We present PIQARD, a question answering system and library that allows users to easily ask open-domain questions and retrieve information from databases and other sources. The system uses advanced natural language processing techniques to understand and provide human-like answers. With its easy-to-use and integrate interface, PIQARD makes it simple for developers to add question answering capabilities to their projects.
- Create the virtual environment using
virtualenv
virtualenv env --python=python3.9- Activate the environment
- Windows
env\scripts\activate- Install the requirements in the environment:
pip install -r requirements.txt- Build the PIQARD library
python setup.py bdist_wheel- Install the PIQARD library
pip install dist/piqard-0.0.1-py3-none-any.whlTo reach the full potential of the PIQARD you should prepare several environment variables. Below we provide a list of all the variables and their description. You can set them up in cmd using:
set VARIABLE_NAME=VARIABLE_VALUEor in python files when needed:
import os
os.environ["VARIABLE_NAME"] = "VARIABLE_VALUE"To use GoogleCustomSearch class you have to set up the following environment variables:
GOOGLE_CUSTOM_SEARCH_API_KEY- API key for Google Custom Search API. To get your own API key you have to visit APIkey.GOOGLE_CUSTOM_SEARCH_ENGINE_ID- ID of the search engine. If you don't have access to google search engine you should visit SearchEngine and create a new one.
Models work thanks to huggingface Inference API.
To use BLOOM176bAPI or GPTj6bAPI class you have to set up the following environment variables:
HUGGINGFACE_API_TOKEN- API token for huggingface. To get your own API token you have to visit huggingface and create a new one.
In the assets directory, you'll find all the additional files needed for prompt templates and configs. We've also included two pre-prepared benchmark datasets with questions and documents. You can download them from the provided link
To get started with PIQARD, check out the example below.
This will give you a basic understanding of how to use the library.
For more advanced usage, refer to the other examples provided in the examples/base_usage.ipynb notebook.
from piqard.utils.prompt_template import PromptTemplate
from piqard.language_models import CohereAPI
from piqard.information_retrievers import WikiAPI
from piqard import PIQARD
prompt_template = PromptTemplate('./../assets/prompting_templates/like_chat_gpt/with_context.txt')
llm = CohereAPI(stop_token="\n")
ir = WikiAPI(k=10)
piqard = PIQARD(prompt_template=prompt_template,
language_model=llm,
information_retriever=ir)
result = piqard("Who is the current CEO of twitter?")
print(result['chain_trace'])PIQARD and extended strategies can be configured using a YAML config file. The file should be set up as a dictionary, with the keys being the names of the components and the values being the parameters for those components.
To help you get started, we've included an example config file for PIQARD. This will give you an idea of how to set up your own file to customize and optimize the tool for your specific needs.
- example config file:
./assets/configs/example.yaml
prompt_template: !PromptTemplate
template: assets/prompting_templates/like_chat_gpt/without_context.txt
language_model: !CohereAPI
stop_token: "\n"
information_retriever: !AnnoyRetriever
k: 2
database: "openbookqa"
database_path: "./../assets/benchmarks/openbookqa/corpus.jsonl",
database_index: "./../assets/benchmarks/openbookqa/corpus_annoy_index_384.ann"- loading PIQARD from config file:
from piqard.PIQARD_loader import PIQARDLoader
piqard = PIQARDLoader().load("./assets/configs/example.yaml")
result = piqard("Who is the current CEO of twitter?")
print(result['chain_trace'])We create user interface in React, which allow playing around custom configurations, prompts and questions.
Additionally, we propose example configuration which use SelfAware strategy.
- PIQARD API server
- Create
config.jsonfile with required API keys and configuration inpiqard_apidirectory.
{
"COHERE_API_KEY": "",
"HUGGINGFACE_API_KEY": "",
"GOOGLE_CUSTOM_SEARCH_API_KEY": "",
"GOOGLE_CUSTOM_SEARCH_ENGINE_ID": "",
"OPEN_SYSTEM_CONFIG": "./assets/configs/self_aware_config_piqard_react.yaml"
}- Run the following command from project root directory:
python piqard_api/api.py- PIQARD APP
- Go to
piqard_appdirectory and run the following commands:
npm install
npm startPIQARD is a project created by Dawid Plaskowski, Mateusz Politycki, Marcin Korcz and Alex Terentowicz from Poznan University of Technology. To see our thesis and learn more about the project, visit our website.
If you have any questions, feel free to contact us.