zRzRzRzRzRzRzR Events in 2024 - Ecosyste.ms: Timeline

Because there are two different models, 0.5 and 1. In the config, the glm-edge series model needs to be set to 1. https://huggingface.co/ZP2HF/glm-edge-4b-chat/blob/6a5e92d0092bba5f94abd471720238b...

View on GitHub

zRzRzRzRzRzRzR created a review on a pull request on huggingface/transformers

November 21, 2024 2:51am

View on GitHub

zRzRzRzRzRzRzR created a comment on an issue on THUDM/CogVideo

November 21, 2024 2:44am

No, it is not 21 in the calculation process, but 22, the first frame will be copied once.

View on GitHub

zRzRzRzRzRzRzR created a comment on an issue on THUDM/CogVideo

November 20, 2024 12:20pm

You need to set it to 81 and clone the source code of the diffusers library, as this is where the implementation of version 1.5 is located.

View on GitHub

zRzRzRzRzRzRzR created a comment on an issue on THUDM/CogVideo

November 20, 2024 12:19pm

这个模型是文生视频的，文生图是cogview

View on GitHub

zRzRzRzRzRzRzR created a comment on an issue on THUDM/CogVideo

November 20, 2024 12:18pm

因为你要先载入T5进行编码，这个是要显存的，接着你还要调用VAE进行编码。这俩的显存占用差不多是40G显存。

View on GitHub

zRzRzRzRzRzRzR created a comment on a pull request on huggingface/transformers

November 20, 2024 10:52am

Now this implementation is compatible with both GLM-Edge and GLM-4 models, @cyril-k, I would like to know how to modify modular_glm.py to achieve automatic updates, because some parts of the implem...

View on GitHub

zRzRzRzRzRzRzR created a comment on an issue on THUDM/CogVideo

November 20, 2024 10:44am

now is https://github.com/huggingface/diffusers/tree/main

View on GitHub

zRzRzRzRzRzRzR created a comment on an issue on THUDM/CogVideo

November 20, 2024 10:41am

很明显不是模型的问题，你下载的路径是正确的。我在这个环境下是正常运行的。 ``` CUDA_VISIBLE_DEVICES=1 python cli_demo.py --prompt "Leaves gently cascade from the rooftop of a quaint, rustic cottage, their vibrant hues contrasting wit...

View on GitHub

zRzRzRzRzRzRzR pushed 28 commits to main zRzRzRzRzRzRzR/diffusers

November 20, 2024 10:35am

[Core] introduce `controlnet` module (#8768) * move vae flax module. * controlnet module. * prepare for PR. ... ded3db1
[Flux] reduce explicit device transfers and typecasting in flux. (#9817) reduce explicit device transfers and typeca... 5588725
Improve downloads of sharded variants (#9869) * update * update * update * update --------- Co-author... 1b39254
[fix] Replaced shutil.copy with shutil.copyfile (#9885) fix shutil.copy 0be52c0
Enabling gradient checkpointing in eval() mode (#9878) * refactored 5b972fb
[FIX] Fix TypeError in DreamBooth SDXL when use_dora is False (#9879) * fix use_dora * fix style and quality *... 9cc96a6
[Advanced LoRA v1.5] fix: gradient unscaling problem (#7018) fix: gradient unscaling problem Co-authored-by: Lino... d720b21
Revert "[Flux] reduce explicit device transfers and typecasting in flux." (#9896) Revert "[Flux] reduce explicit dev... 8d6dc2b
Feature IP Adapter Xformers Attention Processor (#9881) * Feature IP Adapter Xformers Attention Processor: this fix ... dac623b
Notebooks for Community Scripts Examples (#9905) * Add Notebooks on Community Scripts 1dbd26f
Fix Progress Bar Updates in SD 1.5 PAG Img2Img pipeline (#9925) fix progress bar updates in SD 1.5 PAG Img2Img pipeline d74483c
Update pipeline_flux_img2img.py (#9928) * Update pipeline_flux_img2img.py Added FromSingleFileMixin to this pipel... 5c94937
add depth controlnet sd3 pre-trained checkpoints to docs (#9937) 40ab1c0
Move Wuerstchen Dreambooth to research_projects (#9935) update file paths to research_projects folder. Co-authore... e255920
Update ip_adapter.py (#8882) update comments of load_ip_adapter function d38c50c
Modify apply_overlay for inpainting with padding_mask_crop (Inpainting area: "Only Masked") (#8793) * Modify apply_o... 1d2204d
Correct pipeline_output.py to the type Mochi (#9945) Correct pipeline_output.py 07d0fbf
Add all AttnProcessor classes in `AttentionProcessor` type (#9909) Add all AttnProcessor in `AttentionProcessor` type 345907f
Fixed Nits in Docs and Example Script (#9940) Fixed nits in docs and example script. Co-authored-by: Sayak Paul <... 365a938
Add server example (#9918) * Add server example. * Minor updates to README. * Add fixes after local testing. ... c3c94fe
and 8 more ...

View on GitHub

zRzRzRzRzRzRzR created a comment on an issue on THUDM/CogVideo

November 20, 2024 10:24am

emm, is your model self-converted or are you using the CogVideoX1.5-5B model that we have converted, because if it's the latter, this issue wouldn't occur, right? If it's to be self-converted, firs...

View on GitHub

zRzRzRzRzRzRzR created a comment on an issue on THUDM/CogVideo

November 20, 2024 10:22am

During training, the first frame is copied once to ensure that the length can be evenly divided by t_patch (2).

View on GitHub

zRzRzRzRzRzRzR created a comment on an issue on THUDM/CogVideo

November 20, 2024 10:21am

我们现在在维护diffusres的版本，sat版本会出现OOM且后续不会维护，使用diffusers版本进行微调，并使用cogvideox-factory或许能解决问题

View on GitHub