16GB Radeon 7800 XT Image Generation Success with the Q6 GGUF Model
I basically hit the limits of my 16GB card using the below workflow, but I was able to successfully generate using this technique on an AMD 7800 XT with 16GB of memory. I used the JSON workflow provided by the tencent SRPO team post and then modified it to point to the unet GGUF loader, removing the diffusion and standard unet loader. Results were good with some small artifacts, though I did not attempt to tune the parameters at all for better images. Attempted 1024x1024 and 1280x720 with both resolutions successful. My launch parameters for the GPU were:
'python3 main.py --listen --port 7870 --disable-smart-memory --use-split-cross-attention'
Some of those launch parameters might be out of date let me know if they are but cross attestation was the only way I could get my card to work often. I don't know if AMD is still a struggle for image generation, but I do know that 16GB of memory can be with new models so I wanted to share this information here since others have shared simplified workflows. Thank you Befox for sharing these quants!
