How to install OptimizedSD(basujindal) in already cloned stable-diffusion-webui?

[Skymaster3]

I want to try an optimizedSD(https://github.com/basujindal/stable-diffusion) because I get an "Out of memory" error when I try to turn on the "Hires. fix". Can you tell me which files need to be replaced in my repository? Do I just need to replace img2img.py and txt2img.py with optimized_img2img.py and optimized_txt2img.py? Or do I need something else? I apologize in advance for stupid questions, I'm new here.

[BetaDoggo]

From what I can tell the optimizations in that repo are already in the webui. The low and medium vram arguments should enable them. Also check the optimizations page on the wiki to see other options for decreasing memory usage.

[Skymaster3]

Thank you for answer. I already use some of these parameters:

set COMMANDLINE_ARGS=--lowvram --always-batch-cond-uncond --opt-split-attention
set PYTORCH_CUDA_ALLOC_CONF=garbage_collection_threshold:0.6,max_split_size_mb:128

Because my GPU is really bad (old laptop, GTX 960M) and txt2img generation without "Hires. fix" works fine. But if I turn on "Hires. fix", the error appears:

CUDA out of memory. Tried to allocate 6.89 GiB (GPU 0; 2.00 GiB total capacity; 140.48 MiB already allocated; 973.29 MiB free; 150.00 MiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation.

I don't understand why with all these parameters it tries to allocate so much memory.

[Nacurutu]

with my 1660ti 6gb vram i just can use hires fix x1.5 for images 512x768.. if I go further, I get a memory allocation error...