Async Ingestion Pipeline + Metadata Extraction#
Recently, LlamaIndex has introduced async metadata extraction. Let’s compare metadata extraction speeds in an ingestion pipeline using a newer and older version of LlamaIndex.
We will test a pipeline using the classic Paul Graham essay.
!mkdir -p 'data/paul_graham/'
!wget 'https://raw.githubusercontent.com/run-llama/llama_index/main/docs/examples/data/paul_graham/paul_graham_essay.txt' -O 'data/paul_graham/paul_graham_essay.txt'
import os
os.environ[
"OPENAI_API_KEY"
] = "sk-...
New LlamaIndex Ingestion#
Using a version of LlamaIndex greater or equal to v0.9.7, we can take advantage of improved async metadata extraction within ingestion pipelines.
NOTE: Restart your notebook after installing a new version!
!pip install "llama_index>=0.9.7"
NOTE: The num_workers kwarg controls how many requests can be outgoing at a given time using an async semaphore. Setting it higher may increase speeds, but can also lead to timeouts or rate limits, so set it wisely.
from llama_index.embeddings import OpenAIEmbedding
from llama_index.llms import OpenAI
from llama_index.ingestion import IngestionPipeline
from llama_index.extractors import TitleExtractor, SummaryExtractor
from llama_index.text_splitter import SentenceSplitter
from llama_index.schema import MetadataMode
def build_pipeline():
llm = OpenAI(model="gpt-3.5-turbo-1106", temperature=0.1)
transformations = [
SentenceSplitter(chunk_size=1024, chunk_overlap=20),
TitleExtractor(
llm=llm, metadata_mode=MetadataMode.EMBED, num_workers=8
),
SummaryExtractor(
llm=llm, metadata_mode=MetadataMode.EMBED, num_workers=8
),
OpenAIEmbedding(),
]
return IngestionPipeline(transformations=transformations)
from llama_index import SimpleDirectoryReader
documents = SimpleDirectoryReader("./data/paul_graham").load_data()
import time
times = []
for _ in range(3):
time.sleep(30) # help prevent rate-limits/timeouts, keeps each run fair
pipline = build_pipeline()
start = time.time()
nodes = await pipline.arun(documents=documents)
end = time.time()
times.append(end - start)
print(f"Average time: {sum(times) / len(times)}")
100%|██████████| 5/5 [00:01<00:00, 3.99it/s]
100%|██████████| 18/18 [00:07<00:00, 2.36it/s]
100%|██████████| 5/5 [00:01<00:00, 2.97it/s]
100%|██████████| 18/18 [00:06<00:00, 2.63it/s]
100%|██████████| 5/5 [00:01<00:00, 3.84it/s]
100%|██████████| 18/18 [01:07<00:00, 3.75s/it]
Average time: 31.196589946746826
The current openai python client package is a tad unstable – sometimes async jobs will timeout, skewing the average. You can see the last progress bar took 1 minute instead of the previous 6 or 7 seconds, skewing the average.
Old LlamaIndex Ingestion#
Now, lets compare to an older version of LlamaIndex, which was using “fake” async for metadata extraction.
NOTE: Restart your notebook after installing the new version!
!pip install "llama_index<0.9.6"
Collecting llama_index<0.9.6
Obtaining dependency information for llama_index<0.9.6 from https://files.pythonhosted.org/packages/ac/3c/dee8ec4fecaaeabbd8a61ade9ddb6af09d05553c2a0acbebd1b559eaeb30/llama_index-0.9.5-py3-none-any.whl.metadata
Downloading llama_index-0.9.5-py3-none-any.whl.metadata (8.2 kB)
Requirement already satisfied: SQLAlchemy[asyncio]>=1.4.49 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from llama_index<0.9.6) (2.0.23)
Requirement already satisfied: aiohttp<4.0.0,>=3.8.6 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from llama_index<0.9.6) (3.8.6)
Requirement already satisfied: aiostream<0.6.0,>=0.5.2 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from llama_index<0.9.6) (0.5.2)
Requirement already satisfied: beautifulsoup4<5.0.0,>=4.12.2 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from llama_index<0.9.6) (4.12.2)
Requirement already satisfied: dataclasses-json in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from llama_index<0.9.6) (0.5.14)
Requirement already satisfied: deprecated>=1.2.9.3 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from llama_index<0.9.6) (1.2.14)
Requirement already satisfied: fsspec>=2023.5.0 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from llama_index<0.9.6) (2023.10.0)
Requirement already satisfied: httpx in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from llama_index<0.9.6) (0.24.1)
Requirement already satisfied: nest-asyncio<2.0.0,>=1.5.8 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from llama_index<0.9.6) (1.5.8)
Requirement already satisfied: nltk<4.0.0,>=3.8.1 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from llama_index<0.9.6) (3.8.1)
Requirement already satisfied: numpy in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from llama_index<0.9.6) (1.24.4)
Requirement already satisfied: openai>=1.1.0 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from llama_index<0.9.6) (1.3.5)
Requirement already satisfied: pandas in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from llama_index<0.9.6) (2.0.3)
Requirement already satisfied: tenacity<9.0.0,>=8.2.0 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from llama_index<0.9.6) (8.2.3)
Requirement already satisfied: tiktoken>=0.3.3 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from llama_index<0.9.6) (0.5.1)
Requirement already satisfied: typing-extensions>=4.5.0 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from llama_index<0.9.6) (4.8.0)
Requirement already satisfied: typing-inspect>=0.8.0 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from llama_index<0.9.6) (0.8.0)
Requirement already satisfied: urllib3<2 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from llama_index<0.9.6) (1.26.18)
Requirement already satisfied: attrs>=17.3.0 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from aiohttp<4.0.0,>=3.8.6->llama_index<0.9.6) (23.1.0)
Requirement already satisfied: charset-normalizer<4.0,>=2.0 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from aiohttp<4.0.0,>=3.8.6->llama_index<0.9.6) (3.3.2)
Requirement already satisfied: multidict<7.0,>=4.5 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from aiohttp<4.0.0,>=3.8.6->llama_index<0.9.6) (6.0.4)
Requirement already satisfied: async-timeout<5.0,>=4.0.0a3 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from aiohttp<4.0.0,>=3.8.6->llama_index<0.9.6) (4.0.3)
Requirement already satisfied: yarl<2.0,>=1.0 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from aiohttp<4.0.0,>=3.8.6->llama_index<0.9.6) (1.9.2)
Requirement already satisfied: frozenlist>=1.1.1 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from aiohttp<4.0.0,>=3.8.6->llama_index<0.9.6) (1.4.0)
Requirement already satisfied: aiosignal>=1.1.2 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from aiohttp<4.0.0,>=3.8.6->llama_index<0.9.6) (1.3.1)
Requirement already satisfied: soupsieve>1.2 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from beautifulsoup4<5.0.0,>=4.12.2->llama_index<0.9.6) (2.5)
Requirement already satisfied: wrapt<2,>=1.10 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from deprecated>=1.2.9.3->llama_index<0.9.6) (1.15.0)
Requirement already satisfied: click in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from nltk<4.0.0,>=3.8.1->llama_index<0.9.6) (8.1.7)
Requirement already satisfied: joblib in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from nltk<4.0.0,>=3.8.1->llama_index<0.9.6) (1.3.2)
Requirement already satisfied: regex>=2021.8.3 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from nltk<4.0.0,>=3.8.1->llama_index<0.9.6) (2023.10.3)
Requirement already satisfied: tqdm in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from nltk<4.0.0,>=3.8.1->llama_index<0.9.6) (4.66.1)
Requirement already satisfied: anyio<4,>=3.5.0 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from openai>=1.1.0->llama_index<0.9.6) (3.7.1)
Requirement already satisfied: distro<2,>=1.7.0 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from openai>=1.1.0->llama_index<0.9.6) (1.8.0)
Requirement already satisfied: pydantic<3,>=1.9.0 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from openai>=1.1.0->llama_index<0.9.6) (1.10.12)
Requirement already satisfied: certifi in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from httpx->llama_index<0.9.6) (2023.7.22)
Requirement already satisfied: httpcore<0.18.0,>=0.15.0 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from httpx->llama_index<0.9.6) (0.17.3)
Requirement already satisfied: idna in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from httpx->llama_index<0.9.6) (3.4)
Requirement already satisfied: sniffio in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from httpx->llama_index<0.9.6) (1.3.0)
Requirement already satisfied: greenlet!=0.4.17 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from SQLAlchemy[asyncio]>=1.4.49->llama_index<0.9.6) (3.0.1)
Requirement already satisfied: requests>=2.26.0 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from tiktoken>=0.3.3->llama_index<0.9.6) (2.31.0)
Requirement already satisfied: mypy-extensions>=0.3.0 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from typing-inspect>=0.8.0->llama_index<0.9.6) (1.0.0)
Requirement already satisfied: marshmallow<4.0.0,>=3.18.0 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from dataclasses-json->llama_index<0.9.6) (3.20.1)
Requirement already satisfied: python-dateutil>=2.8.2 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from pandas->llama_index<0.9.6) (2.8.2)
Requirement already satisfied: pytz>=2020.1 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from pandas->llama_index<0.9.6) (2023.3.post1)
Requirement already satisfied: tzdata>=2022.1 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from pandas->llama_index<0.9.6) (2023.3)
Requirement already satisfied: h11<0.15,>=0.13 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from httpcore<0.18.0,>=0.15.0->httpx->llama_index<0.9.6) (0.14.0)
Requirement already satisfied: packaging>=17.0 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from marshmallow<4.0.0,>=3.18.0->dataclasses-json->llama_index<0.9.6) (23.2)
Requirement already satisfied: six>=1.5 in /home/loganm/.cache/pypoetry/virtualenvs/llama-index-4a-wkI5X-py3.11/lib/python3.11/site-packages (from python-dateutil>=2.8.2->pandas->llama_index<0.9.6) (1.16.0)
Downloading llama_index-0.9.5-py3-none-any.whl (893 kB)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 893.9/893.9 kB 6.0 MB/s eta 0:00:00a 0:00:01
?25hInstalling collected packages: llama_index
Attempting uninstall: llama_index
Found existing installation: llama-index 0.9.8.post1
Uninstalling llama-index-0.9.8.post1:
Successfully uninstalled llama-index-0.9.8.post1
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
trulens-eval 0.18.0 requires llama-index==0.8.69, but you have llama-index 0.9.5 which is incompatible.
trulens-eval 0.18.0 requires typing-extensions==4.5.0, but you have typing-extensions 4.8.0 which is incompatible.
Successfully installed llama_index-0.9.5
[notice] A new release of pip is available: 23.2.1 -> 23.3.1
[notice] To update, run: pip install --upgrade pip
import os
os.environ["OPENAI_API_KEY"] = "sk-..."
from llama_index.embeddings import OpenAIEmbedding
from llama_index.llms import OpenAI
from llama_index.ingestion import IngestionPipeline
from llama_index.extractors import TitleExtractor, SummaryExtractor
from llama_index.text_splitter import SentenceSplitter
from llama_index.schema import MetadataMode
def build_pipeline():
llm = OpenAI(model="gpt-3.5-turbo-1106", temperature=0.1)
transformations = [
SentenceSplitter(chunk_size=1024, chunk_overlap=20),
TitleExtractor(llm=llm, metadata_mode=MetadataMode.EMBED),
SummaryExtractor(llm=llm, metadata_mode=MetadataMode.EMBED),
OpenAIEmbedding(),
]
return IngestionPipeline(transformations=transformations)
from llama_index import SimpleDirectoryReader
documents = SimpleDirectoryReader("./data/paul_graham").load_data()
import time
times = []
for _ in range(3):
time.sleep(30) # help prevent rate-limits/timeouts, keeps each run fair
pipline = build_pipeline()
start = time.time()
nodes = await pipline.arun(documents=documents)
end = time.time()
times.append(end - start)
print(f"Average time: {sum(times) / len(times)}")
Average time: 106.17690531412761