GIMKit

Guided Infilling Modeling Toolkit — structured text generation and information extraction using language models.

Write a template with typed placeholders. The LLM fills them in. Get structured, named results back.

from gimkit import guide as g

query = f"""Extract from: "Hi, I'm John Smith, reach me at john@gmail.com"

Name: {g.person_name(name="name")}
Email: {g.e_mail(name="email")}"""

result = model(query, use_gim_prompt=True)
result.tags["name"].content   # → "John Smith"
result.tags["email"].content  # → "john@gmail.com"

Installation

pip install gimkit

For vLLM support:

pip install gimkit[vllm]

What Can You Do With GIMKit?

GIMKit is a general-purpose information extraction framework. Write a natural-language template with embedded tags, and the model extracts structured data from any text.

See the Classic IE Use Cases, Privacy and PII Use Cases, and Other Use Cases pages for full examples.

Why GIMKit?

• Template-driven — describe what you want in natural language, not label lists
• Format control — regex constraints, enumerated choices, type-safe tags
• Named access — results are keyed by field name, not token positions
• Small-model friendly — works with compact open-source models (4B+)
• Multiple backends — OpenAI, vLLM (server and offline)

Quick Start

from openai import OpenAI
from gimkit import from_openai, guide as g

client = OpenAI()
model = from_openai(client, model_name="gpt-4")

# Simple extraction
result = model(f"Hello, {g(desc='a single word')}!", use_gim_prompt=True)
print(result)  # Hello, world!

# Structured form
query = f"""
Name: {g.person_name(name="name")}
Email: {g.e_mail(name="email")}
Favorite color: {g.select(name="color", choices=["red", "green", "blue"])}
"""
result = model(query, use_gim_prompt=True)
print(result.tags["name"].content)
print(result.tags["email"].content)
print(result.tags["color"].content)