Chonky

Chonky is a Python library that intelligently segments text into meaningful semantic chunks using a fine-tuned transformer model. This library can be used in the RAG systems.

Installation

pip install chonky

Usage:

from chonky import ParagraphSplitter

# on the first run it will download the transformer model
splitter = ParagraphSplitter(device="cpu")

# Or you can select the model
# splitter = ParagraphSplitter(
#  model_id="mirth/chonky_modernbert_base_1",
#  device="cpu"
# )

text = (
    "Before college the two main things I worked on, outside of school, were writing and programming. "
    "I didn't write essays. I wrote what beginning writers were supposed to write then, and probably still are: short stories. "
    "My stories were awful. They had hardly any plot, just characters with strong feelings, which I imagined made them deep. "
    "The first programs I tried writing were on the IBM 1401 that our school district used for what was then called 'data processing.' "
    "This was in 9th grade, so I was 13 or 14. The school district's 1401 happened to be in the basement of our junior high school, "
    "and my friend Rich Draves and I got permission to use it. It was like a mini Bond villain's lair down there, with all these alien-looking machines — "
    "CPU, disk drives, printer, card reader — sitting up on a raised floor under bright fluorescent lights."
)

for chunk in splitter(text):
  print(chunk)
  print("--")

Sample Output

Before college the two main things I worked on, outside of school, were writing and programming. I didn't write essays. I wrote what beginning writers were supposed to write then, and probably still are: short stories. My stories were awful. They had hardly any plot, just characters with strong feelings, which I imagined made them deep.
--
The first programs I tried writing were on the IBM 1401 that our school district used for what was then called "data processing." This was in 9th grade, so I was 13 or 14. The school district's 1401 happened to be in the basement of our junior high school, and my friend Rich Draves and I got permission to use it.
--
 It was like a mini Bond villain's lair down there, with all these alien-looking machines — CPU, disk drives, printer, card reader — sitting up on a raised floor under bright fluorescent lights.
--

The usage pattern is the following: strip all the markup tags to produce pure text and feed this text into the splitter. For this purpose there is helper class MarkupRemover (it automatically detects the content format):

from chonky.markup_remover import MarkupRemover
from chonky import ParagraphSplitter

remover = MarkupRemover()
splitter = ParagraphSplitter()

text = remover("# Header 1 ...")
splitter(text)

Supported formats: markdown, xml, html.

Supported models

Model ID	Seq Length	Number of Params	Multilingual
mirth/chonky_modernbert_large_1	1024	396M	❌
mirth/chonky_modernbert_base_1	1024	150M	❌
mirth/chonky_mmbert_small_multilingual_1 🆕	1024	140M	✅
mirth/chonky_distilbert_base_uncased_1	512	66.4M	❌

Benchmarks

The following values are token based F1 scores computed on first 1M tokens of each datasets (due to performance reasons).

Various English datasets:

The do_ps fragment for SaT models here is do_paragraph_segmentation flag.

Model	bookcorpus	en_judgements	paul_graham	20_newsgroups
chonkY_modernbert_large_1	0.79 ❗	0.29 ❗	0.69 ❗	0.17
chonkY_modernbert_base_1	0.72	0.08	0.63	0.15
chonkY_mmbert_small_multilingual_1 🆕	0.72	0.2	0.56	0.13
chonkY_distilbert_base_uncased_1	0.69	0.05	0.52	0.15
SaT(sat-12l-sm, do_ps=False)	0.33	0.03	0.43	0.31
SaT(sat-12l-sm, do_ps=True)	0.33	0.06	0.42	0.3
SaT(sat-3l, do_ps=False)	0.28	0.03	0.42	0.34 ❗
SaT(sat-3l, do_ps=True)	0.09	0.07	0.41	0.15
chonkIE SemanticChunker(bge-small-en-v1.5)	0.21	0.01	0.12	0.06
chonkIE SemanticChunker(potion-base-8M)	0.19	0.01	0.15	0.08
chonkIE RecursiveChunker	0.07	0.01	0.05	0.02
langchain SemanticChunker(all-mpnet-base-v2)	0	0	0	0
langchain SemanticChunker(bge-small-en-v1.5)	0	0	0	0
langchain SemanticChunker(potion-base-8M)	0	0	0	0
langchain RecursiveChar	0	0	0	0
llamaindex SemanticSplitter(bge-small-en-v1.5)	0.06	0	0.06	0.02

Project Gutenberg validation:

Model	de	en	es	fr	it	nl	pl	pt	ru	sv	zh
chonky_mmbert_small_multi_1 🆕	0.88 ❗	0.78 ❗	0.91 ❗	0.93 ❗	0.86 ❗	0.81 ❗	0.81 ❗	0.88 ❗	0.97 ❗	0.91 ❗	0.11
chonky_modernbert_large_1	0.53	0.43	0.48	0.51	0.56	0.21	0.65	0.53	0.87	0.51	0.33 ❗
chonky_modernbert_base_1	0.42	0.38	0.34	0.4	0.33	0.22	0.41	0.35	0.27	0.31	0.26
chonky_distilbert_base_uncased_1	0.19	0.3	0.17	0.2	0.18	0.04	0.27	0.21	0.22	0.19	0.15
Number of val tokens	1M	1M	1M	1M	1M	1M	38K	1M	24K	1M	132K