diff --git a/README.md b/README.md index c3b18547..3338c8fb 100644 --- a/README.md +++ b/README.md @@ -228,9 +228,7 @@ You can also launch an HTTP service to generate images with xDiT. #### HunyuanVideo -- Supports `diffusers<=0.32.2` (breaking commit diffusers @ [8907a70](https://github.com/huggingface/diffusers/commit/8907a70a366c96b2322656f57b24e442ea392c7b)) - - For this constraint, supports `xfuser<=0.4.3post3` (breaking commit xDiT @ [0c20492](https://github.com/xdit-project/xDiT/commit/0c2049281d5f698622d006658fb99d1aa92a9d50)) - +- Supports `diffusers>=0.34.0` - Supports bs=1 only

📚 Develop Guide

diff --git a/examples/hunyuan_video_usp_example.py b/examples/hunyuan_video_usp_example.py index 9871c02c..c7a6cbb1 100644 --- a/examples/hunyuan_video_usp_example.py +++ b/examples/hunyuan_video_usp_example.py @@ -76,7 +76,7 @@ def new_forward( image_rotary_emb = self.rope(hidden_states) # 2. Conditional embeddings - temb = self.time_text_embed(timestep, guidance, pooled_projections) + temb,_ = self.time_text_embed(timestep=timestep, pooled_projection=pooled_projections, guidance=guidance) hidden_states = self.x_embedder(hidden_states) encoder_hidden_states = self.context_embedder(encoder_hidden_states, timestep, diff --git a/setup.py b/setup.py index 98daf8e9..19bda118 100644 --- a/setup.py +++ b/setup.py @@ -26,7 +26,7 @@ def get_cuda_version(): author_email="fangjiarui123@gmail.com", packages=find_packages(), install_requires=[ - "torch==2.4.1", + "torch>=2.4.1", "accelerate>=0.33.0", "transformers>=4.39.1", "sentencepiece>=0.1.99", @@ -37,7 +37,7 @@ def get_cuda_version(): ], extras_require={ "diffusers": [ - "diffusers>=0.31.0", # NOTE: diffusers>=0.32.0.dev is necessary for CogVideoX and Flux + "diffusers>=0.31.0", # NOTE: diffusers>=0.32.0.dev is necessary for CogVideoX and Flux. HunyuanVideo depends on diffusers>=0.34.0 ], "flash-attn": [ "flash-attn>=2.6.0", # NOTE: flash-attn is necessary if ring_degree > 1