chatmux 0.1.3

Convert between common LLM chat schemas

chatmux

Convert to and from common LLM chat schemas.

All implementations are built on pydantic base models to be network ready.

Installation

pip install chatmux

Usage

Fully OpenAI compatible pydantic models

from chatmux.openai import UserMessage, TextContentPart, ImageContentPart, ImageUrl

# Example using Pydantic models (more robust):
user_message = UserMessage(
    role="user",
    content=[
        TextContentPart(type="text", text="Describe this image:"),
        ImageContentPart(
            type="image_url",
            image_url=ImageUrl(url="https://example.com/image.jpg")
        )
    ]
)
openai_messages = [user_message.model_dump(exclude_none=True)] # Convert model to dict

Conversion methods to and from other schemas

from chatmux.convert import oai_to_qwen

qwen_messages = oai_to_qwen(openai_messages)
print(qwen_messages)
# Expected Output (structure based on oai_to_qwen):
# [
#     {
#         'role': 'user',
#         'content': [
#             {'type': 'text', 'text': 'Describe this image:'},
#             {'type': 'image', 'image': 'https://example.com/image.jpg'}
#         ]
#     }
# ]

Supported Conversions (as implemented in src/chatmux/convert.py):

• OpenAI-like -> Unsloth Inference: Converts multimodal messages, downloads images, and formats for Unsloth. Returns new schema and PIL Images. (convert_to_unsloth_inference)
• OpenAI-like -> Unsloth: Converts messages (potentially from JSONL), handling image URLs/base64, embedding PIL Images. (oai_to_unsloth)
• OpenAI-like -> Qwen: Converts multimodal messages, keeping image URLs as strings but adjusting the content structure. (oai_to_qwen)

Contributing

Please open an issue before submitting a PR.

License

MIT