Catalog PyPI
PACKAGE EXPLORERView in registry

ai-factory-model

PyPI

AI Library to authenticate and manage LLM models and vector databases RAGs

Versions collected1
Direct dependencies24From release 0.0.12
Last observed21 Oct, 10:40 UTC

README

Version 0.0.12 documentation

ai-factory-model

PyPI version Supported Python Versions Build Status License Coverage Status PyPI - Downloads

Description

ai-factory-model is a modular Python library aimed at integrating with multiple language models (LLMs), cloud providers, and auxiliary utilities for development and infrastructure.

This library is designed to facilitate interaction with LLMs from OpenAI, Azure, Google, and Ollama, also integrating authentication, external configuration, Jinja2 templates, and reusable components.

Features

  • Support for multiple LLM models:
    • OpenAI (chat and embeddings)
    • Azure OpenAI
    • Google Generative AI
    • Ollama
    • LangChain and variants
  • Configuration modules (decouple, YAML)
  • Authentication via Azure Identity
  • Content generation via Jinja2 templates
  • Clear separation of responsibilities with modules such as:
    • logger
    • security
    • auth_clients
    • model_* (interfaces for different LLMs)

Installation

From PyPI:

pip install ai-factory-model

Optional modules:

pip install ai-factory-model[google_genai]
pip install ai-factory-model[community]
pip install ai-factory-model[ollama]
pip install ai-factory-model[cohere]
pip install ai-factory-model[pgvector]

Setup

To use the model factory, you need to define a series of environment variables that allow connection to the various model hosting services:

AZURE_TENANT_ID = <id_tenant_azure>
AZURE_CLIENT_ID = <id_client_azure>
AZURE_CLIENT_SECRET = <secret_passphrase_azure_client>
AZURE_TOKEN_URL = <azure_url_token>

For enhanced security, there is a connection to KeyVault. To define the connection to the corresponding key store, use:

KV_NAME = <kv_name>
KV_TENANT_ID = <id_kv_tenant>
KV_CLIENT_ID = <id_kv_client>
KV_SECRET = <secret_passphrase_kv>

With the KeyVault connection established, the values to be retrieved from the key store should be specified using the following nomenclature:

VARIABLE_SECRET = kv{name-of-secret-at-kv}

For example:

# Transition from having the secret in raw form
AZURE_CLIENT_SECRET = <secret_passphrase_azure_client>

# To retrieving it from the KV
AZURE_CLIENT_SECRET = kv{<name_secret_azure_client>}

Additionally, if you have a file containing the various model configurations you wish to use, you should specify it with the corresponding variable.

MODELS_CONFIG_FILE = <path_to_models_declarations_file>

Basic usage

Using prompt:

from ai_factory_model import ModelFactory

model = ModelFactory.get_model("azai_gtp4o")
params = ["Eres un guía turístico", "¿Dónde está Plasencia?"]

response = model.prompt(params=params)

print(type(response))
# Output:
# <class 'str'>

print(response)
# Output:
# Plasencia es una ciudad situada en la comunidad autónoma de Extremadura, en el oeste de España. Se encuentra en la provincia de Cáceres, a orillas del río Jerte. Plasencia está aproximadamente a unos 80 kilómetros al norte de la ciudad de Cáceres y a unos 250 kilómetros al oeste de Madrid. Es conocida por su casco histórico, que incluye la Catedral de Plasencia, y por su cercanía al Valle del Jerte, famoso por sus cerezos en flor.

Using langchain instance:

from ai_factory_model import ModelFactory

model = ModelFactory.get_model("azai_gtp4o")
params = ["Eres un guía turístico", "¿Cuál es la capital de España?"]

response = model.get_client.invoke([
    {"role": "system", "content": params[0]},
    {"role": "user", "content": params[1]}
])

print(type(response))
# Output:
# <class 'langchain_core.messages.ai.AIMessage'>

print(f"{response.content}")
# Output:
# La capital de España es Madrid. Es una ciudad vibrante y llena de historia, conocida por su rica cultura, su arquitectura impresionante y su animada vida nocturna. Además, Madrid alberga importantes museos como el Museo del Prado y el Museo Reina Sofía, así como el Palacio Real y el Parque del Retiro.

Render a template:

from ai_factory_model import ModelFactory, SEP_PATTERN

model = ModelFactory.get_model("azai_gtp4o")
params = {"system": "Eres un guía turístico", "user": "¿Qué visitar en Mérida de Extremadura?"}

template_content = (
    f"{{{{ system }}}}"
    f"{SEP_PATTERN}"
    f"{{{{ user }}}}"
)
prompt_template = Template(template_content)

response = model.prompt_render(
    template=prompt_template,
    params=params,
    sep_pattern=SEP_PATTERN
)

print(type(response))
# Output:
# <class 'str'>

print(f"{response}")
# Output:
# ¡Mérida es una ciudad fascinante llena de historia y patrimonio! Es conocida por su impresionante legado romano, ya que fue una de las ciudades más importantes de la antigua Hispania. Aquí tienes una lista de los lugares imprescindibles que deberías visitar en Mérida: [...]

Project structure

ai_factory_model/
├── ai_factory_model/
│   ├── __init__.py
│   ├── config/
│   ├── llm/
│   ├── logger/
│   ├── security/
│   ├── vectordb/
├── pyproject.toml
├── requirements.txt
└── README.md

Dependencies

This package requires the followind external libraries:

  • python-decouple
  • PyYAML
  • openai
  • jinja2
  • azure-core
  • azure-identity
  • azure-keyvault-secrets
  • langchain
  • langchain-openai
  • langchain-azure-ai
  • azure-search-documents

With optional extra libraries

pip install ai-factory-model[google_genai]
  • langchain-google-genai
pip install ai-factory-model[community]
  • langchain-community
pip install ai-factory-model[ollama]
  • langchain-ollama
pip install ai-factory-model[cohere]
  • langchain-cohere
pip install ai-factory-model[pgvector]
  • psycopg[binary]

Requirements

  • Python 3.12 o superior
  • Credential access/API keys to your needed providers (OpenAI, Azure, etc.)

Downloads & popularity

Awaiting observations

No metric observations yet

Download counts and popularity will appear here as the collector receives them.

Versions 1

Collected releases
VersionPublishedDirect dependenciesOpen version
0.0.123 May 202624

Package advisories

0
No matching advisories collected

This describes our collected data; it is not a security assessment.

Metadata history

Meaningful changes
Current collected state Current

Observed since 21 Oct, 10:40 UTC

No superseded metadata states collected. Unchanged observations do not create new revisions.

Record identity

package:f20096bf6f17118b4b3b2e7c2d80c82d2a7cf4ff5d64bf2cf57e265497de56cb