support for open-sora-plan v1.2?

[philipwan]

open-sora-plan update to v1.2 version, model struct changed from spatial + temporal to (THW) transformer block; It has a huge sequence_length, This inference cost is unacceptable，Can PAB provide additional support? Or are there any related plans in the future?

[oahzxl]

yeah! our pab can also be applied to 3d attention models like cogvideo and open_sora_plan v1.2. will support soon!

[philipwan]

yeah! our pab can also be applied to 3d attention models like cogvideo and open_sora_plan v1.2. will support soon!

Wow！I am looking forward to 3d attention PAB release

Thank you

[philipwan]

yeah! our pab can also be applied to 3d attention models like cogvideo and open_sora_plan v1.2. will support soon!

I see PAB strategy support open-sora-plan 1.2 / cogvideoX， how about accuracy， does it work now？

[oahzxl]

cogvideox is supported but we haven't support open sora plan v1.2. the result is good

---

发件人: philipwan @.> 发送时间: Wednesday, September 4, 2024 9:56:56 AM 收件人: NUS-HPC-AI-Lab/VideoSys @.> 抄送: Xuanlei Zhao @.>; Comment @.> 主题: Re: [NUS-HPC-AI-Lab/VideoSys] support for open-sora-plan v1.2? (Issue #153)

yeah! our pab can also be applied to 3d attention models like cogvideo and open_sora_plan v1.2. will support soon!

I see PAB strategy support open-sora-plan 1.2 / cogvideoX， how about accuracy， does it work now？

― Reply to this email directly, view it on GitHubhttps://github.com/NUS-HPC-AI-Lab/VideoSys/issues/153#issuecomment-2328171381, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AKOZKWVLKUDACJ4F72ZX7VDZU24MRAVCNFSM6AAAAABMEBZPQ2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMRYGE3TCMZYGE. You are receiving this because you commented.Message ID: @.***>

[philipwan]

cogvideox is supported but we haven't support open sora plan v1.2. the result is good … ________________________________ 发件人: philipwan @.> 发送时间: Wednesday, September 4, 2024 9:56:56 AM 收件人: NUS-HPC-AI-Lab/VideoSys @.> 抄送: Xuanlei Zhao @.>; Comment @.> 主题: Re: [NUS-HPC-AI-Lab/VideoSys] support for open-sora-plan v1.2? (Issue #153) yeah! our pab can also be applied to 3d attention models like cogvideo and open_sora_plan v1.2. will support soon! I see PAB strategy support open-sora-plan 1.2 / cogvideoX， how about accuracy， does it work now？ ― Reply to this email directly, view it on GitHub<#153 (comment)>, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AKOZKWVLKUDACJ4F72ZX7VDZU24MRAVCNFSM6AAAAABMEBZPQ2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMRYGE3TCMZYGE. You are receiving this because you commented.Message ID: @.***>

Sorry for my mistake, maybe v1.2 has the same struction with cogvideo, and PAB's result is good, I can try support PAB in plan v1.2...

[oahzxl]

oh that will be great. we welcome your contribution if you implement this on v1.2!