Cargo.toml

[package]

name = "kitoken"
description = "Fast and versatile tokenizer for language models, supporting BPE, Unigram and WordPiece tokenization"
version = "0.0.0"
authors = ["Christian Sdunek <me@systemcluster.me>"]
keywords = ["tokenizer", "nlp", "bpe", "unigram", "wordpiece"]
categories = ["text-processing", "algorithms", "wasm", "no-std"]
license = "BSD-2-Clause"
repository = "https://github.com/Systemcluster/kitoken"
readme = "README.md"
edition = "2021"
resolver = "2"
include = [
    "Cargo.toml",
    "src/**/*",
    "LICENCE",
]


[lib]

name = "kitoken"
path = "src/lib.rs"


[features]

default = [
    "std",
    "serialization",
    "normalization",
    "convert",
    "regex-perf",
]

# Enables all stable features
all = [
    "std",
    "serialization",
    "normalization",
    "convert",
    "split",
    "regex-unicode",
    "regex-perf",
]

# Enables standard library features
std = ["dep:thiserror", "orx-priority-queue/std", "memchr/std", "multiversion?/std"]

# Enables serialization and deserialization
serialization = ["dep:serde", "dep:postcard"]

# Enables all input normalization features
normalization = ["normalization-unicode", "normalization-charsmap"]
# Enables unicode input normalization support
normalization-unicode = ["dep:unicode-normalization"]
# Enables precompiled charsmap input normalization support
normalization-charsmap = ["bstr/unicode"]

# Enables all input split features
split = ["split-unicode-script"]
# Enables input split by unicode scripts
split-unicode-script = ["dep:unicode-script"]

# Enables conversion for all supported tokenizer data formats
convert = ["convert-tiktoken", "convert-sentencepiece", "convert-tokenizers", "convert-tekken"]
# Enables conversion for the HuggingFace Tokenizers format
convert-tokenizers = ["dep:base64", "dep:serde", "dep:serde_json", "hashbrown/serde"]
# Enables conversion for the SentencePiece format
convert-sentencepiece = ["dep:sentencepiece-model"]
# Enables conversion for the OpenAI Tiktoken format
convert-tiktoken = ["dep:base64"]
# Enables conversion for the Mistral Tekken format
convert-tekken = ["dep:base64", "dep:serde", "dep:serde_json"]

# Enables support for additional regex unicode patterns
regex-unicode = ["fancy-regex/unicode"]
# Enables additional regex performance optimizations
regex-perf = ["fancy-regex/perf"]
# Use the oniguruma regex engine instead of fancy-regex
regex-onig = ["dep:onig"]

# Enables the use of multiversion for generating multiple code paths with different CPU feature utilization (requires nightly Rust)
multiversion = ["dep:multiversion"]


[dependencies]

bstr = { version = "1.10", default-features = false, features = ["alloc"] }
log = { version = "0.4" }
fancy-regex = { version = "0.13", default-features = false }
regex-automata = { version = "0.4", default-features = false, features = [
    "alloc",
    "syntax",
    "meta",
    "nfa",
    "dfa",
    "hybrid",
    "unicode-perl",
    "unicode-gencat",
    "unicode-case",
] }
regex-syntax = { version = "0.8", default-features = false, features = [
    "unicode-perl",
    "unicode-gencat",
    "unicode-case",
] }
hashbrown = { version = "0.14", default-features = false, features = ["ahash", "inline-more", "allocator-api2"] }
ahash = { version = "0.8", default-features = false, features = ["const-random"] }
once_cell = { version = "1.19", default-features = false, features = ["alloc", "race"] }
orx-priority-queue = { version = "1.2", default-features = false }
memchr = { version = "2.7", default-features = false, features = ["alloc"] }
derive_more = { version = "1.0", default-features = false, features = ["deref", "deref_mut", "as_ref", "index", "index_mut"] }

# optional dependencies for the std feature
thiserror = { version = "1.0", optional = true }

# optional dependencies for the serialization feature
serde = { version = "1.0", default-features = false, features = ["alloc", "derive"], optional = true }
postcard = { version = "1.0", default-features = false, features = ["alloc"], optional = true }

# optional dependencies for the regex-onig feature
onig = { version = "6.4", default-features = false, optional = true }

# optional dependencies for the normalization-unicode feature
unicode-normalization = { version = "0.1", optional = true }

# optional dependencies for the split-unicode-script feature
unicode-script = { version = "0.5", optional = true }

# optional dependencies for the convert-tiktoken and convert-tokenizers features
base64 = { version = "0.22", default-features = false, features = ["alloc"], optional = true }

# optional dependencies for the convert-sentencepiece feature
sentencepiece-model = { version = "0.1", default-features = false, optional = true }

# optional dependencies for the convert-tokenizers and convert-tekken features
serde_json = { version = "1.0", default-features = false, features = ["alloc"], optional = true }

# optional dependencies for the multiversion feature
multiversion = { version = "0.7", default-features = false, optional = true }

[dev-dependencies]

kitoken = { path = ".", default-features = false, features = [
    "std",
    "serialization",
    "convert",
    "normalization",
    "split",
    "regex-perf",
] }

console = { version = "0.15", features = ["windows-console-colors"] }
simple_logger = { version = "5.0" }
criterion = { version = "0.5", default-features = false, features = ["cargo_bench_support"] }


[package.metadata.docs.rs]

rustdoc-args = ["--cfg", "docsrs"]


[profile.release]

codegen-units = 1
debug = false
debug-assertions = false
incremental = false
lto = "fat"
opt-level = 's'
overflow-checks = false
panic = "abort"
strip = "symbols"

[profile.release.build-override]

opt-level = 's'

[profile.performance]

inherits = "release"
opt-level = 3

[profile.performance.build-override]

opt-level = 3

[profile.bench]

inherits = "release"
opt-level = 3
debug = true
strip = "none"

[profile.bench.build-override]

opt-level = 3


[[bench]]
name = "encode_cl100k"
path = "benches/bench_encode_cl100k.rs"
harness = false
[[bench]]
name = "encode_gpt2"
path = "benches/bench_encode_gpt2.rs"
harness = false
[[bench]]
name = "encode_llama2"
path = "benches/bench_encode_llama2.rs"
harness = false
[[bench]]
name = "encode_xlnet"
path = "benches/bench_encode_xlnet.rs"
harness = false