Owner

thu-ml

2 indexed repositories · View on GitHub

TurboDiffusion
TurboDiffusion: 100–200× Acceleration for Video Diffusion Models

Python · 3735 stars
Causal-Forcing
[ICML 2026] Official codebase for "Causal Forcing: Autoregressive Diffusion Distillation Done Right for High-Quality Real-Time Interactiv...

Python · 969 stars

Beginner friendly Everything

Open All

Compute needed for the Causal Forcing Chunkwise setting Open

question

Difficulty 5/5 Over a week Newbie friendliness 35/100

thu-ml/Causal-Forcing#56 · 4 comments · Opened Aug 27, 2026
Higher saturation and contrast in Causal ODE initialization Open

question

Difficulty 4/5 3-5 days Newbie friendliness 35/100

thu-ml/Causal-Forcing#54 · 1 comment · Opened Aug 14, 2026
Problems during the DMD‑phase training Open

question

Difficulty 5/5 Over a week Newbie friendliness 25/100

thu-ml/Causal-Forcing#53 · 7 comments · Opened Aug 14, 2026
Is progressive blurring after ~150 frames expected after the teacher-forcing stage? Open

question

Difficulty 5/5 Over a week Newbie friendliness 35/100

thu-ml/Causal-Forcing#52 · 2 comments · Opened Aug 11, 2026
1 chunk Causal Forcing ++ Training Open

question

Difficulty 5/5 Over a week Newbie friendliness 35/100

thu-ml/Causal-Forcing#51 · 11 comments · Opened Jul 29, 2026
How was the 80K training dataset for Causal Forcing++ constructed? Open

question

Difficulty 5/5 Over a week Newbie friendliness 35/100

thu-ml/Causal-Forcing#49 · 1 comment · Opened Jul 17, 2026
VBench Evaluation Format Open

question

Difficulty 4/5 3-5 days Newbie friendliness 35/100

thu-ml/Causal-Forcing#48 · 4 comments · Opened Jul 5, 2026
DreamX-World: A General-Purpose Interactive World Model Open

Difficulty 5/5 Over a week Newbie friendliness 20/100

thu-ml/TurboDiffusion#135 · Opened Jun 18, 2026
AutoRegressive Diffusion Training Results Open

question

Difficulty 5/5 Over a week Newbie friendliness 20/100

thu-ml/Causal-Forcing#46 · 26 comments · Opened Jun 17, 2026
请求新增 RTX 5090 平台 FP8 推理优化分支 Open

Difficulty 5/5 Over a week Newbie friendliness 28/100

thu-ml/TurboDiffusion#131 · Opened Jun 10, 2026
Question bout Autoregressive Diffusion Training Open

good first issue question solved

Difficulty 4/5 3-5 days Newbie friendliness 35/100

thu-ml/Causal-Forcing#45 · 7 comments · Opened Jun 5, 2026
关于merge的一些疑问 Open

Difficulty 5/5 Over a week Newbie friendliness 20/100

thu-ml/TurboDiffusion#130 · 7 reactions · Opened Jun 3, 2026
About the loss curves. Open

question

Difficulty 1/5 Under an hour Newbie friendliness 35/100

thu-ml/Causal-Forcing#42 · 3 comments · Opened Jun 2, 2026
能否支持多卡并行推理呢？ Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

thu-ml/TurboDiffusion#128 · Opened May 29, 2026
Causal CD leads to static and blurry results Open

question

Difficulty 4/5 3-5 days Newbie friendliness 35/100

thu-ml/Causal-Forcing#41 · 8 comments · Opened May 29, 2026
请问当前支持在NPU安装，推理吗？ Open

Difficulty 5/5 Over a week Newbie friendliness 20/100

thu-ml/TurboDiffusion#127 · Opened May 27, 2026
Release additional Causal Forcing++ artifacts on Hugging Face Open

Difficulty 4/5 3-5 days Newbie friendliness 38/100

thu-ml/Causal-Forcing#38 · Opened May 15, 2026
slow for "python utils/merge_and_get_clean.py" Open

question

Difficulty 3/5 1-2 days Newbie friendliness 38/100

thu-ml/Causal-Forcing#37 · 15 comments · Opened May 13, 2026
Question about the three schedulers from generator and real/fake teacher Open

good first issue question

Difficulty 4/5 3-5 days Newbie friendliness 45/100

thu-ml/Causal-Forcing#35 · 5 comments · Opened May 6, 2026
Question about Instruction following Open

question

Difficulty 4/5 3-5 days Newbie friendliness 35/100

thu-ml/Causal-Forcing#34 · 1 comment · Opened May 6, 2026
Strategy for applying rCM with Autoregressive Student Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

thu-ml/TurboDiffusion#126 · 1 comment · Opened Apr 26, 2026
Open source the Causal CD ckpt? Open

solved

Difficulty 5/5 Over a week Newbie friendliness 25/100

thu-ml/Causal-Forcing#32 · 1 comment · Opened Apr 24, 2026
Temporal inconsistency when decoding chunk-wise ODE latents Open

question

Difficulty 4/5 3-5 days Newbie friendliness 35/100

thu-ml/Causal-Forcing#31 · 1 comment · Opened Apr 19, 2026
Causal-Forcing on Wan2.2-TI2V-5B? Open

question

Difficulty 5/5 Over a week Newbie friendliness 30/100

thu-ml/Causal-Forcing#29 · 3 comments · Opened Apr 13, 2026
Inquiry about video extension Open

enhancement

Difficulty 3/5 1-2 days Newbie friendliness 35/100

thu-ml/Causal-Forcing#28 · 3 comments · 1 reaction · Opened Apr 12, 2026
Question about VisionReward scores Open

question

Difficulty 3/5 1-2 days Newbie friendliness 35/100

thu-ml/Causal-Forcing#25 · 17 comments · Opened Mar 26, 2026
关于wan 2.2 5B模型支持 Open

Difficulty 5/5 Over a week Newbie friendliness 25/100

thu-ml/TurboDiffusion#124 · Opened Mar 20, 2026
Question about Vbench scores Open

question solved vbench

Difficulty 4/5 3-5 days Newbie friendliness 35/100

thu-ml/Causal-Forcing#18 · 14 comments · 1 reaction · Opened Mar 11, 2026
About the data preprcess code Open

question

Difficulty 2/5 1-3 hours Newbie friendliness 35/100

thu-ml/Causal-Forcing#17 · 1 comment · Opened Mar 11, 2026
net.load_state_dict(state_dict, assign=True) 失败 Open

Difficulty 4/5 3-5 days Newbie friendliness 35/100

thu-ml/TurboDiffusion#122 · 1 comment · Opened Mar 9, 2026
Generalization of the merging operation Open

Difficulty 5/5 Over a week Newbie friendliness 20/100

thu-ml/TurboDiffusion#121 · 2 comments · Opened Mar 9, 2026
hello,请教下关于模型兼容性的问题 Open

Difficulty 5/5 Over a week Newbie friendliness 20/100

thu-ml/TurboDiffusion#120 · 3 comments · Opened Mar 5, 2026
Questions on I2V support and multi-GPU inference Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 25/100

thu-ml/Causal-Forcing#15 · 6 comments · Opened Mar 5, 2026
Illegal Memory Access on 5090 Open

Difficulty 4/5 3-5 days Newbie friendliness 30/100

thu-ml/TurboDiffusion#119 · 1 comment · Opened Mar 5, 2026
Is there a script to test whether the first-stage model performs as expected? Open

question

Difficulty 3/5 1-2 days Newbie friendliness 25/100

thu-ml/Causal-Forcing#14 · 1 comment · Opened Mar 3, 2026
QA What benchmark/dataset is used for evaluations? Open

Difficulty 2/5 1-3 hours Newbie friendliness 35/100

thu-ml/TurboDiffusion#118 · Opened Feb 25, 2026
sla是全参微调吗，还是只训练SparseLinearAttention？ Open

Difficulty 1/5 Under an hour Newbie friendliness 20/100

thu-ml/TurboDiffusion#117 · 3 comments · Opened Feb 25, 2026
About the Guidance Scale Open

question

Difficulty 2/5 1-3 hours Newbie friendliness 35/100

thu-ml/Causal-Forcing#12 · 1 comment · Opened Feb 24, 2026
Question about Image-to-Video Comparison with CausVid and Self-Forcing (Checkpoints Request) Open

question

Difficulty 4/5 3-5 days Newbie friendliness 20/100

thu-ml/Causal-Forcing#11 · 3 comments · Opened Feb 23, 2026
About self.delta_mean Open

question

Difficulty 3/5 1-2 days Newbie friendliness 25/100

thu-ml/Causal-Forcing#9 · 1 comment · 1 reaction · Opened Feb 17, 2026
About Autoregressive Diffusion Training Dataset Open

question

Difficulty 3/5 1-2 days Newbie friendliness 25/100

thu-ml/Causal-Forcing#8 · 4 comments · Opened Feb 15, 2026
Question about different patch embedding implementation compared to official Wan-Video Open

Difficulty 5/5 Over a week Newbie friendliness 20/100

thu-ml/TurboDiffusion#114 · 1 comment · Opened Feb 14, 2026
Awesome work! Open

question

Difficulty 5/5 Over a week Newbie friendliness 5/100

thu-ml/Causal-Forcing#7 · 1 comment · Opened Feb 12, 2026
Support for WAN2.2 checkpoint Open

enhancement

Difficulty 5/5 Over a week Newbie friendliness 20/100

thu-ml/Causal-Forcing#4 · 2 comments · Opened Feb 6, 2026
Frame-level non-injectivtiy and Causal CD as substitute for ODE distillation Open

enhancement question

Difficulty 4/5 3-5 days Newbie friendliness 35/100

thu-ml/Causal-Forcing#3 · 6 comments · 1 reaction · Opened Feb 5, 2026
How to create long video Open

enhancement

Difficulty 3/5 1-2 days Newbie friendliness 35/100

thu-ml/Causal-Forcing#2 · 10 comments · Opened Feb 3, 2026
How is the performance if we train with sCM loss only? Open

Difficulty 5/5 Over a week Newbie friendliness 20/100

thu-ml/TurboDiffusion#111 · 1 comment · Opened Feb 1, 2026
Sparse Linear Layer weights are 0 for 14B 480p model Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

thu-ml/TurboDiffusion#109 · Opened Jan 27, 2026
帧率比较低？ Open

Difficulty 4/5 3-5 days Newbie friendliness 30/100

thu-ml/TurboDiffusion#108 · 1 comment · Opened Jan 27, 2026
推理效果 Open

Difficulty 4/5 3-5 days Newbie friendliness 25/100

thu-ml/TurboDiffusion#107 · 1 comment · Opened Jan 27, 2026

Showing the newest 100

This page lists what was indexed most recently. Advanced filter has the whole inventory, narrowed by language, difficulty and how long a task takes.

Open advanced filter

thu-ml

Showing the newest 100

Get new issues in your inbox