Skip to content

perf(fast): batch ingestion in Rust with gigatoken's O200k pretokenizer #55

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
AmitMY wants to merge 2 commits into
base: main
Choose a base branch
from
+3,827 −94
Draft

perf(fast): batch ingestion in Rust with gigatoken's O200k pretokenizer #55

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
2bd61b7
perf(fast): batch ingestion in Rust with gigatoken's O200k pretokenizer
AmitMY Jul 23, 2026
5335786
perf(fast): full-scale round — index sets, changed-only registry, sha…
AmitMY Jul 23, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
205 changes: 205 additions & 0 deletions fast/Cargo.lock
View file
Open in desktop

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

2 changes: 2 additions & 0 deletions fast/Cargo.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,3 +14,5 @@ unicode-segmentation = "1.12"
regex = "1"
rayon = "1"
rustc-hash = "2"
# For the vendored gigatoken O200k pretokenizer (see src/pretok/).
icu_properties = { version = "2", features = ["compiled_data"] }
17 changes: 14 additions & 3 deletions fast/python/complex_tokenization_fast/tokenizer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
from functools import reduce

from complex_tokenization_fast._rs import Node, Trainer
from complex_tokenization_fast._rs import Node, Trainer, has_cluster_handlers_py, trainer_from_texts
from complex_tokenization_fast.graphs.settings import GraphSettings
from complex_tokenization_fast.graphs.units import characters, register_script, utf8, utf8_clusters
from complex_tokenization_fast.graphs.words import GPTPretokenizer, words
Expand Down Expand Up @@ -42,8 +42,19 @@ def make_trainer(self, texts):
GraphSettings.ONLY_MINIMAL_MERGES = True
GraphSettings.MAX_MERGE_SIZE = self.merge_size

graphs = self._build_graphs(texts)
trainer = Trainer(graphs=graphs)
# Default configuration ingests entirely in Rust (one boundary
# crossing; gigatoken's O200k pretokenizer is byte-identical to
# GPTPretokenizer). Custom pretokenizers/units/script handlers take
# the per-document Python path.
if (
self.units is utf8_clusters
and self.pretokenizer is GPTPretokenizer
and not has_cluster_handlers_py()
):
trainer = trainer_from_texts(list(texts), connected=self.connected)
else:
graphs = self._build_graphs(texts)
trainer = Trainer(graphs=graphs)

for merge_strs in self.merges:
nodes = tuple(Node(value=s.encode("utf-8")) for s in merge_strs)
Expand Down
Loading
Loading