StableDiffusion/WebUI
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

revert default cross attention optimization to Doggettx

Browse Source 
make --disable-opt-split-attention command line option work again
This commit is contained in:
AUTOMATIC 2023-06-01 08:12:06 +03:00
parent 17a66931da
commit 3ee1238630
3 changed files with 6 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
modules/cmd_args.py
View File

@ -62,7 +62,7 @@ parser.add_argument("--opt-split-attention-invokeai", action='store_true', help=
parser.add_argument("--opt-split-attention-v1", action='store_true', help="prefer older version of split attention optimization for automatic choice of optimization") parser.add_argument("--opt-split-attention-v1", action='store_true', help="prefer older version of split attention optimization for automatic choice of optimization")
parser.add_argument("--opt-sdp-attention", action='store_true', help="prefer scaled dot product cross-attention layer optimization for automatic choice of optimization; requires PyTorch 2.*") parser.add_argument("--opt-sdp-attention", action='store_true', help="prefer scaled dot product cross-attention layer optimization for automatic choice of optimization; requires PyTorch 2.*")
parser.add_argument("--opt-sdp-no-mem-attention", action='store_true', help="prefer scaled dot product cross-attention layer optimization without memory efficient attention for automatic choice of optimization, makes image generation deterministic; requires PyTorch 2.*") parser.add_argument("--opt-sdp-no-mem-attention", action='store_true', help="prefer scaled dot product cross-attention layer optimization without memory efficient attention for automatic choice of optimization, makes image generation deterministic; requires PyTorch 2.*")
parser.add_argument("--disable-opt-split-attention", action='store_true', help="does not do anything") parser.add_argument("--disable-opt-split-attention", action='store_true', help="prefer no cross-attention layer optimization for automatic choice of optimization")
parser.add_argument("--disable-nan-check", action='store_true', help="do not check if produced images/latent spaces have nans; useful for running without a checkpoint in CI") parser.add_argument("--disable-nan-check", action='store_true', help="do not check if produced images/latent spaces have nans; useful for running without a checkpoint in CI")
parser.add_argument("--use-cpu", nargs='+', help="use CPU as torch device for specified modules", default=[], type=str.lower) parser.add_argument("--use-cpu", nargs='+', help="use CPU as torch device for specified modules", default=[], type=str.lower)
parser.add_argument("--listen", action='store_true', help="launch gradio with 0.0.0.0 as server name, allowing to respond to network requests") parser.add_argument("--listen", action='store_true', help="launch gradio with 0.0.0.0 as server name, allowing to respond to network requests")

2
modules/sd_hijack.py
View File

@ -68,6 +68,8 @@ def apply_optimizations():
if selection == "None": if selection == "None":
matching_optimizer = None matching_optimizer = None
elif selection == "Automatic" and shared.cmd_opts.disable_opt_split_attention:
matching_optimizer = None
elif matching_optimizer is None: elif matching_optimizer is None:
matching_optimizer = optimizers[0] matching_optimizer = optimizers[0]

6
modules/sd_hijack_optimizations.py
View File

@ -59,7 +59,7 @@ class SdOptimizationSdpNoMem(SdOptimization):
name = "sdp-no-mem" name = "sdp-no-mem"
label = "scaled dot product without memory efficient attention" label = "scaled dot product without memory efficient attention"
cmd_opt = "opt_sdp_no_mem_attention" cmd_opt = "opt_sdp_no_mem_attention"
priority = 90 priority = 80
def is_available(self): def is_available(self):
return hasattr(torch.nn.functional, "scaled_dot_product_attention") and callable(torch.nn.functional.scaled_dot_product_attention) return hasattr(torch.nn.functional, "scaled_dot_product_attention") and callable(torch.nn.functional.scaled_dot_product_attention)
@ -73,7 +73,7 @@ class SdOptimizationSdp(SdOptimizationSdpNoMem):
name = "sdp" name = "sdp"
label = "scaled dot product" label = "scaled dot product"
cmd_opt = "opt_sdp_attention" cmd_opt = "opt_sdp_attention"
priority = 80 priority = 70
def apply(self): def apply(self):
ldm.modules.attention.CrossAttention.forward = scaled_dot_product_attention_forward ldm.modules.attention.CrossAttention.forward = scaled_dot_product_attention_forward
@ -116,7 +116,7 @@ class SdOptimizationInvokeAI(SdOptimization):
class SdOptimizationDoggettx(SdOptimization): class SdOptimizationDoggettx(SdOptimization):
name = "Doggettx" name = "Doggettx"
cmd_opt = "opt_split_attention" cmd_opt = "opt_split_attention"
priority = 20 priority = 90
def apply(self): def apply(self):
ldm.modules.attention.CrossAttention.forward = split_cross_attention_forward ldm.modules.attention.CrossAttention.forward = split_cross_attention_forward