diff --git a/CHANGELOG.md b/CHANGELOG.md
index 6a9831dfb..7bc2a5401 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -43,6 +43,7 @@
   - **reference models** additional example images, thanks @liutyi  
   - **video** support for configurable multi-stage models such as WAN-2.2-14B  
   - **video** new LTX model selection  
+  - replace `pynvml` with `nvidia-ml-py` for gpu monitoring  
 - **Fixes**
   - framepack: add explicit hf-login before framepack load  
   - benchmark: remove forced sampler from system info benchmark  
diff --git a/installer.py b/installer.py
index 97a502d40..f5e3bad73 100644
--- a/installer.py
+++ b/installer.py
@@ -877,6 +877,7 @@ def install_torch_addons():
         install('pillow-jxl-plugin==1.3.4', 'pillow-jxl-plugin')
     if not args.experimental:
         uninstall('wandb', quiet=True)
+        uninstall('pynvml', quiet=True)
     ts('addons', t_start)
 
 
diff --git a/modules/processing_diffusers.py b/modules/processing_diffusers.py
index 1d88ab2a0..4de4e0a47 100644
--- a/modules/processing_diffusers.py
+++ b/modules/processing_diffusers.py
@@ -80,13 +80,13 @@ def process_pre(p: processing.StableDiffusionProcessing):
         pag.apply(p)
         cfgzero.apply(p)
         linfusion.apply(shared.sd_model)
+        cachedit.apply_cache_dit(shared.sd_model)
 
         # apply-only
         sd_hijack_freeu.apply_freeu(p)
         transformer_cache.set_cache()
         para_attention.apply_first_block_cache()
         teacache.apply_teacache(p)
-        cachedit.apply_cache_dit(shared.sd_model)
     except Exception as e:
         shared.log.error(f'Processing apply: {e}')
         errors.display(e, 'apply')