I am loading the SwarmUI and noticed this in the debug logs. It's taking several minutes for the mod
I am loading the SwarmUI and noticed this in the debug logs. It's taking several minutes for the model to load:
W2024-08-14 15:14:22.415 [Debug] [ComfyUI-0/STDERR] model weight dtype torch.float8_e4m3fn, manual cast: torch.bfloat16
WHow do I keep it from casting it to bfloat16?
Flet me ask this
WIt seems like its loading into main system memory first, even though I have a 4090 which can easily hold it. Nvitop shows system memory usage steadily climbing
Wthis is using the dev model as well.
Fye i think it is designed that way
WOk, but it's been over 9 minutes and the model still hasn't loaded.
W2024-08-14 15:14:17.991 [Info] User local requested 1 image with model 'flux1-dev-fp8.safetensors'...
2024-08-14 15:14:17.997 [Debug] [BackendHandler] Backend request #1 for model flux1-dev-fp8.safetensors, maxWait=7.00:00:00.
2024-08-14 15:14:18.002 [Debug] [BackendHandler] backend #0 will load a model: flux1-dev-fp8.safetensors, with 1 requests waiting for 0 seconds
2024-08-14 15:14:20.374 [Debug] [ComfyUI-0/STDERR] got prompt
2024-08-14 15:14:22.415 [Debug] [ComfyUI-0/STDERR] model weight dtype torch.float8_e4m3fn, manual cast: torch.bfloat16
2024-08-14 15:14:22.604 [Debug] [ComfyUI-0/STDERR] model_type FLUX
2024-08-14 15:14:17.997 [Debug] [BackendHandler] Backend request #1 for model flux1-dev-fp8.safetensors, maxWait=7.00:00:00.
2024-08-14 15:14:18.002 [Debug] [BackendHandler] backend #0 will load a model: flux1-dev-fp8.safetensors, with 1 requests waiting for 0 seconds
2024-08-14 15:14:20.374 [Debug] [ComfyUI-0/STDERR] got prompt
2024-08-14 15:14:22.415 [Debug] [ComfyUI-0/STDERR] model weight dtype torch.float8_e4m3fn, manual cast: torch.bfloat16
2024-08-14 15:14:22.604 [Debug] [ComfyUI-0/STDERR] model_type FLUX
WAt this point it's 13 minutes
Fsomething seems wrong
Fhow much RAM you have?
W24gb, 64gb of system ram
Fok both good
Whold on it finally returned let me get the log
Fcan you check your hdd reading speed
Wmessage.txt4.93KB
Fi think your hdd was too slow
Whmm.. ok I'll move it to a faster drive and see if it changes.
Fsure give it a try
Fi am also adding new downloads to downloader
F4bit bnb
Fmerged model
WThat fixed it. Thank you!
Fgreat
Fman i love claude so much 
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=40)
D
DFlux Canny
D
Fnice
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=16)
Fcant wait to train flux haha :d
Dits next level - so fast everyday theres something new, I fear theres gonna be more calls to regulate it , due to the sheer quality of some of the images harder and harder to tell if its Ai or not
Dsome of the lora images are really impressive and still in early days
AIs the tutorial up for fine tuning Dr?
Fnot yet
AI'm stoked, do you think itll be doable on 4090?
Fyep
Fi am waiting kohya
AI see some folks using some other method is it the ai toolkit?
NAnyone else having trouble getting SUPIR to attach the face upscale results to the final output again? I've updated to the latest SUPIR release too.
V@Dr. Furkan Gözükara hello Furkan, morning, do you know if gligen model can be used for SDXL?
Vor any similar area conditioning can be done?
Vlike dense diffusion or anything similar
Manyone has this flux+lora link ?
Mhttps://huggingface.co/XLabs-AI/flux-lora-collection/commit/78eaa8614cdf96325471c3eb2ee7da139433edcc#d2h-592819
whats the relation between this tensors and midjourney anyone know ?
whats the relation between this tensors and midjourney anyone know ?
Fye it happens at some images
Fwe couldnt solve sadly
Mare they weights for the midjourney v6 ?