msglm makes it a little easier to create messages for language models like Claude and OpenAI GPTs.
Install the latest version from pypi
$ pip install msglm
To use an LLM we need to structure our messages in a particular format.
Here’s an example of a text chat from the OpenAI docs.
from openai import OpenAI
client = OpenAI()
completion = client.chat.completions.create(
model="gpt-4o",
messages=[
{"role": "user", "content": [{"type": "text", "text": "What's the Wild Atlantic Way?"}]}
]
)
Generating the correct format for a particular API can get tedious. The goal of msglm is to make it easier.
The examples below will show you how to use msglm for text and image chats with OpenAI and Anthropic.
For a text chat simply pass a list of strings and the api format (e.g. “openai”) to mk_msgs and it will generate the correct format.
mk_msgs(["Hello, world!", "some LLM response"], api="openai")
[
{"role": "user", "content": [{"type": "text", "text": "Hello, world!"}]},
{"role": "assistant", "content": [{"type": "text", "text": "Some assistant response"}]}
]
from msglm import mk_msgs_anthropic as mk_msgs
from anthropic import Anthropic
client = Anthropic()
r = client.messages.create(
model="claude-3-haiku-20240307",
max_tokens=1024,
messages=[mk_msgs(["Hello, world!", "some LLM response"])]
)
print(r.content[0].text)
from msglm import mk_msgs_openai as mk_msgs
from openai import OpenAI
client = OpenAI()
r = client.chat.completions.create(
model="gpt-4o-mini",
messages=[mk_msgs(["Hello, world!", "some LLM response"])]
)
print(r.choices[0].message.content)
For an image chat simply pass the raw image bytes in a list with your question to mk_msgs and it will generate the correct format.
mk_msg([img, "What's in this image?"], api="anthropic")
[
{
"role": "user",
"content": [
{"type": "image", "source": {"type": "base64", "media_type": media_type, "data": img}}
{"type": "text", "text": "What's in this image?"}
]
}
]
import httpx
from msglm import mk_msg_anthropic as mk_msg
from anthropic import Anthropic
client = Anthropic()
img_url = "https://www.atshq.org/wp-content/uploads/2022/07/shutterstock_1626122512.jpg"
img = httpx.get(img_url).content
r = client.messages.create(
model="claude-3-haiku-20240307",
max_tokens=1024,
messages=[mk_msg([img, "Describe the image"])]
)
print(r.content[0].text)
import httpx
from msglm import mk_msg_openai as mk_msg
from openai import OpenAI
img_url = "https://www.atshq.org/wp-content/uploads/2022/07/shutterstock_1626122512.jpg"
img = httpx.get(img_url).content
client = OpenAI()
r = client.chat.completions.create(
model="gpt-4o-mini",
messages=[mk_msg([img, "Describe the image"])]
)
print(r.choices[0].message.content)
To make your life a little easier msglm comes with api specific wrappers
for mk_msg and
mk_msgs.
For Anthropic use
from msglm import mk_msg_anthropic as mk_msg, mk_msgs_anthropic as mk_msgs
For OpenAI use
from msglm import mk_msg_openai as mk_msg, mk_msgs_openai as mk_msgs
msglm supports prompt caching for Anthropic models. Simply pass cache=True to mk_msg or mk_msgs.
from msglm import mk_msg_anthropic as mk_msg
mk_msg("please cache my message", cache=True)
This generates the expected cache block below
{
"role": "user",
"content": [
{"type": "text", "text": "Please cache my message", "cache_control": {"type": "ephemeral"}}
]
}
msglm supports text only models such as DeepSeek that use the OpenAI API format. Simply pass text_only=True to mk_msg or mk_msgs
from msglm import mk_msg_openai as mk_msg
mk_msg("please format my text only prompt", text_only=True)
This generates the expected cache block below
{
"role": "user",
"content": "please format my text only prompt"
}
We hope msglm will make your life a little easier when chatting to LLMs. To learn more about the package please read this doc.