Components

VectorDB

• VectorDB
  • Submodules
  • Base
    • VectorDB
      • VectorDB.aadd_docs()
      • VectorDB.add_docs()
      • VectorDB.aget_chunk()
      • VectorDB.delete()
      • VectorDB.get_chunk()
  • Chroma Client
    • ChromaClient()
    • host
    • port
    • collection_name
    • embedding_type
    • embedding_model
    • embedding_function
    • client
    • collection
    • langchain_client
  • Deeplake Client
    • DeepLakeClient()
    • store_path
    • embedding_type
    • embedding_model
    • embedding_function
    • client
    • collection_name
    • langchain_client
  • Module Contents

Embedding

Class for embedding.

This module provides:

— Embedding

class grag.components.embedding.Embedding(embedding_type: str, embedding_model: str)

Bases: object

A class for vector embeddings.

Supports:

huggingface sentence transformers -> model_type = ‘sentence-transformers’ huggingface instructor embeddings -> model_type = ‘instructor-embedding’

embedding_type

embedding model type, refer above for supported types

embedding_model

huggingface model name

embedding_function

langchain embedding type

LLM

Class for LLM.

grag.components.llm.LLM(model_name: str, quantization: str, pipeline: str, device_map: str = 'auto', task: str = 'text-generation', max_new_tokens: str = '1024', temperature: str | float = 0.1, n_batch: str | int = 1024, n_ctx: str | int = 6000, n_gpu_layers: str | int = -1, std_out: bool | str = True, base_dir: str | Path = PosixPath('models'), callbacks=None)

A class for managing and utilizing large language models (LLMs).

This class facilitates the loading and operation of large language models using different pipelines and settings. It supports both local and Hugging Face-based model management, with adjustable parameters for quantization, computational specifics, and output control.

grag.components.llm.model_name

Name of the model to be loaded.

Type:

str

grag.components.llm.quantization

Quantization setting for the model, affecting performance and memory usage.

Type:

str

grag.components.llm.pipeline

Type of pipeline (‘llama_cpp’ or ‘hf’) used for model operations.

Type:

str

grag.components.llm.device_map

Device mapping for model execution, defaults to ‘auto’.

Type:

str

grag.components.llm.task

The task for which the model is being used, defaults to ‘text-generation’.

Type:

str

grag.components.llm.max_new_tokens

Maximum number of new tokens to be generated, defaults to 1024.

Type:

int

grag.components.llm.temperature

Sampling temperature for generation, affecting randomness.

Type:

float

grag.components.llm.n_batch

Number of batches for GPU CPP, impacting batch processing.

Type:

int

grag.components.llm.n_ctx

Context size for CPP, defining the extent of context considered.

Type:

int

grag.components.llm.n_gpu_layers

Number of GPU layers for CPP, specifying computational depth.

Type:

int

grag.components.llm.std_out

Flag or descriptor for standard output during operations.

Type:

bool or str

grag.components.llm.base_dir

Base directory path for model files, defaults to ‘models’.

Type:

str or Path

grag.components.llm.callbacks

List of callback functions for additional processing.

Type:

list or None

Retriever

Class for retriever.

This module provides:

— Retriever

grag.components.multivec_retriever.Retriever(vectordb: VectorDB | None = None, store_path: str | Path = PosixPath('data/doc_store'), top_k: str | int = 3, id_key: str = 'doc_id', namespace: str = '71e4b558187b270922923569301f1039', client_kwargs: Dict[str, Any] | None = None)

A class for multi vector retriever.

It connects to a vector database and a local file store. It is used to return most similar chunks from a vector store but has the additional functionality to return a linked document, chunk, etc.

grag.components.multivec_retriever.vectordb

ChromaClient class instance from components.client (Optional, if the user provides it, store_path, id_key and namespace is not considered)

grag.components.multivec_retriever.store_path

Path to the local file store

grag.components.multivec_retriever.id_key

A key prefix for identifying documents

grag.components.multivec_retriever.store

langchain.storage.LocalFileStore object, stores the key value pairs of document id and parent file

grag.components.multivec_retriever.retriever

langchain.retrievers.multi_vector.MultiVectorRetriever class instance, langchain’s multi-vector retriever

grag.components.multivec_retriever.splitter

TextSplitter class instance from components.text_splitter

grag.components.multivec_retriever.namespace

Namespace for producing unique id

grag.components.multivec_retriever.top_k

Number of top chunks to return from similarity search.

Parse PDF

Classes for parsing files.

This module provides:

— ParsePDF

grag.components.parse_pdf.ParsePDF(single_text_out: bool = True, strategy: str = 'hi_res', infer_table_structure: bool = True, extract_images: bool = True, image_output_dir: str | None = None, add_captions_to_text: bool = True, add_captions_to_blocks: bool = True, add_caption_first: bool = True, table_as_html: bool = False)

Parsing and partitioning PDF documents into Text, Table or Image elements.

grag.components.parse_pdf.single_text_out

Whether to combine all text elements into a single output document.

Type:

bool

grag.components.parse_pdf.strategy

The strategy for PDF partitioning; default is “hi_res” for better accuracy.

Type:

str

grag.components.parse_pdf.infer_table_structure

Whether to extract tables during partitioning.

Type:

bool

grag.components.parse_pdf.extract_images

Whether to extract images.

Type:

bool

grag.components.parse_pdf.image_output_dir

Directory to save extracted images, if any.

Type:

str

grag.components.parse_pdf.add_captions_to_text

Whether to include figure captions in text output. Default is True.

Type:

bool

grag.components.parse_pdf.add_captions_to_blocks

Whether to add captions to table and image blocks. Default is True.

Type:

bool

grag.components.parse_pdf.add_caption_first

Whether to place captions before their corresponding image or table in the output. Default is True.

Type:

bool

grag.components.parse_pdf.table_as_html

Whether to add table elements as HTML. Default is False.

Type:

bool

Prompt

Classes for prompts.

This module provides:

— Prompt: for generic prompts

— FewShotPrompt: for few-shot prompts

class grag.components.prompt.FewShotPrompt(*, name: str = 'custom_prompt', llm_type: str = 'None', task: str = 'QA', source: str = 'NoSource', doc_chain: str = 'stuff', language: str = 'en', filepath: str | None = None, input_keys: List[str], template: str, prompt: PromptTemplate | None = None, output_keys: List[str], examples: List[Dict[str, Any]], prefix: str, suffix: str, example_template: str)

Bases: Prompt

A class for generic prompts.

name

The prompt name (Optional, defaults to “custom_prompt”) (Parent Class)

Type:

str

llm_type

The type of llm, llama2, etc (Optional, defaults to “None”) (Parent Class)

Type:

str

task

The task (Optional, defaults to QA) (Parent Class)

Type:

str

source

The source of the prompt (Optional, defaults to “NoSource”) (Parent Class)

Type:

str

doc_chain

The doc chain for the prompt (“stuff”, “refine”) (Optional, defaults to “stuff”) (Parent Class)

Type:

str

language

The language of the prompt (Optional, defaults to “en”) (Parent Class)

Type:

str

filepath

The filepath of the prompt (Optional) (Parent Class)

Type:

str

input_keys

The input keys for the prompt (Parent Class)

Type:

List[str]

input_keys

The output keys for the prompt

Type:

List[str]

prefix

The template prefix for the prompt

Type:

str

suffix

The template suffix for the prompt

Type:

str

example_template

The template for formatting the examples

Type:

str

examples

The list of examples, each example is a dictionary with respective keys

Type:

List[Dict[str, Any]]

example_template: str

examples: List[Dict[str, Any]]

model_computed_fields: ClassVar[dict[str, ComputedFieldInfo]] = {}

A dictionary of computed field names and their corresponding ComputedFieldInfo objects.

model_config: ClassVar[ConfigDict] = {}

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].

model_fields: ClassVar[dict[str, FieldInfo]] = {'doc_chain': FieldInfo(annotation=str, required=False, default='stuff'), 'example_template': FieldInfo(annotation=str, required=True), 'examples': FieldInfo(annotation=List[Dict[str, Any]], required=True), 'filepath': FieldInfo(annotation=Union[str, NoneType], required=False, default=None, exclude=True), 'input_keys': FieldInfo(annotation=List[str], required=True), 'language': FieldInfo(annotation=str, required=False, default='en'), 'llm_type': FieldInfo(annotation=str, required=False, default='None'), 'name': FieldInfo(annotation=str, required=False, default='custom_prompt'), 'output_keys': FieldInfo(annotation=List[str], required=True), 'prefix': FieldInfo(annotation=str, required=True), 'prompt': FieldInfo(annotation=Union[PromptTemplate, NoneType], required=False, default=None, exclude=True, repr=False), 'source': FieldInfo(annotation=str, required=False, default='NoSource'), 'suffix': FieldInfo(annotation=str, required=True), 'task': FieldInfo(annotation=str, required=False, default='QA'), 'template': FieldInfo(annotation=str, required=True)}

Metadata about the fields defined on the model, mapping of field names to [FieldInfo][pydantic.fields.FieldInfo].

This replaces Model.__fields__ from Pydantic V1.

output_keys: List[str]

prefix: str

suffix: str

classmethod validate_examples(v) → List[Dict[str, Any]]

Validate the examples field.

classmethod validate_output_keys(v) → List[str]

Validate the output_keys field.

class grag.components.prompt.Prompt(*, name: str = 'custom_prompt', llm_type: str = 'None', task: str = 'QA', source: str = 'NoSource', doc_chain: str = 'stuff', language: str = 'en', filepath: str | None = None, input_keys: List[str], template: str, prompt: PromptTemplate | None = None)

Bases: BaseModel

A class for generic prompts.

name

The prompt name (Optional, defaults to “custom_prompt”)

Type:

str

llm_type

The type of llm, llama2, etc (Optional, defaults to “None”)

Type:

str

task

The task (Optional, defaults to QA)

Type:

str

source

The source of the prompt (Optional, defaults to “NoSource”)

Type:

str

doc_chain

The doc chain for the prompt (“stuff”, “refine”) (Optional, defaults to “stuff”)

Type:

str

language

The language of the prompt (Optional, defaults to “en”)

Type:

str

filepath

The filepath of the prompt (Optional)

Type:

str

input_keys

The input keys for the prompt

Type:

List[str]

template (str): The template for the prompt

doc_chain: str

filepath: str | None

format(**kwargs) → str

Formats the prompt with provided keys and returns a string.

input_keys: List[str]

language: str

llm_type: str

classmethod load(filepath: Path | str)

Loads a json file and returns a Prompt class.

model_computed_fields: ClassVar[dict[str, ComputedFieldInfo]] = {}

A dictionary of computed field names and their corresponding ComputedFieldInfo objects.

model_config: ClassVar[ConfigDict] = {}

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].

model_fields: ClassVar[dict[str, FieldInfo]] = {'doc_chain': FieldInfo(annotation=str, required=False, default='stuff'), 'filepath': FieldInfo(annotation=Union[str, NoneType], required=False, default=None, exclude=True), 'input_keys': FieldInfo(annotation=List[str], required=True), 'language': FieldInfo(annotation=str, required=False, default='en'), 'llm_type': FieldInfo(annotation=str, required=False, default='None'), 'name': FieldInfo(annotation=str, required=False, default='custom_prompt'), 'prompt': FieldInfo(annotation=Union[PromptTemplate, NoneType], required=False, default=None, exclude=True, repr=False), 'source': FieldInfo(annotation=str, required=False, default='NoSource'), 'task': FieldInfo(annotation=str, required=False, default='QA'), 'template': FieldInfo(annotation=str, required=True)}

Metadata about the fields defined on the model, mapping of field names to [FieldInfo][pydantic.fields.FieldInfo].

This replaces Model.__fields__ from Pydantic V1.

name: str

prompt: PromptTemplate | None

save(filepath: Path | str | None, overwrite=False) → None | ValueError

Saves the prompt class into a json file.

source: str

task: str

template: str

classmethod validate_doc_chain(v: str) → str

Validate the doc_chain field.

classmethod validate_input_keys(v) → List[str]

Validate the input_keys field.

classmethod validate_task(v: str) → str

Validate the task field.

Text Splitter

Class for splitting/chunking text.

This module provides:

— TextSplitter

grag.components.text_splitter.TextSplitter(chunk_size: int | str = 2000, chunk_overlap: int | str = 400)

Class for recursively chunking text, it prioritizes ‘/n/n then ‘/n’ and so on.

grag.components.text_splitter.chunk_size

maximum size of chunk

grag.components.text_splitter.chunk_overlap

chunk overlap size

Utils

Utils functions.

This module provides:

— stuff_docs: concats langchain documents into string

— load_prompt: loads json prompt to langchain prompt

— find_config_path: finds the path of the ‘config.ini’ file by traversing up the directory tree from the current path.

— get_config: retrieves and parses the configuration settings from the ‘config.ini’ file.

— configure_args: a decorator to configure class instantiation arguments from a ‘config.ini’ file.

grag.components.utils.configure_args(cls)

Decorator to configure class instantiation arguments from a ‘config.ini’ file, based on the class’s module name.

This function reads configuration specific to a class’s module from ‘config.ini’, then uses it to override or provide defaults for keyword arguments passed during class instantiation.

Parameters:

cls (class) – The class whose instantiation is to be configured.

Returns:

A wrapped class constructor that uses modified arguments based on the configuration.

Return type:

function

Raises:

TypeError – If there is a mismatch in provided arguments and class constructor requirements.

grag.components.utils.find_config_path(current_path: Path)

Finds the path of the ‘config.ini’ file by traversing up the directory tree from the current path.

This function starts at the current path and moves up the directory tree until it finds a file named ‘config.ini’. If ‘config.ini’ is not found by the time the root of the directory tree is reached, None is returned.

Parameters:

current_path (Path) – The starting point for the search, typically the location of the script being executed.

Returns:

None or the path to the found ‘config.ini’ file.

Return type:

Path

grag.components.utils.get_config(load_env=False)

Retrieves and parses the configuration settings from the ‘config.ini’ file.

This function locates the ‘config.ini’ file by calling find_config_path using the script’s current location. It initializes a ConfigParser object to read the configuration settings from the located ‘config.ini’ file. Optionally, it can also load environment variables from a .env file specified in the config. If a config file cannot be read, a default dictionary is returned.

Parameters:

load_env (bool) – If True, load environment variables from the path specified in the ‘config.ini’. Defaults to False.

Returns:

A parser object containing the configuration settings from ‘config.ini’, or a defaultdict

with None if the file is not found or an empty dict{dict{}}.

Return type:

ConfigParser

grag.components.utils.stuff_docs(docs: List[Document]) → str

Concatenates langchain documents into a string using ‘nn’ seperator.

Parameters:

docs – List of langchain_core.documents.Document

Returns:

string of document page content joined by ‘nn’

Module contents