Merge branch 'release_candidate'

make --disable-opt-split-attention command line option work again

CHANGELOG.md

@@ -1,3 +1,17 @@
+## 1.3.1
+
+### Features:
+ * revert default cross attention optimization to Doggettx
+
+### Bug Fixes:
+ * fix bug: LoRA don't apply on dropdown list sd_lora
+ * fix png info always added even if setting is not enabled
+ * fix some fields not applying in xyz plot
+ * fix "hires. fix" prompt sharing same labels with txt2img_prompt
+ * fix lora hashes not being added properly to infotex if there is only one lora
+ * fix --use-cpu failing to work properly at startup
+ * make --disable-opt-split-attention command line option work again
+
 ## 1.3.0
 
 ### Features:

modules/cmd_args.py

@@ -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-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("--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("--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")

modules/sd_hijack.py

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

modules/sd_hijack_optimizations.py

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

modules/sd_models.py

@@ -313,8 +313,6 @@ def load_model_weights(model, checkpoint_info: CheckpointInfo, state_dict, timer
 
         timer.record("apply half()")
 
-    devices.dtype = torch.float32 if shared.cmd_opts.no_half else torch.float16
-    devices.dtype_vae = torch.float32 if shared.cmd_opts.no_half or shared.cmd_opts.no_half_vae else torch.float16
     devices.dtype_unet = model.model.diffusion_model.dtype
     devices.unet_needs_upcast = shared.cmd_opts.upcast_sampling and devices.dtype == torch.float16 and devices.dtype_unet == torch.float16
 

modules/shared.py

@@ -6,6 +6,7 @@ import threading
 import time
 
 import gradio as gr
+import torch
 import tqdm
 
 import modules.interrogate
@@ -76,6 +77,9 @@ cmd_opts.disable_extension_access = (cmd_opts.share or cmd_opts.listen or cmd_op
 devices.device, devices.device_interrogate, devices.device_gfpgan, devices.device_esrgan, devices.device_codeformer = \
     (devices.cpu if any(y in cmd_opts.use_cpu for y in [x, 'all']) else devices.get_optimal_device() for x in ['sd', 'interrogate', 'gfpgan', 'esrgan', 'codeformer'])
 
+devices.dtype = torch.float32 if cmd_opts.no_half else torch.float16
+devices.dtype_vae = torch.float32 if cmd_opts.no_half or cmd_opts.no_half_vae else torch.float16
+
 device = devices.device
 weight_load_location = None if cmd_opts.lowram else "cpu"