xdit-project/xDiT

Issues

• RoadMap and Looking for Contributions

  #213 opened 3 months ago by feifeibear
  0

• parallel bug for CogVideoX

  #295 opened 2 months ago by oahzxl
  0

• Contrains on Parallel Setting for CogVideoX and Performance issue

  #265 opened 2 months ago by feifeibear
  2

• Model flux fail

  #282 opened 2 months ago by gty111
  4

• latte模型的pos_embed没有被正确初始化

  #255 opened 2 months ago by weyeeji
  3

• Add AuraFlow support

  #291 opened 2 months ago by isidentical
  1

• Parallel VAE fails when the degree of parallelism > 1

  #271 opened 2 months ago by gty111
  0

• num_frames is not defined for latte

  #283 opened 2 months ago by littletomatodonkey
  1

• hunyuanDiT PipeFusion=8 on L40

  #284 opened 2 months ago by feifeibear
  0

• Flux.1 needs diffusers built from source code

  #267 opened 2 months ago by feifeibear
  2

• profile support

  #272 opened 2 months ago by kuangdao
  0

• [Feature] Disaggregating Warmup and Stale-KV Phases for Long Sequence DiT

  #127 opened 3 months ago by dannyxiaocn
  0

• Tensor size mismatch in CogVideoX transformer forward pass

  #260 opened 3 months ago by chen-yy20
  1

• 变量is_dp_last_group未定义

  #256 opened 3 months ago by weyeeji
  1

• FLUX with SP 并行生成图像差异

  #262 opened 3 months ago by lixiang007666
  2

• ValueError: Attention Processor class CogVideoXAttnProcessor2_0 is not supported by xFuser

  #254 opened 3 months ago by weyeeji
  4

• 运行CogVideoX-2b模型显示缺少bin文件

  #253 opened 3 months ago by weyeeji
  4

• Flux sp=2 errors

  #252 opened 3 months ago by feifeibear
  0

• SD3 example failed when batch size is greater than 1 .

  #247 opened 3 months ago by wushixong
  1

• Why Latte Model can‘t support PipeFusion

  #225 opened 3 months ago by philipwan
  3

• TypeError: _flash_attn_forward() missing 1 required positional argument: 'softcap'

  #217 opened 3 months ago by Pumbaa-peng
  2

• Flux Speed Up with PipeFusion

  #230 opened 3 months ago by mali-afridi
  2

• RuntimeError: expected scalar type Float but found Half

  #226 opened 3 months ago by ptkang
  1

• RTX 4090, flux model, out of memory; approach is not compatible with quantization

  #218 opened 3 months ago by csdY123
  4

• Running with torch.compile on Flux with ulysses_degree=2 results in incorrect outcomes.

  #220 opened 3 months ago by feifeibear
  0

• Image Difference for SD3 in the refactored code, when pipefusion_parallel_degree changes

  #163 opened 3 months ago by khadijairfan2345
  1

• [feature] Loras support for SD and Flux models

  #192 opened 3 months ago by tsubasakong
  2

• 为什么没有像Megatron一样，使用batch_isend_irecv做p2p通信？

  #132 opened 3 months ago by taozhiwei
  0

• [SERVER] Can give a http interface example?

  #185 opened 3 months ago by senlyu163
  3

• Whether HunyuanDiT-v1.2-Diffusers supports the method Sequence Parallelism

  #167 opened 3 months ago by linuxmi
  2

• How to set the offline download model path

  #189 opened 3 months ago by lonngxiang
  1

• [bug] latte output bug when sp=8

  #206 opened 3 months ago by dannyxiaocn
  0

• Flux Sequence Parallel Poor Scalability Issue

  #204 opened 3 months ago by feifeibear
  1

• RTX 4090, flux model, out of memory

  #195 opened 3 months ago by csdY123
  1

• Cannot allocate memory

  #190 opened 4 months ago by lonngxiang
  7

• how to run the PixArt-alpha/PixArt-XL-2-1024-MS with multiple gpus?

  #136 opened 4 months ago by lambda7xx
  3

• Thoughts to implement this on ODD number of GPUs?

  #139 opened 4 months ago by alivisionrd
  10

• Benchmark running error？

  #154 opened 4 months ago by tensorflowt
  12

• [feature] runtime check diffusers version

  #170 opened 4 months ago by feifeibear
  0

• Why this is inference only but not for training?

  #164 opened 4 months ago by spacegoing
  1

• add pipefusers to pypi

  #133 opened 4 months ago by feifeibear
  0

• Process stuck after all timesteps are finished

  #147 opened 4 months ago by Eigensystem
  0

• The best communication pattern for Sequence Parallel(SP)+PipeFusion(PP) hybrid parallelism

  #146 opened 4 months ago by feifeibear
  0

• [REFACTOR] Managing Configuration with a Single Instance

  #143 opened 4 months ago by Eigensystem
  1

• Performance issue of sequence parallelism on A100 after refactor

  #137 opened 4 months ago by feifeibear
  1

• # of parameters on each device

  #135 opened 4 months ago by wonkyoc
  3

• HunyuanDiT 2GPUs PipeFusion Bugs

  #121 opened 4 months ago by feifeibear
  0

• Adding support for pixart-sigma model?

  #115 opened 4 months ago by foreverpiano
  2

• HunyuanDiT pipefusion use_split_batch True errors

  #122 opened 4 months ago by feifeibear
  0

• Tensor Parallelism with Stale Activations to Reduce Communication Overhead

  #116 opened 5 months ago by SaudxInu
  7