The Llama Parser

The Llama parser uses llama under the hood to parse pdfs and extract text from them. It is a wrapper around the llama library that provides a more user-friendly interface.

Documentation

LlamaPDFParser

Parse a PDF file using the LlamaParse library.

Parameters:

Name	Type	Description	Default
llama_options	Optional[Dict]	A dictionary containing the options for the LlamaParse converter.	{'show_progress': True, 'ignore_errors': False, 'split_by_page': False, 'invalidate_cache': False, 'do_not_cache': False, 'result_type': 'markdown', 'continuous_mode': True, 'take_screenshot': True, 'disable_ocr': False, 'is_formatting_instruction': False, 'premium_mode': True, 'verbose': False}

Raises:

Type	Description
ValueError	An error occurred while initializing the LlamaParse converter.

Source code in parsestudio/parsers/llama_parser.py

class LlamaPDFParser:
    """
    Parse a PDF file using the LlamaParse library.

    Args:
        llama_options (Optional[Dict], optional): A dictionary containing the options for the LlamaParse converter.

    Raises:
        ValueError: An error occurred while initializing the LlamaParse converter.
    """
    def __init__(
            self,
            llama_options: Optional[Dict] = {
                "show_progress": True,
                "ignore_errors": False,
                "split_by_page": False,
                "invalidate_cache": False,
                "do_not_cache": False,
                "result_type": "markdown",
                "continuous_mode": True,
                "take_screenshot": True,
                "disable_ocr": False,
                "is_formatting_instruction": False,
                "premium_mode": True,
                "verbose": False
            }
            ):

        try:
            self.converter = LlamaParse(
                api_key=os.environ.get("LLAMA_PARSE_KEY"),
                **llama_options
            )

        except Exception as e:
            raise ValueError(
                f"An error occurred while initializing the LlamaParse converter: {e}"
            )

    def load_documents(self, paths: List[str]) -> Generator[Dict, None, None]:
        """
        Load the documents from the given paths and yield the JSON result.

        Args:
            paths (List[str]): A list of paths to the PDF files.

        Yields:
            result (Generator[Dict, None, None]): A generator that yields the JSON result of the document.
        """

        document: List[Dict] = self.converter.get_json_result(paths)
        yield from document

    def _validate_modalities(self, modalities: List[str]) -> None:
        """
        Validate the modalities provided by the user. The valid modalities are: ["text", "tables", "images"]

        Args:
            modalities (List[str]): List of modalities to validate

        Raises:
            ValueError: If the modality is not valid
        """
        valid_modalities = ["text", "tables", "images"]
        for modality in modalities:
            if modality not in valid_modalities:
                raise ValueError(
                    f"Invalid modality: {modality}. The valid modalities are: {valid_modalities}"
                )

    def parse(
            self,
            paths: Union[str, List[str]],
            modalities: List[str] = ["text", "tables", "images"],
        ) -> List[ParserOutput]:
        """
        Parse the PDF file and return the extracted the specified modalities.

        Args:
            paths (Union[str, List[str]]): A path or a list of paths to the PDF files.
            modalities (List[str], optional): List of modalities to extract. Defaults to ["text", "tables", "images"].

        Returns:
            data (List[ParserOutput]): A list of ParserOutput objects containing the extracted modalities.

        Raises:
            ValueError: If the modality is not valid

        Example:
        !!! example
            ```python
            parser = LlamaPDFParser()
            data = parser.parse("path/to/file.pdf", modalities=["text", "tables", "images"])
            print(len(data)) 
            # Output: 1
            text = data[0].text # TextElement
            tables = data[0].tables # List of TableElement
            images = data[0].images # List of ImageElement

            # Access the text
            text = text.text


            # Access the first table
            table = tables[0]
            # Access the markdown representation of the table
            table_md = table.markdown
            # Access the dataframe representation of the table
            table_df = table.dataframe
            # Access the metadata of the table
            page_number = table.metadata.page_number
            bbox = table.metadata.bbox

            # Access the first image
            image = images[0]
            # Access the image object
            image_obj = image.image # PIL Image object
            # Access the metadata of the image
            page_number = image.metadata.page_number
            bbox = image.metadata.bbox 
            ```
        """
        self._validate_modalities(modalities)

        if isinstance(paths, str):
            paths = [paths]

        data = []
        for result in self.load_documents(paths):
            output = self.__export_result(result, modalities)
            data.append(output)

        return data

    def __export_result(
            self, 
            json_result: dict, 
            modalities: List[str]
        ) -> ParserOutput:
        """
        Export the result to the ParserOutput object.

        Args:
            json_result (dict): The JSON result of the document.
            modalities (List[str]): List of modalities to extract.

        Returns:
            output (ParserOutput): The ParserOutput object containing the extracted modalities.
        """
        text = TextElement(text="")
        tables: List[TableElement] = []
        images: List[ImageElement] = []

        job_id: str = json_result["job_id"]
        pages: List[Dict] = json_result["pages"]

        for page in pages:
            if "text" in modalities:
                text.text += self._extract_text(page).text + "\n"

            if "tables" in modalities:
                tables += self._extract_tables(page)

            if "images" in modalities:
                images += self._extract_images(page, job_id)

        return ParserOutput(text=text, tables=tables, images=images)

    @staticmethod
    def _extract_text(page: Dict) -> TextElement:
        """
        Extract the text from the page dict.

        Args:
            page (Dict): A dictionary containing the page information.

        Returns:
            text (TextElement): TextElement object

        Examples:
        !!! example
            ```python
            parser = LlamaPDFParser()
            text = parser._extract_text(page)
            text = text.text
            ```
        """
        return TextElement(text=page["text"])

    @staticmethod
    def _extract_tables(page: Dict) -> List[TableElement]:
        """
        Extract the tables from the page dict.

        Args:
            page (Dict): A dictionary containing the page information.

        Returns:
            tables (List[TableElement]): List of TableElement objects

        Examples:
        !!! example
            ```python
            parser = LlamaPDFParser()
            table = parser._extract_tables(table_item)[0]
            table_md = table.markdown
            table_df = table.dataframe
            page_number = table.metadata.page_number
            bbox = table.metadata.bbox
            ```
        """
        tables: List[TableElement] = []
        for item in page["items"]:
            if item["type"] == "table":
                table_md = item["md"]
                try:
                    table_df = pd.read_csv(io.StringIO(item["csv"]), sep=",")
                except Exception as e:
                    print(f"Error converting table {table_md} to dataframe: {e}")
                    table_df = None

                tables.append(
                    TableElement(
                        markdown=table_md, 
                        dataframe=table_df, 
                        metadata=Metadata(page_number=page["page"])
                    )
                )
        return tables

    def _extract_images(self, page: Dict, job_id: str) -> List[ImageElement]:
        """
        Extract the images from the page dict.

        Args:
            page (Dict): A dictionary containing the page information.
            job_id (str): The job_id of the document.

        Returns:
            images (List[ImageElement]): List of ImageElement objects

        Examples:
        !!! example
            ```python
            parser = LlamaPDFParser()
            image = parser._extract_images(page, job_id)[0]
            image_obj = image.image
            page_number = image.metadata.page_number
            bbox = image.metadata.bbox
            ```
        """
        images: List[ImageElement] = []
        image_dicts = self.converter.get_images([{
            "job_id": job_id,
            "pages": [page]
            }], download_path="llama_images")
        for img in image_dicts:
            image_path = img["path"]
            image = Image.open(image_path).convert("RGB")
            images.append(
                ImageElement(
                    image=image, 
                    metadata=Metadata(page_number=page["page"])
                    )
                )
            os.remove(image_path)
        return images

__export_result(json_result, modalities)

Export the result to the ParserOutput object.

Parameters:

Name	Type	Description	Default
json_result	dict	The JSON result of the document.	required
modalities	List[str]	List of modalities to extract.	required

Returns:

Name	Type	Description
output	ParserOutput	The ParserOutput object containing the extracted modalities.

Source code in parsestudio/parsers/llama_parser.py

def __export_result(
        self, 
        json_result: dict, 
        modalities: List[str]
    ) -> ParserOutput:
    """
    Export the result to the ParserOutput object.

    Args:
        json_result (dict): The JSON result of the document.
        modalities (List[str]): List of modalities to extract.

    Returns:
        output (ParserOutput): The ParserOutput object containing the extracted modalities.
    """
    text = TextElement(text="")
    tables: List[TableElement] = []
    images: List[ImageElement] = []

    job_id: str = json_result["job_id"]
    pages: List[Dict] = json_result["pages"]

    for page in pages:
        if "text" in modalities:
            text.text += self._extract_text(page).text + "\n"

        if "tables" in modalities:
            tables += self._extract_tables(page)

        if "images" in modalities:
            images += self._extract_images(page, job_id)

    return ParserOutput(text=text, tables=tables, images=images)

_extract_images(page, job_id)

Extract the images from the page dict.

Parameters:

Name	Type	Description	Default
page	Dict	A dictionary containing the page information.	required
job_id	str	The job_id of the document.	required

Returns:

Name	Type	Description
images	List[ImageElement]	List of ImageElement objects

Examples:

Example

parser = LlamaPDFParser()
image = parser._extract_images(page, job_id)[0]
image_obj = image.image
page_number = image.metadata.page_number
bbox = image.metadata.bbox

Source code in parsestudio/parsers/llama_parser.py

def _extract_images(self, page: Dict, job_id: str) -> List[ImageElement]:
    """
    Extract the images from the page dict.

    Args:
        page (Dict): A dictionary containing the page information.
        job_id (str): The job_id of the document.

    Returns:
        images (List[ImageElement]): List of ImageElement objects

    Examples:
    !!! example
        ```python
        parser = LlamaPDFParser()
        image = parser._extract_images(page, job_id)[0]
        image_obj = image.image
        page_number = image.metadata.page_number
        bbox = image.metadata.bbox
        ```
    """
    images: List[ImageElement] = []
    image_dicts = self.converter.get_images([{
        "job_id": job_id,
        "pages": [page]
        }], download_path="llama_images")
    for img in image_dicts:
        image_path = img["path"]
        image = Image.open(image_path).convert("RGB")
        images.append(
            ImageElement(
                image=image, 
                metadata=Metadata(page_number=page["page"])
                )
            )
        os.remove(image_path)
    return images

_extract_tables(page) staticmethod

Extract the tables from the page dict.

Parameters:

Name	Type	Description	Default
page	Dict	A dictionary containing the page information.	required

Returns:

Name	Type	Description
tables	List[TableElement]	List of TableElement objects

Examples:

Example

parser = LlamaPDFParser()
table = parser._extract_tables(table_item)[0]
table_md = table.markdown
table_df = table.dataframe
page_number = table.metadata.page_number
bbox = table.metadata.bbox

Source code in parsestudio/parsers/llama_parser.py

@staticmethod
def _extract_tables(page: Dict) -> List[TableElement]:
    """
    Extract the tables from the page dict.

    Args:
        page (Dict): A dictionary containing the page information.

    Returns:
        tables (List[TableElement]): List of TableElement objects

    Examples:
    !!! example
        ```python
        parser = LlamaPDFParser()
        table = parser._extract_tables(table_item)[0]
        table_md = table.markdown
        table_df = table.dataframe
        page_number = table.metadata.page_number
        bbox = table.metadata.bbox
        ```
    """
    tables: List[TableElement] = []
    for item in page["items"]:
        if item["type"] == "table":
            table_md = item["md"]
            try:
                table_df = pd.read_csv(io.StringIO(item["csv"]), sep=",")
            except Exception as e:
                print(f"Error converting table {table_md} to dataframe: {e}")
                table_df = None

            tables.append(
                TableElement(
                    markdown=table_md, 
                    dataframe=table_df, 
                    metadata=Metadata(page_number=page["page"])
                )
            )
    return tables

_extract_text(page) staticmethod

Extract the text from the page dict.

Parameters:

Name	Type	Description	Default
page	Dict	A dictionary containing the page information.	required

Returns:

Name	Type	Description
text	TextElement	TextElement object

Examples:

Example

parser = LlamaPDFParser()
text = parser._extract_text(page)
text = text.text

Source code in parsestudio/parsers/llama_parser.py

@staticmethod
def _extract_text(page: Dict) -> TextElement:
    """
    Extract the text from the page dict.

    Args:
        page (Dict): A dictionary containing the page information.

    Returns:
        text (TextElement): TextElement object

    Examples:
    !!! example
        ```python
        parser = LlamaPDFParser()
        text = parser._extract_text(page)
        text = text.text
        ```
    """
    return TextElement(text=page["text"])

_validate_modalities(modalities)

Validate the modalities provided by the user. The valid modalities are: ["text", "tables", "images"]

Parameters:

Name	Type	Description	Default
modalities	List[str]	List of modalities to validate	required

Raises:

Type	Description
ValueError	If the modality is not valid

Source code in parsestudio/parsers/llama_parser.py

def _validate_modalities(self, modalities: List[str]) -> None:
    """
    Validate the modalities provided by the user. The valid modalities are: ["text", "tables", "images"]

    Args:
        modalities (List[str]): List of modalities to validate

    Raises:
        ValueError: If the modality is not valid
    """
    valid_modalities = ["text", "tables", "images"]
    for modality in modalities:
        if modality not in valid_modalities:
            raise ValueError(
                f"Invalid modality: {modality}. The valid modalities are: {valid_modalities}"
            )

load_documents(paths)

Load the documents from the given paths and yield the JSON result.

Parameters:

Name	Type	Description	Default
paths	List[str]	A list of paths to the PDF files.	required

Yields:

Name	Type	Description
result	Generator[Dict, None, None]	A generator that yields the JSON result of the document.

Source code in parsestudio/parsers/llama_parser.py

def load_documents(self, paths: List[str]) -> Generator[Dict, None, None]:
    """
    Load the documents from the given paths and yield the JSON result.

    Args:
        paths (List[str]): A list of paths to the PDF files.

    Yields:
        result (Generator[Dict, None, None]): A generator that yields the JSON result of the document.
    """

    document: List[Dict] = self.converter.get_json_result(paths)
    yield from document

parse(paths, modalities=['text', 'tables', 'images'])

Parse the PDF file and return the extracted the specified modalities.

Parameters:

Name	Type	Description	Default
paths	Union[str, List[str]]	A path or a list of paths to the PDF files.	required
modalities	List[str]	List of modalities to extract. Defaults to ["text", "tables", "images"].	['text', 'tables', 'images']

Returns:

Name	Type	Description
data	List[ParserOutput]	A list of ParserOutput objects containing the extracted modalities.

Raises:

Type	Description
ValueError	If the modality is not valid

Example:

Example

parser = LlamaPDFParser()
data = parser.parse("path/to/file.pdf", modalities=["text", "tables", "images"])
print(len(data)) 
# Output: 1
text = data[0].text # TextElement
tables = data[0].tables # List of TableElement
images = data[0].images # List of ImageElement

# Access the text
text = text.text


# Access the first table
table = tables[0]
# Access the markdown representation of the table
table_md = table.markdown
# Access the dataframe representation of the table
table_df = table.dataframe
# Access the metadata of the table
page_number = table.metadata.page_number
bbox = table.metadata.bbox

# Access the first image
image = images[0]
# Access the image object
image_obj = image.image # PIL Image object
# Access the metadata of the image
page_number = image.metadata.page_number
bbox = image.metadata.bbox 

Source code in parsestudio/parsers/llama_parser.py

def parse(
        self,
        paths: Union[str, List[str]],
        modalities: List[str] = ["text", "tables", "images"],
    ) -> List[ParserOutput]:
    """
    Parse the PDF file and return the extracted the specified modalities.

    Args:
        paths (Union[str, List[str]]): A path or a list of paths to the PDF files.
        modalities (List[str], optional): List of modalities to extract. Defaults to ["text", "tables", "images"].

    Returns:
        data (List[ParserOutput]): A list of ParserOutput objects containing the extracted modalities.

    Raises:
        ValueError: If the modality is not valid

    Example:
    !!! example
        ```python
        parser = LlamaPDFParser()
        data = parser.parse("path/to/file.pdf", modalities=["text", "tables", "images"])
        print(len(data)) 
        # Output: 1
        text = data[0].text # TextElement
        tables = data[0].tables # List of TableElement
        images = data[0].images # List of ImageElement

        # Access the text
        text = text.text


        # Access the first table
        table = tables[0]
        # Access the markdown representation of the table
        table_md = table.markdown
        # Access the dataframe representation of the table
        table_df = table.dataframe
        # Access the metadata of the table
        page_number = table.metadata.page_number
        bbox = table.metadata.bbox

        # Access the first image
        image = images[0]
        # Access the image object
        image_obj = image.image # PIL Image object
        # Access the metadata of the image
        page_number = image.metadata.page_number
        bbox = image.metadata.bbox 
        ```
    """
    self._validate_modalities(modalities)

    if isinstance(paths, str):
        paths = [paths]

    data = []
    for result in self.load_documents(paths):
        output = self.__export_result(result, modalities)
        data.append(output)

    return data