@KBlueleaf

Graduate student in Taiwan. Leader of KohakuLab. Researcher/Dev in ComfyOrg

 Taiwan
kblueleaf.net/
Joined May 2021
  • Tweets3.8K
  • Following1K
  • Followers4.8K
726Media
TT-VidT is accepted at #NeurIPS2026 🎉 A temporal-decoupled methodology in video pretraining to provide real "motion" info extracted from the video and only requires small scale training to reach high quality results. Huge thanks to all collaborators, @NVIDIAAI and @spellbrush
3
9
2
66
9,786
I have 2x2080ti-22G (modded) and I found 2080ti also have similar unlock stuff from community, I'm trying to make it become initramfs module (easier to recover than the UEFI module design) and see if it works. UwUb
琥珀青葉@KohakuLab
@KBlueleafOct 23
Confirmed that 3090 unlock is successful with 360watt power wall, I got 75T -> 135T (measured in Pytorch) Which is insane I used github.com/Zagitta/gpu-unloc… with commit a9b0c75
7
9
1
118
4,327
2080ti success xitter.cf/KBlueleaf/status/21141…
琥珀青葉@KohakuLab
@KBlueleaf13h
Now I also confirmed that 2080ti unlock is successful with similar approach as the 30-series' method it reach nearly same speed as 350w 3090(locked) with only 250w in mmapeak and mamf-finder github.com/KohakuBlueleaf/Ko…
7
612
Confirmed that 3090 unlock is successful with 360watt power wall, I got 75T -> 135T (measured in Pytorch) Which is insane I used github.com/Zagitta/gpu-unloc… with commit a9b0c75
137
325
80
4,504
874,925
Remember to check my latest news on 2080ti! xitter.cf/KBlueleaf/status/21141…
琥珀青葉@KohakuLab
@KBlueleaf13h
Now I also confirmed that 2080ti unlock is successful with similar approach as the 30-series' method it reach nearly same speed as 350w 3090(locked) with only 250w in mmapeak and mamf-finder github.com/KohakuBlueleaf/Ko…
1
1
42
23,462
琥珀青葉@KohakuLab retweeted
琥珀青葉@KohakuLab
@KBlueleaf13h
Now I also confirmed that 2080ti unlock is successful with similar approach as the 30-series' method it reach nearly same speed as 350w 3090(locked) with only 250w in mmapeak and mamf-finder github.com/KohakuBlueleaf/Ko…
琥珀青葉@KohakuLab
@KBlueleafOct 23
Confirmed that 3090 unlock is successful with 360watt power wall, I got 75T -> 135T (measured in Pytorch) Which is insane I used github.com/Zagitta/gpu-unloc… with commit a9b0c75
33
79
17
816
88,159
Now I also confirmed that 2080ti unlock is successful with similar approach as the 30-series' method it reach nearly same speed as 350w 3090(locked) with only 250w in mmapeak and mamf-finder github.com/KohakuBlueleaf/Ko…
琥珀青葉@KohakuLab
@KBlueleafOct 23
Confirmed that 3090 unlock is successful with 360watt power wall, I got 75T -> 135T (measured in Pytorch) Which is insane I used github.com/Zagitta/gpu-unloc… with commit a9b0c75
33
79
17
816
88,159
For those who don't know: 1. This unlock the "nerfed FP32 accumulator" stuff 2. you can rent a A6000/6000 ada to test the mmapeak to know what I'm saying (or check the FP16 accumulation speed of 3090/4090) 3. no unlock core, no overclock, it just remove the nerf
6
7
1
361
59,845
And as this unlock is NOT overclock: 1. if you use kernel mod method, it is riskless 2. won't hurt your card at all (FP16 accumulator already "run in unlocked speed")
1
6
1,450
And as this unlock only happen in tensor core: 1. not help mining, not help gaming 2. only help AI usage (potentially DLSS but not guranteed)
5
3
341
47,465
琥珀青葉@KohakuLab retweeted
琥珀青葉@KohakuLab
@KBlueleafOct 23
Confirmed that 3090 unlock is successful with 360watt power wall, I got 75T -> 135T (measured in Pytorch) Which is insane I used github.com/Zagitta/gpu-unloc… with commit a9b0c75
137
325
80
4,504
874,925
琥珀青葉@KohakuLab retweeted
Tim Pan
@mediastorm_enOct 22
My iPhone Duo video went viral, so I'm giving away 100 iPhones on X 😛
93,477
100,065
6,733
143,842
4,235,287
琥珀青葉@KohakuLab retweeted
DeepGHS@DeepGHSOct 20
Something exciting is coming. 👀✨ We've been working on an anime-focused 2B pretrained DINOv3 model! Our internal evaluations show promising improvements across a wide range of anime-related downstream tasks. We're now making the preparations for release. Stay tuned! 🚀
5
15
147
7,776
琥珀青葉@KohakuLab retweeted
農藝女孩搬家了@loveagrogirlOct 19
李登輝總統、李元簇副總統1995年親手栽植的紀念樹，已經被蔣萬安消滅。 台北市政府1999專線說，這是因為褐根病，你信嗎。
52
259
11
1,155
19,614
琥珀青葉@KohakuLab retweeted
琥珀青葉@KohakuLab
@KBlueleafOct 19
I made FP16 fixed version of MOSS-image It support any FP16 compute card (such as V100/2080ti) without any issue. and if your lora or PEFT model only trained QKVO/MLP layers, it can be used directly without issue! huggingface.co/KBlueLeaf/MOS…
琥珀青葉@KohakuLab
@KBlueleafOct 18
KohakuMossTrainer is released: github.com/KohakuBlueleaf/Ko… **Features** - LyCORIS all algorithms - SM70 ~ SM120 - Intel Arc GPU - Windows and LInux - multi GPU training - Model offload ***Try to test it now!***

GitHub - KohakuBlueleaf/KohakuMossTrainer

Contribute to KohakuBlueleaf/KohakuMossTrainer development by creating an account on GitHub.

github.com
1
7
2
47
4,181
I made FP16 fixed version of MOSS-image It support any FP16 compute card (such as V100/2080ti) without any issue. and if your lora or PEFT model only trained QKVO/MLP layers, it can be used directly without issue! huggingface.co/KBlueLeaf/MOS…
琥珀青葉@KohakuLab
@KBlueleafOct 18
KohakuMossTrainer is released: github.com/KohakuBlueleaf/Ko… **Features** - LyCORIS all algorithms - SM70 ~ SM120 - Intel Arc GPU - Windows and LInux - multi GPU training - Model offload ***Try to test it now!***

GitHub - KohakuBlueleaf/KohakuMossTrainer

Contribute to KohakuBlueleaf/KohakuMossTrainer development by creating an account on GitHub.

github.com
1
7
2
47
4,181
Interestingly (or actually expected), the FP16 fixed model actually more precise than BF16 when compare to FP32 That's why I keep staying in FP16 even when I'm already using 5090 and B300. More mantissa = more accuracy. Range is not a problem, it is a skill issue
10
801
琥珀青葉@KohakuLab retweeted
琥珀青葉@KohakuLab
@KBlueleafOct 18
The style lora test on Moss-image-base with 100step only
琥珀青葉@KohakuLab
@KBlueleafOct 18
KohakuMossTrainer is released: github.com/KohakuBlueleaf/Ko… **Features** - LyCORIS all algorithms - SM70 ~ SM120 - Intel Arc GPU - Windows and LInux - multi GPU training - Model offload ***Try to test it now!***

GitHub - KohakuBlueleaf/KohakuMossTrainer

Contribute to KohakuBlueleaf/KohakuMossTrainer development by creating an account on GitHub.

github.com
7
83
5,479
The style lora test on Moss-image-base with 100step only
琥珀青葉@KohakuLab
@KBlueleafOct 18
KohakuMossTrainer is released: github.com/KohakuBlueleaf/Ko… **Features** - LyCORIS all algorithms - SM70 ~ SM120 - Intel Arc GPU - Windows and LInux - multi GPU training - Model offload ***Try to test it now!***

GitHub - KohakuBlueleaf/KohakuMossTrainer

Contribute to KohakuBlueleaf/KohakuMossTrainer development by creating an account on GitHub.

github.com
7
83
5,479
琥珀青葉@KohakuLab retweeted
琥珀青葉@KohakuLab
@KBlueleafOct 18
KohakuMossTrainer is released: github.com/KohakuBlueleaf/Ko… **Features** - LyCORIS all algorithms - SM70 ~ SM120 - Intel Arc GPU - Windows and LInux - multi GPU training - Model offload ***Try to test it now!***

GitHub - KohakuBlueleaf/KohakuMossTrainer

Contribute to KohakuBlueleaf/KohakuMossTrainer development by creating an account on GitHub.

github.com
1
23
5
256
40,612
KohakuMossTrainer is released: github.com/KohakuBlueleaf/Ko… **Features** - LyCORIS all algorithms - SM70 ~ SM120 - Intel Arc GPU - Windows and LInux - multi GPU training - Model offload ***Try to test it now!***

GitHub - KohakuBlueleaf/KohakuMossTrainer

Contribute to KohakuBlueleaf/KohakuMossTrainer development by creating an account on GitHub.

github.com
1
23
5
256
40,612
The weight offload mechanism allow you to achieve 6GB vram training with most aggresive offloading you can still reach 1img/sec on RTX4090 with bs4 (20k token per batch) on windows and only occupy 10~12GB vram! This is tested "without" latent caching and te caching
12
3,219
Load more