New Text Splitters (#817)
This commit is contained in:
Gabriel Luiz Freitas Almeida
GitHub
commit
f19e745b77
9 changed files with 207 additions and 17 deletions
|
|
@ -1,11 +1,13 @@
|
|||
import Admonition from '@theme/Admonition';
|
||||
import Admonition from "@theme/Admonition";
|
||||
|
||||
# Text Splitters
|
||||
|
||||
<Admonition type="caution" icon="🚧" title="ZONE UNDER CONSTRUCTION">
|
||||
<p>
|
||||
We appreciate your understanding as we polish our documentation – it may contain some rough edges. Share your feedback or report issues to help us improve! 🛠️📝
|
||||
</p>
|
||||
<p>
|
||||
We appreciate your understanding as we polish our documentation – it may
|
||||
contain some rough edges. Share your feedback or report issues to help us
|
||||
improve! 🛠️📝
|
||||
</p>
|
||||
</Admonition>
|
||||
|
||||
A text splitter is a tool that divides a document or text into smaller chunks or segments. It is used to break down large texts into more manageable pieces for analysis or processing.
|
||||
|
|
@ -22,13 +24,13 @@ The `CharacterTextSplitter` is used to split a long text into smaller chunks bas
|
|||
|
||||
- **chunk_overlap:** Determines the number of characters that overlap between consecutive chunks when splitting text. It specifies how much of the previous chunk should be included in the next chunk.
|
||||
|
||||
For example, if the `chunk_overlap` is set to 20 and the `chunk_size` is set to 100, the splitter will create chunks of 100 characters each, but the last 20 characters of each chunk will overlap with the first 20 characters of the next chunk. This allows for a smoother transition between chunks and ensures that no information is lost – defaults to `200`.
|
||||
For example, if the `chunk_overlap` is set to 20 and the `chunk_size` is set to 100, the splitter will create chunks of 100 characters each, but the last 20 characters of each chunk will overlap with the first 20 characters of the next chunk. This allows for a smoother transition between chunks and ensures that no information is lost – defaults to `200`.
|
||||
|
||||
- **chunk_size:** Determines the maximum number of characters in each chunk when splitting a text. It specifies the size or length of each chunk.
|
||||
|
||||
For example, if the chunk_size is set to 100, the splitter will create chunks of 100 characters each. If the text is longer than 100 characters, it will be divided into multiple chunks of equal size, except for the last chunk, which may be smaller if there are remaining characters –defaults to `1000`.
|
||||
For example, if the chunk_size is set to 100, the splitter will create chunks of 100 characters each. If the text is longer than 100 characters, it will be divided into multiple chunks of equal size, except for the last chunk, which may be smaller if there are remaining characters –defaults to `1000`.
|
||||
|
||||
- **separator:** Specifies the character that will be used to split the text into chunks – defaults to `.`
|
||||
- **separator:** Specifies the character that will be used to split the text into chunks – defaults to `.`
|
||||
|
||||
---
|
||||
|
||||
|
|
@ -44,6 +46,18 @@ The `RecursiveCharacterTextSplitter` splits the text by trying to keep paragra
|
|||
|
||||
- **chunk_size:** Determines the maximum number of characters in each chunk when splitting a text. It specifies the size or length of each chunk.
|
||||
|
||||
- **separator_type:** The parameter allows the user to split the code with multiple language support. It supports various languages such as Text, Ruby, Python, Solidity, Java, and more. Defaults to `Text`.
|
||||
- **separators:** The `separators` in RecursiveCharacterTextSplitter are the characters used to split the text into chunks. The text splitter tries to create chunks based on splitting on the first character in the list of `separators`. If any chunks are too large, it moves on to the next character in the list and continues splitting. Defaults to ["\n\n", "\n", " ", ""].
|
||||
|
||||
- **separators:** The `separators` in RecursiveCharacterTextSplitter are the characters used to split the text into chunks. The text splitter tries to create chunks based on splitting on the first character in the list of `separators`. If any chunks are too large, it moves on to the next character in the list and continues splitting. Defaults to `.`
|
||||
### LanguageRecursiveTextSplitter
|
||||
|
||||
The `LanguageRecursiveTextSplitter` is a text splitter that splits the text into smaller chunks based on the (programming) language of the text.
|
||||
|
||||
**Params**
|
||||
|
||||
- **Documents:** Input documents to split.
|
||||
|
||||
- **chunk_overlap:** Determines the number of characters that overlap between consecutive chunks when splitting text. It specifies how much of the previous chunk should be included in the next chunk.
|
||||
|
||||
- **chunk_size:** Determines the maximum number of characters in each chunk when splitting a text. It specifies the size or length of each chunk.
|
||||
|
||||
- **separator_type:** The parameter allows the user to split the code with multiple language support. It supports various languages such as Ruby, Python, Solidity, Java, and more. Defaults to `Python`.
|
||||
|
|
|
|||
4
poetry.lock
generated
4
poetry.lock
generated
|
|
@ -4195,7 +4195,7 @@ files = [
|
|||
name = "pillow"
|
||||
version = "10.0.0"
|
||||
description = "Python Imaging Library (Fork)"
|
||||
optional = true
|
||||
optional = false
|
||||
python-versions = ">=3.8"
|
||||
files = [
|
||||
{file = "Pillow-10.0.0-cp310-cp310-macosx_10_10_x86_64.whl", hash = "sha256:1f62406a884ae75fb2f818694469519fb685cc7eaff05d3451a9ebe55c646891"},
|
||||
|
|
@ -7467,4 +7467,4 @@ local = ["ctransformers", "llama-cpp-python", "sentence-transformers"]
|
|||
[metadata]
|
||||
lock-version = "2.0"
|
||||
python-versions = ">=3.9,<3.11"
|
||||
content-hash = "1effd07e35ba89cc3971f027218032e24e7816d93bccb7fd6470cc56acc04418"
|
||||
content-hash = "36e1f79f4e6d2e55b652d10e43ccde639714ffff2965fa52b466bd854259ebf6"
|
||||
|
|
|
|||
|
|
@ -1,6 +1,6 @@
|
|||
[tool.poetry]
|
||||
name = "langflow"
|
||||
version = "0.4.15"
|
||||
version = "0.4.16"
|
||||
description = "A Python package with a built-in web application"
|
||||
authors = ["Logspace <contact@logspace.ai>"]
|
||||
maintainers = [
|
||||
|
|
@ -79,6 +79,7 @@ psycopg-binary = "^3.1.9"
|
|||
fastavro = "^1.8.0"
|
||||
langchain-experimental = "^0.0.8"
|
||||
metaphor-python = "^0.1.11"
|
||||
pillow = "^10.0.0"
|
||||
|
||||
[tool.poetry.group.dev.dependencies]
|
||||
black = "^23.1.0"
|
||||
|
|
|
|||
|
|
@ -0,0 +1,82 @@
|
|||
from typing import Optional
|
||||
from langflow import CustomComponent
|
||||
from langchain.text_splitter import Language
|
||||
from langchain.schema import Document
|
||||
from langflow.utils.util import build_loader_repr_from_documents
|
||||
|
||||
|
||||
class LanguageRecursiveTextSplitterComponent(CustomComponent):
|
||||
display_name: str = "Language Recursive Text Splitter"
|
||||
description: str = "Split text into chunks of a specified length based on language."
|
||||
documentation: str = "https://docs.langflow.org/components/text-splitters#languagerecursivetextsplitter"
|
||||
|
||||
def build_config(self):
|
||||
options = [x.value for x in Language]
|
||||
return {
|
||||
"documents": {
|
||||
"display_name": "Documents",
|
||||
"info": "The documents to split.",
|
||||
},
|
||||
"separator_type": {
|
||||
"display_name": "Separator Type",
|
||||
"info": "The type of separator to use.",
|
||||
"field_type": "str",
|
||||
"options": options,
|
||||
"value": "Python",
|
||||
},
|
||||
"separators": {
|
||||
"display_name": "Separators",
|
||||
"info": "The characters to split on.",
|
||||
"is_list": True,
|
||||
},
|
||||
"chunk_size": {
|
||||
"display_name": "Chunk Size",
|
||||
"info": "The maximum length of each chunk.",
|
||||
"field_type": "int",
|
||||
"value": 1000,
|
||||
},
|
||||
"chunk_overlap": {
|
||||
"display_name": "Chunk Overlap",
|
||||
"info": "The amount of overlap between chunks.",
|
||||
"field_type": "int",
|
||||
"value": 200,
|
||||
},
|
||||
"code": {"show": False},
|
||||
}
|
||||
|
||||
def build(
|
||||
self,
|
||||
documents: list[Document],
|
||||
chunk_size: Optional[int] = 1000,
|
||||
chunk_overlap: Optional[int] = 200,
|
||||
separator_type: Optional[str] = "Python",
|
||||
) -> list[Document]:
|
||||
"""
|
||||
Split text into chunks of a specified length.
|
||||
|
||||
Args:
|
||||
separators (list[str]): The characters to split on.
|
||||
chunk_size (int): The maximum length of each chunk.
|
||||
chunk_overlap (int): The amount of overlap between chunks.
|
||||
length_function (function): The function to use to calculate the length of the text.
|
||||
|
||||
Returns:
|
||||
list[str]: The chunks of text.
|
||||
"""
|
||||
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
||||
|
||||
# Make sure chunk_size and chunk_overlap are ints
|
||||
if isinstance(chunk_size, str):
|
||||
chunk_size = int(chunk_size)
|
||||
if isinstance(chunk_overlap, str):
|
||||
chunk_overlap = int(chunk_overlap)
|
||||
|
||||
splitter = RecursiveCharacterTextSplitter.from_language(
|
||||
language=Language(separator_type),
|
||||
chunk_size=chunk_size,
|
||||
chunk_overlap=chunk_overlap,
|
||||
)
|
||||
|
||||
docs = splitter.split_documents(documents)
|
||||
self.repr_value = build_loader_repr_from_documents(docs)
|
||||
return docs
|
||||
|
|
@ -0,0 +1,79 @@
|
|||
from typing import Optional
|
||||
from langflow import CustomComponent
|
||||
from langchain.schema import Document
|
||||
|
||||
|
||||
class RecursiveCharacterTextSplitterComponent(CustomComponent):
|
||||
display_name: str = "Recursive Character Text Splitter"
|
||||
description: str = "Split text into chunks of a specified length."
|
||||
documentation: str = "https://docs.langflow.org/components/text-splitters#recursivecharactertextsplitter"
|
||||
|
||||
def build_config(self):
|
||||
return {
|
||||
"documents": {
|
||||
"display_name": "Documents",
|
||||
"info": "The documents to split.",
|
||||
},
|
||||
"separators": {
|
||||
"display_name": "Separators",
|
||||
"info": 'The characters to split on.\nIf left empty defaults to ["\\n\\n", "\\n", " ", ""].',
|
||||
"is_list": True,
|
||||
},
|
||||
"chunk_size": {
|
||||
"display_name": "Chunk Size",
|
||||
"info": "The maximum length of each chunk.",
|
||||
"field_type": "int",
|
||||
"value": 1000,
|
||||
},
|
||||
"chunk_overlap": {
|
||||
"display_name": "Chunk Overlap",
|
||||
"info": "The amount of overlap between chunks.",
|
||||
"field_type": "int",
|
||||
"value": 200,
|
||||
},
|
||||
"code": {"show": False},
|
||||
}
|
||||
|
||||
def build(
|
||||
self,
|
||||
documents: list[Document],
|
||||
separators: Optional[list[str]] = None,
|
||||
chunk_size: Optional[int] = 1000,
|
||||
chunk_overlap: Optional[int] = 200,
|
||||
) -> list[Document]:
|
||||
"""
|
||||
Split text into chunks of a specified length.
|
||||
|
||||
Args:
|
||||
separators (list[str]): The characters to split on.
|
||||
chunk_size (int): The maximum length of each chunk.
|
||||
chunk_overlap (int): The amount of overlap between chunks.
|
||||
length_function (function): The function to use to calculate the length of the text.
|
||||
|
||||
Returns:
|
||||
list[str]: The chunks of text.
|
||||
"""
|
||||
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
||||
|
||||
if separators == "":
|
||||
separators = None
|
||||
elif separators:
|
||||
# check if the separators list has escaped characters
|
||||
# if there are escaped characters, unescape them
|
||||
separators = [x.encode().decode("unicode-escape") for x in separators]
|
||||
|
||||
# Make sure chunk_size and chunk_overlap are ints
|
||||
if isinstance(chunk_size, str):
|
||||
chunk_size = int(chunk_size)
|
||||
if isinstance(chunk_overlap, str):
|
||||
chunk_overlap = int(chunk_overlap)
|
||||
splitter = RecursiveCharacterTextSplitter(
|
||||
separators=separators,
|
||||
chunk_size=chunk_size,
|
||||
chunk_overlap=chunk_overlap,
|
||||
)
|
||||
|
||||
docs = splitter.split_documents(documents)
|
||||
# self.repr_value = build_loader_repr_from_documents(docs)
|
||||
self.repr_value = separators
|
||||
return docs
|
||||
|
|
@ -169,8 +169,6 @@ prompts:
|
|||
textsplitters:
|
||||
CharacterTextSplitter:
|
||||
documentation: "https://python.langchain.com/docs/modules/data_connection/document_transformers/text_splitters/character_text_splitter"
|
||||
RecursiveCharacterTextSplitter:
|
||||
documentation: "https://python.langchain.com/docs/modules/data_connection/document_transformers/text_splitters/recursive_text_splitter"
|
||||
toolkits:
|
||||
OpenAPIToolkit:
|
||||
documentation: ""
|
||||
|
|
|
|||
|
|
@ -5,6 +5,7 @@ from langflow.api.utils import merge_nested_dicts_with_renaming
|
|||
from langflow.interface.agents.base import agent_creator
|
||||
from langflow.interface.chains.base import chain_creator
|
||||
from langflow.interface.custom.constants import CUSTOM_COMPONENT_SUPPORTED_TYPES
|
||||
from langflow.interface.custom.utils import extract_inner_type
|
||||
from langflow.interface.document_loaders.base import documentloader_creator
|
||||
from langflow.interface.embeddings.base import embedding_creator
|
||||
from langflow.interface.importing.utils import get_function_custom
|
||||
|
|
@ -84,6 +85,8 @@ def build_langchain_types_dict(): # sourcery skip: dict-assign-update-to-union
|
|||
|
||||
|
||||
def process_type(field_type: str):
|
||||
if field_type.startswith("list") or field_type.startswith("List"):
|
||||
return extract_inner_type(field_type)
|
||||
return "prompt" if field_type == "Prompt" else field_type
|
||||
|
||||
|
||||
|
|
@ -100,6 +103,7 @@ def add_new_custom_field(
|
|||
# if it is, update the value
|
||||
display_name = field_config.pop("display_name", field_name)
|
||||
field_type = field_config.pop("field_type", field_type)
|
||||
field_contains_list = "list" in field_type.lower()
|
||||
field_type = process_type(field_type)
|
||||
field_value = field_config.pop("value", field_value)
|
||||
field_advanced = field_config.pop("advanced", False)
|
||||
|
|
@ -110,7 +114,9 @@ def add_new_custom_field(
|
|||
# If options is a list, then it's a dropdown
|
||||
# If options is None, then it's a list of strings
|
||||
is_list = isinstance(field_config.get("options"), list)
|
||||
field_config["is_list"] = is_list or field_config.get("is_list", False)
|
||||
field_config["is_list"] = (
|
||||
is_list or field_config.get("is_list", False) or field_contains_list
|
||||
)
|
||||
|
||||
if "name" in field_config:
|
||||
warnings.warn(
|
||||
|
|
@ -172,7 +178,7 @@ def extract_type_from_optional(field_type):
|
|||
Returns:
|
||||
str: The extracted type, or an empty string if no type was found.
|
||||
"""
|
||||
match = re.search(r"\[(.*?)\]", field_type)
|
||||
match = re.search(r"\[(.*?)\]$", field_type)
|
||||
return match[1] if match else None
|
||||
|
||||
|
||||
|
|
|
|||
|
|
@ -2,7 +2,7 @@ import re
|
|||
import inspect
|
||||
import importlib
|
||||
from functools import wraps
|
||||
from typing import Optional, Dict, Any, Union
|
||||
from typing import List, Optional, Dict, Any, Union
|
||||
|
||||
from docstring_parser import parse # type: ignore
|
||||
|
||||
|
|
@ -10,6 +10,7 @@ from langflow.template.frontend_node.constants import FORCE_SHOW_FIELDS
|
|||
from langflow.utils import constants
|
||||
from langflow.utils.logger import logger
|
||||
from multiprocess import cpu_count # type: ignore
|
||||
from langchain.schema import Document
|
||||
|
||||
|
||||
def build_template_from_function(
|
||||
|
|
@ -462,3 +463,12 @@ def get_number_of_workers(workers=None):
|
|||
workers = (cpu_count() * 2) + 1
|
||||
logger.debug(f"Number of workers: {workers}")
|
||||
return workers
|
||||
|
||||
|
||||
def build_loader_repr_from_documents(documents: List[Document]) -> str:
|
||||
if documents:
|
||||
avg_length = sum(len(doc.page_content) for doc in documents) / len(documents)
|
||||
return f"""{len(documents)} documents
|
||||
\nAvg. Document Length (characters): {int(avg_length)}
|
||||
Documents: {documents[:3]}..."""
|
||||
return "0 documents"
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue