Installation

pip install llmloader

Or install from GitHub directly:

pip install git+https://github.com/rbturnbull/llmloader.git

Usage

Load the LLM with the llmloader.load function. e.g.

import llmloader

llm = llmloader.load("gpt-6-luna")
result = llm.invoke("Write me a haiku about love")

llm = llmloader.load("claude-sonnet-5")
result = llm.invoke("Write me a haiku about love")

llm = llmloader.load("grok-4-latest")
result = llm.invoke("Write me a haiku about love")

llm = llmloader.load("mistral-small-latest")
result = llm.invoke("Write me a haiku about love")

llm = llmloader.load("meta-llama/Llama-3.3-70B-Instruct")
result = llm.invoke("Write me a haiku about love")

To pass an image, it needs to be base64 encoded, and reformatted with LLMWrapper.

import base64
from langchain_core.messages import HumanMessage
from llmloader.wrappers import LLMWrapper

with open("path/to/image.png", "rb") as image_file:
    encoded_string = base64.b64encode(image_file.read()).decode("utf-8")

formatted_image = LLMWrapper.format(llm, "image", {
    "data": encoded_string,
    "mime_type": "image/png"
})

message = HumanMessage(content=[{"type": "text", "text": "Here is an image for you:"}, formatted_image])

result = llm.invoke(message)

Get the token usage

from llmloader.wrappers import LLMWrapper

result = llm.invoke(message)

count = LLMWrapper.get_token_count(result, record="optional_path/to/record.yaml")

CLI

You can test out prompts and models on the command line. Make sure you have your API keys set in your environment or add the key with the --api-key flag.

llmloader "Write me a haiku about love" --model gpt-5-mini
llmloader "Write me a haiku about love" --model gpt-6-luna
llmloader "Write me a haiku about love" --model claude-sonnet-5
llmloader "Write me a haiku about love" --model grok-4-latest
llmloader "Write me a haiku about love" --model mistral-small-latest
llmloader "Write me a haiku about love" --model gemini-3-pro-preview
# Using OpenRouter
llmloader "Write me a haiku about love" --model openai/gpt-5-mini
# Local deployment models
llmloader "Write me a haiku about love" --model meta-llama/Meta-Llama-3-8B-Instruct
llmloader "Write me a haiku about love" --model meta-llama/Llama-3.3-70B-Instruct
llmloader --help

Environment Variables

You can pass an API key for the model provider using the command line flag --api-key, kwarg api_key=..., or by setting the appropriate environment variable as described below.

Testing

pytest -m manual

pytest -m manual tests/test_manual.py::test_name

Credit

• Robert Turnbull (Melbourne Data Analytics Platform, University of Melbourne)

• James Quang (Melbourne Data Analytics Platform, University of Melbourne)