	32.76x
median_ae_n1000_k2	0.007	0.246	36.13x	0.007	0.246	36.13x
r2_n1000_k2	0.003	0.281	82.21x	0.003	0.281	82.20x
confusion_matrix_n1000_k10	0.007	0.804	109.46x	0.007	0.804	109.47x
accuracy_n1000_k10	0.001	0.428	375.17x	0.001	0.428	375.13x
precision_recall_f1_macro_n1000_k10	0.007	1.491	207.73x	0.007	1.491	207.72x
classification_report_n1000_k10	0.015	5.804	396.09x	0.015	5.804	396.09x
roc_auc_ovr_macro_n1000_k10	0.521	8.316	15.96x	0.521	8.316	15.96x
balanced_accuracy_n1000_k10	0.007	0.870	124.71x	0.007	0.870	124.71x
matthews_n1000_k10	0.007	1.674	240.07x	0.007	1.674	240.05x
cohen_kappa_n1000_k10	0.007	0.914	123.99x	0.007	0.914	123.99x
mse_n1000_k10	0.007	0.231	32.84x	0.007	0.231	32.84x
mae_n1000_k10	0.007	0.231	32.71x	0.007	0.230	32.71x
median_ae_n1000_k10	0.007	0.243	35.99x	0.007	0.243	35.99x
r2_n1000_k10	0.003	0.280	82.09x	0.003	0.280	82.10x
confusion_matrix_n100000_k2	0.829	10.913	13.16x	0.829	10.912	13.16x
accuracy_n100000_k2	0.180	3.764	20.88x	0.180	3.763	20.88x
precision_recall_f1_macro_n100000_k2	0.744	12.380	16.65x	0.744	12.380	16.65x
classification_report_n100000_k2	0.753	26.734	35.49x	0.753	26.734	35.49x
roc_auc_binary_n100000_k2	4.079	25.349	6.22x	4.079	25.349	6.21x
balanced_accuracy_n100000_k2	0.746	10.964	14.69x	0.746	10.964	14.69x
matthews_n100000_k2	0.753	21.924	29.13x	0.753	21.923	29.13x
cohen_kappa_n100000_k2	0.745	10.989	14.74x	0.745	10.989	14.74x
mse_n100000_k2	0.281	0.514	1.83x	0.281	0.514	1.83x
mae_n100000_k2	0.268	0.505	1.88x	0.268	0.505	1.88x
median_ae_n100000_k2	0.761	1.669	2.19x	0.777	1.669	2.15x
r2_n100000_k2	0.336	0.854	2.54x	0.336	0.854	2.54x
confusion_matrix_n100000_k10	0.903	10.863	12.03x	0.903	10.863	12.03x
accuracy_n100000_k10	0.276	3.764	13.62x	0.276	3.764	13.62x
precision_recall_f1_macro_n100000_k10	0.862	13.020	15.11x	0.862	13.019	15.11x
classification_report_n100000_k10	0.867	29.265	33.75x	0.867	29.263	33.76x
roc_auc_ovr_macro_n100000_k10	45.793	205.712	4.49x	45.787	205.694	4.49x
balanced_accuracy_n100000_k10	0.855	10.969	12.82x	0.855	10.969	12.82x
matthews_n100000_k10	0.853	22.536	26.41x	0.853	22.536	26.41x
cohen_kappa_n100000_k10	0.984	11.003	11.18x	0.984	11.003	11.19x
mse_n100000_k10	0.281	0.514	1.83x	0.281	0.514	1.83x
mae_n100000_k10	0.268	0.506	1.89x	0.268	0.506	1.89x
median_ae_n100000_k10	0.821	1.669	2.03x	0.887	1.669	1.88x
r2_n100000_k10	0.336	0.852	2.53x	0.336	0.851	2.53x
confusion_matrix_n1000000_k2	7.833	104.503	13.34x	7.833	104.488	13.34x
accuracy_n1000000_k2	1.930	34.402	17.82x	1.930	34.402	17.82x
precision_recall_f1_macro_n1000000_k2	7.536	112.327	14.91x	7.536	112.315	14.90x
classification_report_n1000000_k2	7.453	223.095	29.93x	7.452	223.065	29.93x
roc_auc_binary_n1000000_k2	63.750	288.226	4.52x	63.749	288.178	4.52x
balanced_accuracy_n1000000_k2	7.446	104.649	14.05x	7.446	104.640	14.05x
matthews_n1000000_k2	7.444	210.737	28.31x	7.444	210.727	28.31x
cohen_kappa_n1000000_k2	7.458	104.812	14.05x	7.458	104.810	14.05x
mse_n1000000_k2	2.626	3.401	1.30x	2.625	3.401	1.30x
mae_n1000000_k2	2.667	3.095	1.16x	2.667	3.095	1.16x
median_ae_n1000000_k2	7.506	14.364	1.91x	7.543	14.362	1.90x
r2_n1000000_k2	3.294	6.400	1.94x	3.293	6.399	1.94x
confusion_matrix_n1000000_k10	9.067	104.381	11.51x	9.066	104.343	11.51x
accuracy_n1000000_k10	2.990	34.358	11.49x	2.990	34.353	11.49x
precision_recall_f1_macro_n1000000_k10	8.682	118.604	13.66x	8.682	118.596	13.66x
classification_report_n1000000_k10	8.600	250.319	29.11x	8.599	250.317	29.11x
balanced_accuracy_n1000000_k10	8.567	104.559	12.20x	8.567	104.556	12.20x
matthews_n1000000_k10	8.554	216.820	25.35x	8.554	216.792	25.34x
cohen_kappa_n1000000_k10	8.554	104.742	12.25x	8.554	104.737	12.24x
mse_n1000000_k10	2.832	3.515	1.24x	2.832	3.515	1.24x
mae_n1000000_k10	2.813	3.047	1.08x	2.813	3.047	1.08x
median_ae_n1000000_k10	7.828	14.728	1.88x	7.945	14.727	1.85x
r2_n1000000_k10	3.859	6.416	1.66x	3.858	6.416	1.66x

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.

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	3.976	9.892	2.49x	4.283	9.891	2.31x	228,891	228,891
tokenizer_json_wordpiece	11.403	15.477	1.36x	11.741	15.477	1.32x	706,526	706,526
tokenizer_json_unigram	11.678	41.488	3.55x	12.060	41.483	3.44x	1,990,038	1,990,038
spiece_model	4.646	28.771	6.19x	4.945	28.770	5.82x	533,084	533,084
tfidf_save	1.860	2.496	1.34x	1.882	2.496	1.33x	581,787	591,922
tfidf_load	4.359	3.890	0.89x	4.607	3.890	0.84x	581,787	591,922
embedding_index_save	5.254	2.527	0.48x	5.452	2.527	0.46x	20,589,007	15,360,128
embedding_index_save_file	50.362	36.496	0.72x	10.704	5.280	0.49x	20,589,007	15,360,128
embedding_index_load	9.543	1.631	0.17x	10.291	1.631	0.16x	20,589,007	15,360,128
embedding_index_load_file	10.366	1.148	0.11x	11.168	1.142	0.10x	20,589,007	15,360,128
embedding_index_load_memory	5.599	1.662	0.30x	5.991	1.662	0.28x	20,589,007	15,360,128
embedding_index_view_floor	0.000	0.001	115.06x	0.000	0.001	115.07x	20,589,007	15,360,128
embedding_index_save_gzip	424.529	562.664	1.33x	425.739	562.652	1.32x	15,250,490	14,022,374
embedding_index_load_gzip	83.508	71.661	0.86x	84.952	71.661	0.84x	15,250,490	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.