ezllm-python

The goal of EzLLM is to provide the simplest interface for creating LLM based applications

EzLLM simplifies the process of

1. File parsing
2. Storage
3. Retrieval
4. LLM optimization
5. LLM output parsing

Home Page --- Docs --- Developer Console

for more information

Quick Start

Install

pip install ezllm

Authenticate the SDK

1. go to the API keys page in the developer dashboard
2. create an API key
3. save the API key and API secret to a .env file

EZLLM_KEY=abc
EZLLM_SECRET=xyz

by saving it to a .env file, ezllm will authenticate itself

alternatively you can pass the key and secret to ezllm to create a client

from ezllm import Client

client = Client(key='...', secret='...')

Upload Documents

import ezllm
# or
from ezllm import Collection

col = ezllm.Collection("documentation")



# Upload a single file
file = open('./file.pdf', 'r')
await col.upload(file)

# Upload by path
await col.upload(path='./file.pdf')
# Or
await col.upload(path='./file.pdf', name="My File Name", type='pdf')

Retrieve Relevant Documents

for more information on document retrieval click here

response = col.search('search query').get()

print(response)

# Output
SearchResponse(
    costs=ResponseCosts(
        total_cost=10.0,
        costs=[
            ResponseCost(
                cost=10,
                date='2024-01-12T05:38:08.266074',
                step='retrieve',
                usage=1,
                limitType='search',
                limitId='search',
                desc=None,
                reportedCost=None
            )
        ]
    )
    doc_groups=SearchDocOutputGroups(
        groups=[
            SearchDocOutputGroup(
                id="collection_id",
                type="collection",
                data=SearchResponseDocs(
                    docs=[
                        SearchResponseDoc(
                            name="Test File",
                            SubDocs=1
                        )
                    ]
                )
            )
        ]
    )
)

# it's heavily nested from the aggregation step
# so you can reference all of the underlying docs with .docs
print(response.docs)

# Output
ResponseDocs(
    docs=[
        SearchResponseDoc(
            name="Test File"
            SubDocs=1
        )
    ]
)

for doc in response.docs:
    print(doc)

    # Output
    SearchResponseDoc(
        name="Test File"
        SubDocs=1
    )

Run Methods against a Retrieval

from ezllm.methods import QAMethod

col.search('search query').run(QAMethod(questions=["Is this document relevant to my research in XYZ"]))

# Output
QAMethodResponse(
    costs=ResponseCosts(
        total_cost=0.08994,
        costs=[
            ResponseCost(cost=0.01061, date='2023-12-27T18:11:18.722257', step='method', usage=835, limitType='llm', limitId='gpt-4-1106-preview', desc=None, reportedCost=None)
            ...
        ]
    )
    # add the include_docs=True argument to return docs
    doc_groups=DocOutputGroups(
        groups=[

        ]
    )
    output_groups=QAOutputGroups(
        groups=[
            QAOutputGroup(
                type="document"
                id="658b4e441f41ead7592562de"
                output=QAOutputData(
                    data=[
                        {
                            "question": "Who are the authors?",
                            "answer": "Sylvain Burns, Joseph Brehaut, Jamie Britton",
                            "relevant": 100
                        },
                        {
                            "question": "What organizations is this document from?",
                            "answer": "Journal of Professional Engineering, University of Ottawa",
                            "relevant": 100
                        }
                    ]
                )
            )
            ...
        ]
    )
)

Tests

tests are located in the /test directory and can be a good resource for understanding how to interact with entities in the SDK.

Read more about tests here

to run tests, you need to provide a .env file with the following values.

EZLLM_ACCESS_KEY = access_key
EZLLM_SECRET = secret
EZLLM_API_URL = https://api.ezllm.io # optional
EZLLM_RUN_URL = https://run.ezllm.io # optional

Deploy

pip show ezllm

# 1. Build
python3 setup.py sdist bdist_wheel

2. # Install
## local
pip3 install --upgrade dist/ezllm-0.1-py3-none-any.whl

## test-pypi
pip3 install --index-url https://test.pypi.org/simple/ ezllm

## pypi
pip3 install ezllm

## Uninstall (necessary to test each version)
pip3 uninstall ezllm 

3. # Upload
## test-pypi
python3 -m twine upload --repository testpypi dist/*

## pypi
python3 -m twine upload dist/*