llm-factory

Mevcutta 0.X.Y versiyon llm factory sorunsuz çalışıyor ve değişiklik gerekmiyorsa; breaking change olmaması için güvenlik, vs gibi gereksinim olmadığı sürece versiyon ilerletmeyelim.
- örn: (tcell_llm_factory==0.2.10)
Query Engine implement ederken cloud ve onprem fark etmeksizin "response" ve "consumption" return edelim. Onpremler için (hic dönülmemesi yerine) "consumption" None döndürülür.
tests/test_integration.py altında tüm query engine'ler tek tip input/output akışı gerçeklemesi test edilir.
Async runtime'lar için LLMFactry & AsyncLLMFactory class'ları ayrı ayrı kullanılır ihtiyaca göre. Async tarafında olmayan QEngine ilgili developer implement edebilir.

Build

python -m pip install -U wheel setuptools build
increment project_id on pyproject.toml
archrose (for mac users)
python -m build

Upload

artifactory urls:
- local-pypi-dist-dev/com/turkcell/sensai/llm_factory
- https://artifactory.turkcell.com.tr/ui/repos/tree/General/local-pypi-dist-dev/com/turkcell/sensai/llm_factory/
Deploy tıkla
Multiple Deploy seçerek yükle yeni build dosyaları
target path güncelle:
- com/turkcell/sensai/llm_factory/0.2.0

Improvement Notes (TODO):

azure provider altında langchain'den azure class import edilmiş; yukarıda openai'dan farklı frameworkler eşitlenmeli mi?
provider'ların hepsi kendi içinde langchain, vs llm dev frameworkü yönetsin. Biz karışmayalım. LLMFactory pyproject.yml'dan dependency olarak çıkartmalı.
partial pip installation için (örn: pip install llm_factory[ollama]) dependency importlar da dinamik olmalı (aşağıda örnek implementasyon var fakat paralelde best practice literatür taraması yapalım)

import os
from typing import List, Optional
from importlib.util import find_spec

# Import core dependencies
from . import query_engine as qe

try:
    from ..errors import LLMFactoryError
except ImportError:  # for debug runtime
    from errors import LLMFactoryError

# Initialize variables to hold imported classes
ChatOpenAI = None
AzureChatOpenAI = None
OllamaClient = None
OllamaAsyncClient = None
AzureOpenAI = None
AsyncAzureOpenAI = None

# Helper function to check if a module is installed
def is_package_installed(package_name: str) -> bool:
    return find_spec(package_name) is not None

# Dynamic imports for Azure/OpenAI
if is_package_installed("langchain_openai"):
    from langchain_openai import ChatOpenAI, AzureChatOpenAI

if is_package_installed("openai"):
    from openai import AzureOpenAI, AsyncAzureOpenAI

# Dynamic imports for Ollama
if is_package_installed("ollama"):
    from ollama import Client as OllamaClient, AsyncClient as OllamaAsyncClient

# You might want to add error classes for missing dependencies
class DependencyNotInstalledError(LLMFactoryError):
    """Raised when a required dependency is not installed."""
    pass

# Then in your code, you can check before using these classes
def get_azure_client(*args, **kwargs):
    if ChatOpenAI is None or AzureOpenAI is None:
        raise DependencyNotInstalledError(
            "Azure/OpenAI support requires additional dependencies. "
            "Please install with: pip install 'llm-factory-tcell[azure]'"
        )
    # Your Azure client initialization code here
    return AzureOpenAI(*args, **kwargs)

def get_ollama_client(*args, **kwargs):
    if OllamaClient is None:
        raise DependencyNotInstalledError(
            "Ollama support requires additional dependencies. "
            "Please install with: pip install 'llm-factory-tcell[ollama]'"
        )
    # Your Ollama client initialization code here
    return OllamaClient(*args, **kwargs)

# Example usage in a factory method
def create_llm_client(provider: str, *args, **kwargs):
    if provider == "azure":
        return get_azure_client(*args, **kwargs)
    elif provider == "ollama":
        return get_ollama_client(*args, **kwargs)
    # Add other providers as needed
    else:
        raise ValueError(f"Unsupported provider: {provider}")

[project]
name = "llm-factory-tcell"
version = "0.2.12"
authors = [
  { name="Mithat Sinan Ergen", email="mithat.ergen@turkcell.com.tr" },
  { name="Emre Çalışır", email="emre.calisir@atmoswere.turkcell.com.tr" },
  { name="Umut Alihan Dikel", email="alihan.dikel@turkcell.com.tr" },
  { name="Mahmut Yılmaz", email="yilmaz.mahmut@turkcell.com.tr" }
]
description = "LLM Factory package"
readme = "README.md"
requires-python = ">=3.11"
classifiers = [
    "Programming Language :: Python :: 3",
    "Operating System :: OS Independent",
]

# Core dependencies that are always needed
dependencies = [
    "fastapi>=0.110.0",
    "langchain>=0.2.0",
    "langchain-community>=0.2.0",
    "langchain-core>=0.2.0",
    "pydantic>=2.0.0",
    "SQLAlchemy>=2.0.0",
    "loguru>=0.7.0",
    "httpx>=0.24.0",
    "rich>=13.0.0",
]

[project.optional-dependencies]
dev = [
    "pytest>=8.0.0",
    "pytest-asyncio>=0.23.0",
    "pytest-cov>=5.0.0",
]
ollama = [
    "ollama>=0.4.6",
]
vllm = [
    "vllm>=0.2.0",
]
azure = [
    "langchain-openai>=0.1.0",
    "openai>=1.0.0",
    "tiktoken>=0.7.0",
]

[build-system]
requires = ["setuptools>=61.0", "wheel"]
build-backend = "setuptools.build_meta"

[tool.setuptools.packages.find]
where = ["src"]

[build-system]
requires = ["setuptools>=40.8.0"]
build-backend = "setuptools.build_meta"
 
[tool.setuptools]
packages = ["your_library"]
 
[tool.setuptools.dependencies]
# Required dependencies for your library
install_requires = [
    "requests>=2.25.0"
]
 
[tool.setuptools.extras]
# Optional dependencies (extras)
data-analysis = ["pandas", "numpy"]
web-scraping = ["beautifulsoup4", "requests"]
ml = ["scikit-learn", "tensorflow"]
dev = ["black", "flake8", "pre-commit"]
testing = ["pytest", "coverage"]

README.md Unescape Escape

Notes to Developer

Build

Upload

Improvement Notes (TODO):

README.md