Added option to use cudnn as backend for pytorch, this should help fixing an issue with nvidia 16xx cards getting a black or green square instead of a proper image. (#1699)

configs/webui/webui_streamlit.yaml

@@ -59,6 +59,7 @@ general:
     no_half: False
     use_float16: False
     precision: "autocast"
+    use_cudnn: False
     optimized: False
     optimized_turbo: False
     optimized_config: "optimizedSD/v1-inference.yaml"

environment.yaml

@@ -23,13 +23,13 @@ channels:
 dependencies:
   - conda-forge::nodejs=18.11.0
   - yarn=1.22.19
-  - cudatoolkit=11.3
+  - cudatoolkit=11.7
   - git
-  - numpy=1.22.3
+  - numpy=1.23.3
   - pip=20.3
   - python=3.8.5
-  - pytorch=1.11.0
+  - pytorch=1.13.0
   - scikit-image=0.19.2
-  - torchvision=0.12.0
+  - torchvision=0.14.0
   - pip:
     - -r requirements.txt

scripts/sd_utils/__init__.py

@@ -95,6 +95,12 @@ except ImportError as e:
 # remove all the annoying python warnings.
 shutup.please()
 
+# the following lines should help fixing an issue with nvidia 16xx cards.
+if "defaults" in st.session_state:
+    if st.session_state["defaults"].general.use_cudnn:
+        torch.backends.cudnn.benchmark = True
+        torch.backends.cudnn.enabled = True 
+
 try:
     # this silences the annoying "Some weights of the model checkpoint were not used when initializing..." message at start.
     from transformers import logging
@@ -1613,6 +1619,10 @@ def ModelLoader(models,load=False,unload=False,imgproc_realesrgan_model_name='Re
 #
 @retry(tries=5)
 def generation_callback(img, i=0):
+    
+    # try to do garbage collection before decoding the image
+    torch_gc()
+    
     if "update_preview_frequency" not in st.session_state:
         raise StopException