Software Engineering Courses (SECourses)•16mo ago

they just suck it happens to frequently

Furkan Gözükara SECoursesOP•9/21/24, 2:05 PM

their biggest advantage is permanent storage but this slowness killing it

FFurkan Gözükara SECourses they just suck it happens to frequently

�

𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮•9/21/24, 2:07 PM

Their ui is much superior to massed compute and I don't get this problem very often and most of the time I am using 8X l40s which works super fast

�𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮 Their ui is much superior to massed compute and I don't get this problem very of...

Furkan Gözükara SECoursesOP•9/21/24, 2:08 PM

8x l40s of course would be fast lol

Furkan Gözükara SECoursesOP•9/21/24, 2:08 PM

it is newest machine

FFurkan Gözükara SECourses it is newest machine

�

𝗡𝗶𝘁𝗶𝗻 𝗣𝗮𝗿𝗲𝘃𝗮•9/21/24, 2:09 PM

oh yes, I forgot that

Furkan Gözükara SECoursesOP•9/21/24, 2:46 PM

20 epoch fine tuning 256 images

Furkan Gözükara SECoursesOP•9/21/24, 2:47 PM

it can do cartoon atm but not very resemblaing

Furkan Gözükara SECoursesOP•9/21/24, 2:47 PM

Furkan Gözükara SECoursesOP•9/21/24, 2:48 PM

it can do beard amazing

Furkan Gözükara SECoursesOP•9/21/24, 2:48 PM

Furkan Gözükara SECoursesOP•9/21/24, 2:48 PM

Furkan Gözükara SECoursesOP•9/21/24, 2:57 PM

this bleeding issue of flux so annoying

Furkan Gözükara SECoursesOP•9/21/24, 2:57 PM

20e poch photo of barrack obama

1457-photo20of20barack20obama-FT_256_images-000020-1690245631.png

Dazzastrous [4090]•9/21/24, 2:58 PM

FFurkan Gözükara SECourses 20e poch photo of barrack obama

druhl•9/21/24, 3:09 PM

That's Furrack Obakara

druhl•9/21/24, 3:09 PM

Furkan Gözükara SECoursesOP•9/21/24, 3:32 PM

new public article

Furkan Gözükara SECoursesOP•9/21/24, 3:32 PM

https://www.patreon.com/posts/112494682?pr=true

Patreon

Multi-GPU FLUX Full Fine Tuning Experiments and Requirements on Run...

Get more from SECourses: Tutorials, Guides, Resources, Training, FLUX, MidJourney, Voice Clone, TTS, ChatGPT, GPT, LLM, Scripts by Furkan Gözü on Patreon

daroth•9/21/24, 5:19 PM

at the moment nothing points to the 5090 having more than 28gigs of vram, even that is hearsay

Ddaroth at the moment nothing points to the 5090 having more than 28gigs of vram, even t...

Furkan Gözükara SECoursesOP•9/21/24, 5:20 PM

Ye so shameless

Furkan Gözükara SECoursesOP•9/21/24, 5:28 PM

500 steps left first checkpoint

daroth•9/21/24, 5:42 PM

thats gonna take a while

Ddaroth thats gonna take a while 🙂

Furkan Gözükara SECoursesOP•9/21/24, 6:16 PM

Maxivi•9/21/24, 6:20 PM

@Dr. Furkan Gözükara any update on the fp8 for dreambooth Furak Obama?

Maxivi•9/21/24, 6:31 PM

@Dr. Furkan Gözükara do you get this issue when extracting lora

flux_extract_lora.py:68
31%|█████████████████████████████████████████████████████████████████████████████ | 93/304 [01:57<09:26, 1.86s/it]
Traceback (most recent call last):
File "/workspace/kohya_ss/sd-scripts/networks/flux_extract_lora.py", line 219, in <module>
svd(**vars(args))
File "/workspace/kohya_ss/sd-scripts/networks/flux_extract_lora.py", line 87, in svd
value_t = f_tuned.get_tensor(key)
File "/workspace/kohya_ss/sd-scripts/library/utils.py", line 249, in get_tensor
return self._deserialize_tensor(tensor_bytes, metadata)
File "/workspace/kohya_ss/sd-scripts/library/utils.py", line 265, in _deserialize_tensor
byte_tensor = torch.frombuffer(tensor_bytes, dtype=torch.uint8)
ValueError: both buffer length (0) and count (-1) must not be 0