Markov text generation

Train a bigram Markov model on a 1 MiB corpus and emit 10,000 words.

Compile time · median (cold)

median of 10 runs

Native Gocompiled

181 ms172×

Piko interpbytecode VM

1.05 msbaseline

CPython 3.13bytecode VM

354 µs0.34×

PyPy 7.3tracing JIT

314 µs0.30×

Ttengobytecode VM

239 µs0.23×

Sscriggobytecode VM

413 µs0.39×

Mmvmbytecode VM

334 µs0.32×

YyaegiAST walker

374 µs0.36×

Runner	N	Compile	Runtime	P95	Stddev	RSS	vs piko	Status
Native Gocompiled	10	181 ms	5.42 ms	6.07 ms	456 µs	68 MiB	172×	OK
Piko interpbytecode VM	10	1.05 ms	23.7 ms	24.7 ms	355 µs	255 MiB	1.00×	OK
CPython 3.13bytecode VM	10	354 µs	27.2 ms	27.6 ms	496 µs	n/a	0.34×	OK
PyPy 7.3tracing JIT	10	314 µs	9.49 ms	9.77 ms	128 µs	n/a	0.30×	OK
tengobytecode VM	10	239 µs	62.8 ms	72.2 ms	6.63 ms	451 MiB	0.23×	OK
scriggobytecode VM	10	413 µs	56.2 ms	74.4 ms	7.09 ms	428 MiB	0.39×	OK
mvmbytecode VM	10	334 µs	62.7 ms	79.5 ms	6.14 ms	66 MiB	0.32×	OK
yaegiAST walker	10	374 µs	58.3 ms	67.0 ms	9.31 ms	66 MiB	0.36×	OK

Workload & symmetry rules

Workload

Build a bigram model from a 1 MiB lorem corpus. Sample 10,000 tokens from a deterministic LCG-seeded RNG. Print a fingerprint of the output.

Source code