genai-forge 0.1.13

Generative AI tools from ToolForge-AI organization

genai-forge

Lightweight utilities to call LLMs and parse their outputs with Pydantic.

Installation

Install from PyPI:

pip install genai-forge

Requirements

• Python 3.10+
• An OpenAI API key in your environment:
  • Create a .env file at your project root with:

OPENAI_API_KEY=sk-...

genai_forge will load .env automatically via python-dotenv when using the OpenAI provider.

Quick start

from genai_forge.llm import create_llm
from genai_forge.prompting import PromptTemplate

template = PromptTemplate(
    system="You are a concise expert assistant.",
    template="Generate one actionable tip.\nAudience: {audience}\nTime: {time}",
)

llm = create_llm("openai:gpt-4o-mini", temperature=0.2)

# Chain: query | template | llm
query = "Provide a short productivity tip."
chain = query | template | llm
print(chain({"audience": "Backend Python developer", "time": "30 minutes"}))

Parsing structured outputs with Pydantic

Use PydanticOutputParser to have the LLM return valid JSON that is validated into a Pydantic model. When you add a parser to the chain, the parser’s format instructions are automatically appended to the prompt; you do not need to place {instructions} in your template.

from typing import List
from pydantic import BaseModel
from genai_forge.llm import create_llm
from genai_forge import PydanticOutputParser
from genai_forge.prompting import PromptTemplate

class CityPlan(BaseModel):
    city: str
    attractions: List[str]
    days: int

template = PromptTemplate(
    system="You are a helpful travel planner.",
    template="Create a city plan.\nCity: {city}\nDays: {days}",
)

llm = create_llm("openai:gpt-4o-mini", temperature=0.1)
parser = PydanticOutputParser(CityPlan)

# Chain: query | template | llm | parser
query = "Create a 2-day city plan for Kyoto."
chain = query | template | llm | parser
result = chain({"city": "Kyoto", "days": 2})  # -> CityPlan
print(result)

What chaining does

The | operator builds a Chain:

• template | llm produces a function that formats the prompt (system + user) and calls the LLM.
• template | llm | parser also injects the parser's format instructions automatically into the prompt and validates the output.

Output parser: how it works

PydanticOutputParser:

• Accepts tolerant output formats (e.g., extra text or ```json fences).
• Extracts JSON from the text (tries direct parse, then finds the first balanced {...} block).
• Validates the JSON against your Pydantic model.
• Provides get_format_instructions(); the Chain appends it for you when a parser is present.

API surface:

from genai_forge import PydanticOutputParser, BaseOutputParser, OutputParserException

Key methods:

• get_format_instructions() -> str: Include in your prompt.
• parse(text: str) -> BaseModel: Parse and validate output.

Provider configuration

By default, models without an explicit provider use openai. You can specify:

create_llm("openai:gpt-4o-mini")
# or
create_llm("gpt-4.1", provider="openai")

The OpenAI provider reads the key from OPENAI_API_KEY (or api_key argument).

Running the example

An example.py is included at the repo root. It demonstrates:

• A chain: user input → PromptTemplate (with system prompt) → LLM → PydanticOutputParser
• Automatic format instructions injection

Ensure you have a .env with OPENAI_API_KEY, then:

python example.py

License

See LICENSE.