Dtype not understood: F8_E4M3！！！！！！
Why doesn't my model understand
I'm getting the same error with the sd3_medium_incl_clips_t5xxlfp8.safetensors file but it works normally with sd3_medium_incl_clips.safetensors
I had a similar issue that i solved by downloading the base model and the other clip and xxxlfp16 model separately and then loading them in using the 3 clip node. My hardware doesn't have the fp8 support, so the fp16 did the trick.
I think you need to update safetensors library, not %100 sure tho.
Got this error myself. I fixed it by updating the safetensors package. Just run "pip install safetensors --upgrade" and you should be good.
I think you need to update safetensors library, not %100 sure tho.
Resolved by updating safetensors package to version 0.4.3. Thank you very much!
Got this error myself. I fixed it by updating the safetensors package. Just run "pip install safetensors --upgrade" and you should be good.
Resolved by updating safetensors package to version 0.4.3. Thank you very much!
Updating Safetensors didn't work for me, even though I was on 4.2 and am now on 4.3.
I'm on AMD - do I need to use the fp16 version?
UPDATE: downloading the fp16 version got rid of the "Dtype not understood" error, but now I get the dreaded memory error, despite my GPU having 16G of RAM.
Could not allocate tensor with 1734000000 bytes. There is not enough GPU video memory available
The largest image I could get to run is 768x768.
SDXL can generate 1024x1024 images on my GPU no problem.
Couldn’t hurt to try. My issue was only resolved after using the fp16 version.
Couldn’t hurt to try. My issue was only resolved after using the fp16 version.
Yeah, that did fix the error. But my system might not be powerful enough to benefit from SD3, at least at this stage in its development.
Even generating a 768x768 image takes 3 times as long as using SDXL based models and for the first time it noticeably slows down my system, causing the mouse to stagger a bit.
I never thought my relatively new rig would feel obsolete so soon.
12th gen i7, RX7800 GPU, 64G system RAM - not a monster, but no slouch in general.
i Fixed just running the " update_comfyui_and_python_dependencies" bat