From 2cb1b65309256763814a006a07b683c0f1013a30 Mon Sep 17 00:00:00 2001
From: Kohaku-Blueleaf <59680068+KohakuBlueleaf@users.noreply.github.com>
Date: Sun, 28 Jan 2024 22:18:46 +0800
Subject: [PATCH 01/10] Bump safetensors' version to 0.4.2

---
 requirements_versions.txt | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/requirements_versions.txt b/requirements_versions.txt
index 2a922f28..5e30b5ea 100644
--- a/requirements_versions.txt
+++ b/requirements_versions.txt
@@ -19,7 +19,7 @@ piexif==1.1.3
 psutil==5.9.5
 pytorch_lightning==1.9.4
 resize-right==0.0.2
-safetensors==0.3.1
+safetensors==0.4.2
 scikit-image==0.21.0
 spandrel==0.1.6
 tomesd==0.1.3

From baaf39b6f92f24275a1b264a634514bac571dfae Mon Sep 17 00:00:00 2001
From: AUTOMATIC1111 <16777216c@gmail.com>
Date: Mon, 29 Jan 2024 10:20:27 +0300
Subject: [PATCH 02/10] fix the typo -- thanks Cyberbeing

---
 modules/sd_samplers_cfg_denoiser.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/modules/sd_samplers_cfg_denoiser.py b/modules/sd_samplers_cfg_denoiser.py
index ef237396..941dff4b 100644
--- a/modules/sd_samplers_cfg_denoiser.py
+++ b/modules/sd_samplers_cfg_denoiser.py
@@ -94,7 +94,7 @@ class CFGDenoiser(torch.nn.Module):
 
     def pad_cond_uncond(self, cond, uncond):
         empty = shared.sd_model.cond_stage_model_empty_prompt
-        num_repeats = (cond.shape[1] - cond.shape[1]) // empty.shape[1]
+        num_repeats = (cond.shape[1] - uncond.shape[1]) // empty.shape[1]
 
         if num_repeats < 0:
             cond = pad_cond(cond, -num_repeats, empty)

From ec124607f47371a6cfd61a795f86a7f1cbd44651 Mon Sep 17 00:00:00 2001
From: wangshuai09 <391746016@qq.com>
Date: Sat, 27 Jan 2024 17:21:32 +0800
Subject: [PATCH 03/10] Add NPU Support

---
 modules/devices.py                            |  9 +++--
 modules/initialize.py                         |  6 +++-
 modules/npu_specific.py                       | 34 +++++++++++++++++++
 .../textual_inversion/textual_inversion.py    |  4 +++
 requirements.txt                              |  4 +++
 requirements_versions.txt                     |  4 +++
 webui.sh                                      |  4 +++
 7 files changed, 62 insertions(+), 3 deletions(-)
 create mode 100644 modules/npu_specific.py

diff --git a/modules/devices.py b/modules/devices.py
index ea1f712f..f1e56501 100644
--- a/modules/devices.py
+++ b/modules/devices.py
@@ -3,7 +3,7 @@ import contextlib
 from functools import lru_cache
 
 import torch
-from modules import errors, shared
+from modules import errors, shared, npu_specific
 
 if sys.platform == "darwin":
     from modules import mac_specific
@@ -40,6 +40,9 @@ def get_optimal_device_name():
     if has_xpu():
         return xpu_specific.get_xpu_device_string()
 
+    if npu_specific.has_npu:
+        return npu_specific.get_npu_device_string()
+
     return "cpu"
 
 
@@ -67,6 +70,9 @@ def torch_gc():
     if has_xpu():
         xpu_specific.torch_xpu_gc()
 
+    if npu_specific.has_npu:
+        npu_specific.torch_npu_gc()
+
 
 def enable_tf32():
     if torch.cuda.is_available():
@@ -164,4 +170,3 @@ def first_time_calculation():
     x = torch.zeros((1, 1, 3, 3)).to(device, dtype)
     conv2d = torch.nn.Conv2d(1, 1, (3, 3)).to(device, dtype)
     conv2d(x)
-
diff --git a/modules/initialize.py b/modules/initialize.py
index ac95fc6f..3285cc3c 100644
--- a/modules/initialize.py
+++ b/modules/initialize.py
@@ -143,13 +143,17 @@ def initialize_rest(*, reload_script_modules=False):
         its optimization may be None because the list of optimizaers has neet been filled
         by that time, so we apply optimization again.
         """
+        from modules import devices
+        # Work around due to bug in torch_npu, revert me after fixed, @see https://gitee.com/ascend/pytorch/issues/I8KECW?from=project-issue
+        if devices.npu_specific.has_npu:
+            import torch
+            torch.npu.set_device(0)
 
         shared.sd_model  # noqa: B018
 
         if sd_hijack.current_optimizer is None:
             sd_hijack.apply_optimizations()
 
-        from modules import devices
         devices.first_time_calculation()
     if not shared.cmd_opts.skip_load_model_at_start:
         Thread(target=load_model).start()
diff --git a/modules/npu_specific.py b/modules/npu_specific.py
new file mode 100644
index 00000000..d8aebf9c
--- /dev/null
+++ b/modules/npu_specific.py
@@ -0,0 +1,34 @@
+import importlib
+import torch
+
+from modules import shared
+
+
+def check_for_npu():
+    if importlib.util.find_spec("torch_npu") is None:
+        return False
+    import torch_npu
+    torch_npu.npu.set_device(0)
+
+    try:
+        # Will raise a RuntimeError if no NPU is found
+        _ = torch.npu.device_count()
+        return torch.npu.is_available()
+    except RuntimeError:
+        return False
+
+
+def get_npu_device_string():
+    if shared.cmd_opts.device_id is not None:
+        return f"npu:{shared.cmd_opts.device_id}"
+    return "npu:0"
+
+
+def torch_npu_gc():
+    # Work around due to bug in torch_npu, revert me after fixed, @see https://gitee.com/ascend/pytorch/issues/I8KECW?from=project-issue
+    torch.npu.set_device(0)
+    with torch.npu.device(get_npu_device_string()):
+        torch.npu.empty_cache()
+
+
+has_npu = check_for_npu()
diff --git a/modules/textual_inversion/textual_inversion.py b/modules/textual_inversion/textual_inversion.py
index 04dda585..9c062503 100644
--- a/modules/textual_inversion/textual_inversion.py
+++ b/modules/textual_inversion/textual_inversion.py
@@ -151,6 +151,10 @@ class EmbeddingDatabase:
         return embedding
 
     def get_expected_shape(self):
+        # workaround
+        if devices.npu_specific.has_npu:
+            import torch
+            torch.npu.set_device(0)
         vec = shared.sd_model.cond_stage_model.encode_embedding_init_text(",", 1)
         return vec.shape[1]
 
diff --git a/requirements.txt b/requirements.txt
index 80b43845..4537402b 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -5,6 +5,8 @@ accelerate
 basicsr
 blendmodes
 clean-fid
+cloudpickle
+decorator
 einops
 fastapi>=0.90.1
 gfpgan
@@ -26,9 +28,11 @@ resize-right
 
 safetensors
 scikit-image>=0.19
+synr==0.5.0
 timm
 tomesd
 torch
 torchdiffeq
 torchsde
+tornado
 transformers==4.30.2
diff --git a/requirements_versions.txt b/requirements_versions.txt
index cb7403a9..95515b55 100644
--- a/requirements_versions.txt
+++ b/requirements_versions.txt
@@ -4,6 +4,8 @@ accelerate==0.21.0
 basicsr==1.4.2
 blendmodes==2022
 clean-fid==0.1.35
+cloudpickle==3.0.0
+decorator==5.1.1
 einops==0.4.1
 fastapi==0.94.0
 gfpgan==1.3.8
@@ -23,10 +25,12 @@ realesrgan==0.3.0
 resize-right==0.0.2
 safetensors==0.3.1
 scikit-image==0.21.0
+synr==0.5.0
 timm==0.9.2
 tomesd==0.1.3
 torch
 torchdiffeq==0.2.3
 torchsde==0.2.6
+tornado==6.4
 transformers==4.30.2
 httpx==0.24.1
diff --git a/webui.sh b/webui.sh
index cff43327..3f6e87fd 100755
--- a/webui.sh
+++ b/webui.sh
@@ -159,6 +159,10 @@ then
     if echo "$gpu_info" | grep -q "AMD" && [[ -z "${TORCH_COMMAND}" ]]
     then
         export TORCH_COMMAND="pip install torch==2.0.1+rocm5.4.2 torchvision==0.15.2+rocm5.4.2 --index-url https://download.pytorch.org/whl/rocm5.4.2"
+    elif echo "$gpu_info" | grep -q "Huawei" && [[ -z "${TORCH_COMMAND}" ]]
+    then
+        export TORCH_COMMAND="pip install torch==2.1.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu; pip install torch_npu"
+    
     fi
 fi
 

From 750dd6014a45397979cad42a74634451d0861581 Mon Sep 17 00:00:00 2001
From: Kohaku-Blueleaf <59680068+KohakuBlueleaf@users.noreply.github.com>
Date: Mon, 29 Jan 2024 22:27:53 +0800
Subject: [PATCH 04/10] Fix potential bugs

---
 modules/devices.py | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/modules/devices.py b/modules/devices.py
index dfffaf24..60f7d6d7 100644
--- a/modules/devices.py
+++ b/modules/devices.py
@@ -141,7 +141,12 @@ def manual_cast_forward(target_dtype):
             args = [arg.to(target_dtype) if isinstance(arg, torch.Tensor) else arg for arg in args]
             kwargs = {k: v.to(target_dtype) if isinstance(v, torch.Tensor) else v for k, v in kwargs.items()}
 
-        org_dtype = torch_utils.get_param(self).dtype
+        org_dtype = target_dtype
+        for param in self.parameters():
+            if param.dtype != target_dtype:
+                org_dtype = param.dtype
+                break
+
         if org_dtype != target_dtype:
             self.to(target_dtype)
         result = self.org_forward(*args, **kwargs)
@@ -170,7 +175,7 @@ def manual_cast(target_dtype):
             continue
         applied = True
         org_forward = module_type.forward
-        if module_type == torch.nn.MultiheadAttention and has_xpu():
+        if module_type == torch.nn.MultiheadAttention:
             module_type.forward = manual_cast_forward(torch.float32)
         else:
             module_type.forward = manual_cast_forward(target_dtype)

From 6e7f0860f7ae4a0ce59f9416fb9b2f3bcab44f1d Mon Sep 17 00:00:00 2001
From: Kohaku-Blueleaf <59680068+KohakuBlueleaf@users.noreply.github.com>
Date: Mon, 29 Jan 2024 22:46:43 +0800
Subject: [PATCH 05/10] linting

---
 modules/devices.py | 1 -
 1 file changed, 1 deletion(-)

diff --git a/modules/devices.py b/modules/devices.py
index 60f7d6d7..8f49f7a4 100644
--- a/modules/devices.py
+++ b/modules/devices.py
@@ -4,7 +4,6 @@ from functools import lru_cache
 
 import torch
 from modules import errors, shared
-from modules import torch_utils
 
 if sys.platform == "darwin":
     from modules import mac_specific

From d243e24f539d717b221992e894a5db5a321bf3cd Mon Sep 17 00:00:00 2001
From: Kohaku-Blueleaf <59680068+KohakuBlueleaf@users.noreply.github.com>
Date: Mon, 29 Jan 2024 22:49:45 +0800
Subject: [PATCH 06/10] Try to reverse the dtype checking mechanism

---
 modules/devices.py | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)

diff --git a/modules/devices.py b/modules/devices.py
index 8f49f7a4..f9648e9a 100644
--- a/modules/devices.py
+++ b/modules/devices.py
@@ -4,6 +4,7 @@ from functools import lru_cache
 
 import torch
 from modules import errors, shared
+from modules import torch_utils
 
 if sys.platform == "darwin":
     from modules import mac_specific
@@ -140,11 +141,7 @@ def manual_cast_forward(target_dtype):
             args = [arg.to(target_dtype) if isinstance(arg, torch.Tensor) else arg for arg in args]
             kwargs = {k: v.to(target_dtype) if isinstance(v, torch.Tensor) else v for k, v in kwargs.items()}
 
-        org_dtype = target_dtype
-        for param in self.parameters():
-            if param.dtype != target_dtype:
-                org_dtype = param.dtype
-                break
+        org_dtype = torch_utils.get_param(self).dtype
 
         if org_dtype != target_dtype:
             self.to(target_dtype)

From f9ba7e648ad5bf7dbdf2b95fa207936179bf784e Mon Sep 17 00:00:00 2001
From: Kohaku-Blueleaf <59680068+KohakuBlueleaf@users.noreply.github.com>
Date: Mon, 29 Jan 2024 22:54:12 +0800
Subject: [PATCH 07/10] Revert "Try to reverse the dtype checking mechanism"

This reverts commit d243e24f539d717b221992e894a5db5a321bf3cd.
---
 modules/devices.py | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/modules/devices.py b/modules/devices.py
index f9648e9a..8f49f7a4 100644
--- a/modules/devices.py
+++ b/modules/devices.py
@@ -4,7 +4,6 @@ from functools import lru_cache
 
 import torch
 from modules import errors, shared
-from modules import torch_utils
 
 if sys.platform == "darwin":
     from modules import mac_specific
@@ -141,7 +140,11 @@ def manual_cast_forward(target_dtype):
             args = [arg.to(target_dtype) if isinstance(arg, torch.Tensor) else arg for arg in args]
             kwargs = {k: v.to(target_dtype) if isinstance(v, torch.Tensor) else v for k, v in kwargs.items()}
 
-        org_dtype = torch_utils.get_param(self).dtype
+        org_dtype = target_dtype
+        for param in self.parameters():
+            if param.dtype != target_dtype:
+                org_dtype = param.dtype
+                break
 
         if org_dtype != target_dtype:
             self.to(target_dtype)

From c4255d12f7531725e591160e1cfe47d7a2fc0f02 Mon Sep 17 00:00:00 2001
From: w-e-w <40751091+w-e-w@users.noreply.github.com>
Date: Wed, 31 Jan 2024 04:36:11 +0900
Subject: [PATCH 08/10] add tooltip create_submit_box

---
 modules/ui_toprow.py | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/modules/ui_toprow.py b/modules/ui_toprow.py
index fbe705be..457fbf52 100644
--- a/modules/ui_toprow.py
+++ b/modules/ui_toprow.py
@@ -96,9 +96,9 @@ class Toprow:
         with gr.Row(elem_id=f"{self.id_part}_generate_box", elem_classes=["generate-box"] + (["generate-box-compact"] if self.is_compact else []), render=not self.is_compact) as submit_box:
             self.submit_box = submit_box
 
-            self.interrupt = gr.Button('Interrupt', elem_id=f"{self.id_part}_interrupt", elem_classes="generate-box-interrupt")
-            self.skip = gr.Button('Skip', elem_id=f"{self.id_part}_skip", elem_classes="generate-box-skip")
-            self.submit = gr.Button('Generate', elem_id=f"{self.id_part}_generate", variant='primary')
+            self.interrupt = gr.Button('Interrupt', elem_id=f"{self.id_part}_interrupt", elem_classes="generate-box-interrupt", tooltip="End generation immediately or after completing current batch")
+            self.skip = gr.Button('Skip', elem_id=f"{self.id_part}_skip", elem_classes="generate-box-skip", tooltip="Stop generation of current batch and continues onto next batch")
+            self.submit = gr.Button('Generate', elem_id=f"{self.id_part}_generate", variant='primary', tooltip="Right click generate forever menu")
 
             self.skip.click(
                 fn=lambda: shared.state.skip(),

From cc3f604310458eed7d26456c1b3934d582283ffe Mon Sep 17 00:00:00 2001
From: wangshuai09 <391746016@qq.com>
Date: Wed, 31 Jan 2024 10:46:53 +0800
Subject: [PATCH 09/10] Update

---
 modules/devices.py                             | 7 +++++++
 modules/initialize.py                          | 5 +----
 modules/launch_utils.py                        | 8 ++++++++
 modules/npu_specific.py                        | 5 +----
 modules/textual_inversion/textual_inversion.py | 5 +----
 requirements.txt                               | 4 ----
 requirements_npu.txt                           | 4 ++++
 requirements_versions.txt                      | 4 ----
 8 files changed, 22 insertions(+), 20 deletions(-)
 create mode 100644 requirements_npu.txt

diff --git a/modules/devices.py b/modules/devices.py
index c737162a..28c0c54d 100644
--- a/modules/devices.py
+++ b/modules/devices.py
@@ -88,9 +88,16 @@ def torch_gc():
         xpu_specific.torch_xpu_gc()
 
     if npu_specific.has_npu:
+        torch_npu_set_device()
         npu_specific.torch_npu_gc()
 
 
+def torch_npu_set_device():
+    # Work around due to bug in torch_npu, revert me after fixed, @see https://gitee.com/ascend/pytorch/issues/I8KECW?from=project-issue
+    if npu_specific.has_npu:
+        torch.npu.set_device(0)
+
+
 def enable_tf32():
     if torch.cuda.is_available():
 
diff --git a/modules/initialize.py b/modules/initialize.py
index cc34fd6f..f7313ff4 100644
--- a/modules/initialize.py
+++ b/modules/initialize.py
@@ -143,10 +143,7 @@ def initialize_rest(*, reload_script_modules=False):
         by that time, so we apply optimization again.
         """
         from modules import devices
-        # Work around due to bug in torch_npu, revert me after fixed, @see https://gitee.com/ascend/pytorch/issues/I8KECW?from=project-issue
-        if devices.npu_specific.has_npu:
-            import torch
-            torch.npu.set_device(0)
+        devices.torch_npu_set_device()
 
         shared.sd_model  # noqa: B018
 
diff --git a/modules/launch_utils.py b/modules/launch_utils.py
index 3ff4576a..107c72b0 100644
--- a/modules/launch_utils.py
+++ b/modules/launch_utils.py
@@ -338,6 +338,7 @@ def prepare_environment():
             torch_index_url = os.environ.get('TORCH_INDEX_URL', "https://pytorch-extension.intel.com/release-whl/stable/xpu/us/")
             torch_command = os.environ.get('TORCH_COMMAND', f"pip install torch==2.0.0a0 intel-extension-for-pytorch==2.0.110+gitba7f6c1 --extra-index-url {torch_index_url}")
     requirements_file = os.environ.get('REQS_FILE', "requirements_versions.txt")
+    requirements_file_for_npu = os.environ.get('REQS_FILE_FOR_NPU', "requirements_npu.txt")
 
     xformers_package = os.environ.get('XFORMERS_PACKAGE', 'xformers==0.0.23.post1')
     clip_package = os.environ.get('CLIP_PACKAGE', "https://github.com/openai/CLIP/archive/d50d76daa670286dd6cacf3bcd80b5e4823fc8e1.zip")
@@ -421,6 +422,13 @@ def prepare_environment():
         run_pip(f"install -r \"{requirements_file}\"", "requirements")
         startup_timer.record("install requirements")
 
+    if not os.path.isfile(requirements_file_for_npu):
+        requirements_file_for_npu = os.path.join(script_path, requirements_file_for_npu)
+
+    if "torch_npu" in torch_command and not requirements_met(requirements_file_for_npu):
+        run_pip(f"install -r \"{requirements_file_for_npu}\"", "requirements_for_npu")
+        startup_timer.record("install requirements_for_npu")
+
     if not args.skip_install:
         run_extensions_installers(settings_file=args.ui_settings_file)
 
diff --git a/modules/npu_specific.py b/modules/npu_specific.py
index d8aebf9c..94100691 100644
--- a/modules/npu_specific.py
+++ b/modules/npu_specific.py
@@ -8,11 +8,10 @@ def check_for_npu():
     if importlib.util.find_spec("torch_npu") is None:
         return False
     import torch_npu
-    torch_npu.npu.set_device(0)
 
     try:
         # Will raise a RuntimeError if no NPU is found
-        _ = torch.npu.device_count()
+        _ = torch_npu.npu.device_count()
         return torch.npu.is_available()
     except RuntimeError:
         return False
@@ -25,8 +24,6 @@ def get_npu_device_string():
 
 
 def torch_npu_gc():
-    # Work around due to bug in torch_npu, revert me after fixed, @see https://gitee.com/ascend/pytorch/issues/I8KECW?from=project-issue
-    torch.npu.set_device(0)
     with torch.npu.device(get_npu_device_string()):
         torch.npu.empty_cache()
 
diff --git a/modules/textual_inversion/textual_inversion.py b/modules/textual_inversion/textual_inversion.py
index d16e3b9a..6d815c0b 100644
--- a/modules/textual_inversion/textual_inversion.py
+++ b/modules/textual_inversion/textual_inversion.py
@@ -150,10 +150,7 @@ class EmbeddingDatabase:
         return embedding
 
     def get_expected_shape(self):
-        # workaround
-        if devices.npu_specific.has_npu:
-            import torch
-            torch.npu.set_device(0)
+        devices.torch_npu_set_device()
         vec = shared.sd_model.cond_stage_model.encode_embedding_init_text(",", 1)
         return vec.shape[1]
 
diff --git a/requirements.txt b/requirements.txt
index d1e4ede9..731a1be7 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -4,8 +4,6 @@ accelerate
 
 blendmodes
 clean-fid
-cloudpickle
-decorator
 einops
 facexlib
 fastapi>=0.90.1
@@ -26,10 +24,8 @@ resize-right
 
 safetensors
 scikit-image>=0.19
-synr==0.5.0
 tomesd
 torch
 torchdiffeq
 torchsde
-tornado
 transformers==4.30.2
diff --git a/requirements_npu.txt b/requirements_npu.txt
new file mode 100644
index 00000000..5e6a4364
--- /dev/null
+++ b/requirements_npu.txt
@@ -0,0 +1,4 @@
+cloudpickle
+decorator
+synr==0.5.0
+tornado
diff --git a/requirements_versions.txt b/requirements_versions.txt
index 1c66cd8c..5e30b5ea 100644
--- a/requirements_versions.txt
+++ b/requirements_versions.txt
@@ -3,8 +3,6 @@ Pillow==9.5.0
 accelerate==0.21.0
 blendmodes==2022
 clean-fid==0.1.35
-cloudpickle==3.0.0
-decorator==5.1.1
 einops==0.4.1
 facexlib==0.3.0
 fastapi==0.94.0
@@ -23,12 +21,10 @@ pytorch_lightning==1.9.4
 resize-right==0.0.2
 safetensors==0.4.2
 scikit-image==0.21.0
-synr==0.5.0
 spandrel==0.1.6
 tomesd==0.1.3
 torch
 torchdiffeq==0.2.3
 torchsde==0.2.6
-tornado==6.4
 transformers==4.30.2
 httpx==0.24.1

From 74b214a92a2959948dcd05a78b7380e046163871 Mon Sep 17 00:00:00 2001
From: Cyberbeing <cyber.spamage@gmail.com>
Date: Mon, 29 Jan 2024 02:06:50 -0800
Subject: [PATCH 10/10] Fix potential autocast NaNs in image upscale

---
 modules/upscaler_utils.py | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/modules/upscaler_utils.py b/modules/upscaler_utils.py
index afed8b40..b5e5a80c 100644
--- a/modules/upscaler_utils.py
+++ b/modules/upscaler_utils.py
@@ -6,7 +6,7 @@ import torch
 import tqdm
 from PIL import Image
 
-from modules import images, shared, torch_utils
+from modules import devices, images, shared, torch_utils
 
 logger = logging.getLogger(__name__)
 
@@ -44,7 +44,8 @@ def upscale_pil_patch(model, img: Image.Image) -> Image.Image:
     with torch.no_grad():
         tensor = pil_image_to_torch_bgr(img).unsqueeze(0)  # add batch dimension
         tensor = tensor.to(device=param.device, dtype=param.dtype)
-        return torch_bgr_to_pil_image(model(tensor))
+        with devices.without_autocast():
+            return torch_bgr_to_pil_image(model(tensor))
 
 
 def upscale_with_model(