Python Usage

Python options

Some parameters are specific to the Python implementation

Number of workers

This controls the number of workers to use sending API requests for parsing. The default is 4.

Python

parser = LlamaParse(
  num_workers=10
)

Check interval

In synchronous mode (see below), Python will poll to check the status of the job. The default is 1 second.

Python

parser = LlamaParse(
  check_interval=10
)

Verbose mode

By default, LlamaParse will print the status of the job as it is uploaded and checked. You can disable this output.

Python

parser = LlamaParse(
  verbose=False
)

Use with SimpleDirectoryReader

You can use LlamaParse directly within LlamaIndex by using SimpleDirectoryReader. This will parse all files in a directory called data and return the parsed documents.

from llama_cloud_services import LlamaParse
from llama_index.core import SimpleDirectoryReader

parser = LlamaParse()

file_extractor = {".pdf": parser}
documents = SimpleDirectoryReader(
    "./data", file_extractor=file_extractor
).load_data()

Direct usage

It is also possible to call the parser directly, in one of 4 modes:

Synchronous parsing"

documents = parser.load_data("./my_file.pdf")

Synchronous batch parsing

documents = parser.load_data(["./my_file1.pdf", "./my_file2.pdf"])

Asynchronous parsing

documents = await parser.aload_data("./my_file.pdf")

Asynchronous batch parsing

documents = await parser.aload_data(["./my_file1.pdf", "./my_file2.pdf"])

Large document partitioning

The LlamaParse.partition_pages option can be used to split parsing of a large document into smaller parse jobs. By default, partitioning is disabled to maintain consistent default behavior between the Python SDK and the raw API.

parser = LlamaParse(
    api_key="llx-...",
    num_workers=4,
    partition_pages=100,  # Split large documents into partitions of up to 100 pages each
)
result = parser.parse("./1000-page-long.pdf")

In this example, parsing of the 1000 page document would be split into 10 jobs (with 100 pages per job), with up to 4 of the jobs being run concurrently at a time. Once all jobs are completed, the partitioned result would be returned.

partition_pages can be used in conjunction with target_pages and max_pages.