Promptify Documentation

Promptify provides several key capabilities for NLP tasks:

• Prompt Generation: Tailor prompts for state-of-the-art generative models.
• Unified Architecture: Utilizes a Prompter, Model, and Pipeline solution.
• Detailed Output Logs: Generates comprehensive structured JSON output within a log folder.
• Wider Model Support: Supports models from OpenAI, Azure, Cohere, Anthropic, Huggingface, and more.
• Robust Parser: Handles incomplete or unstructured JSON outputs from LLMs.
• Ready-Made Jinja Templates: Includes templates for NER, Text Classification, QA, Relation-Extraction, Tabular data, etc.
• Rich Customization: Adapt prompts for specific needs like creative writing or summaries.

Install Promptify using pip. Requires Python 3.9+.

To install the base package:

pip install promptify

To install from the GitHub repository:

pip install git+https://github.com/promptslab/Promptify.git

To install with support for evaluation metrics:

pip install promptify[eval]

Promptify uses LiteLLM as a backend, allowing you to use any provider by simply changing the model string. Supported providers include OpenAI, Anthropic, Google, Ollama, Azure, and 100+ others.

ner_openai = NER(model="gpt-4o-mini")
ner_claude = NER(model="claude-sonnet-4-20250514")
ner_local  = NER(model="ollama/llama3")

Install the stable version of Promptify using pip. Requires Python 3.7+ and openai 0.25+.

$ pip install promptify

Promptify uses OpenAI models (defaulting to text-davinci-003). To use these models, you must provide an OPENAI_API_KEY. You can initialize the Prompter by passing an OpenAI client instance configured with your API key.

$ model = OpenAI(api_key="YOUR_API_KEY")
$ nlp_prompter = Prompter(model)

You can install Promptify directly from the GitHub repository or perform an editable installation for development purposes.

• Direct from GitHub: Installs the package from the repository.
• Editable Install: Use -e . to allow modifications to the source files.
• Development Dependencies: Use requirements.txt to install optional dependencies and tools used for development (e.g., unit testing).

The Pipeline API allows you to immediately perform NLP tasks (like Named Entity Recognition) using an LLM model and a prompt template.

To use it, you need to initialize a model (e.g., OpenAI, HubModel, or Azure), a Prompter with a template (e.g., 'ner.jinja'), and then wrap them in a Pipeline object. Use the .fit() method to run inference on your text.

Key Parameters for .fit():

• sentence: The input text string.
• domain: (Optional) The domain of the text (e.g., "medical").
• labels: (Optional) A list of labels. If omitted, the model automatically infers labels from the text. If provided, the model uses your specific labels.

For production environments, instead of using the free Inference API, use a dedicated Hugging Face Inference Endpoint URL. You can switch to a production-ready setup by passing the endpoint URL and your api_key directly to HubModel.

model = HubModel("https://endpoint-id.region.vendor.endpoints.huggingface.cloud", api_key="hf_***")

To use promptify, first initialize a model provider (e.g., OpenAI) with your API key, then pass that model instance to a Prompter to enable prompt generation.

from promptify import OpenAI
from promptify import Prompter

model = OpenAI(api_key="")
nlp_prompter = Prompter(model)

To use promptify for NLP tasks, import the OpenAI model class and the Prompter class. Initialize the OpenAI model with your API key, then pass that model instance into the Prompter to create an NLP-capable prompter.

from promptify.models.nlp.openai_model import OpenAI
from promptify.prompts.nlp.prompter import Prompter

model = OpenAI(api_key="YOUR_API_KEY")
nlp_prompter = Prompter(model)