MageFlow 5B

Only the model structure is provided. The duplicated blocks accept LFM2.5 text encoder inputs, while the rest of the original blocks accept Qwen encoder inputs.

The output should be the same as the MageFlow model's on initialization.

The number of trainable layers has increased to 20.

Keywords

layer expansion, duplicated layers
shrinking AdaLN

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support