[FEAT] Slice text embedding in MM-DiT

xdit-project / xDiT

xDiT: A Scalable Inference Engine for Diffusion Transformers (DiTs) with Massive Parallelism

Apache License 2.0

726 stars 56 forks source link

Closed xibosun closed 3 days ago

xibosun commented 4 days ago

This pr implements the slicing both text and image/video embedding in MM-DiTs, including SD3, Flux, and CogVideo.