Add support for Janus model from DeepSeek AI

[ighoshsubho]

Model description

Janus is an autoregressive framework that unifies multimodal understanding and generation. Unlike previous approaches that use a single visual encoder for both tasks, Janus decouples visual encoding into separate pathways while utilizing a unified transformer architecture for processing. This decoupling addresses the conflict between visual encoder roles in understanding and generation, enhancing flexibility and performance.

Key features:

• Unified framework for multimodal understanding and generation
• Decoupled visual encoding pathways
• Single, unified transformer architecture for processing
• Improved performance in multimodal understanding tasks
• Flexibility to select optimal encoding methods for each component

Open source status

• [X] The model implementation is available
• [X] The model weights are available

Provide useful links for the implementation

The Janus model is developed by DeepSeek AI. Here are the relevant links for implementation:

Paper: Janus: Bridging the Gap Between Multimodal Understanding and Generation GitHub repository: deepseek-ai/Janus