multiple fixes and update requirements

Signed-off-by: Vladimir Mandic <mandic00@live.com>

CHANGELOG.md

@@ -8,6 +8,9 @@
   - detailer with faceid modules  
   - detailer restore pipeline before run  
   - fix python 3.9 compatibility  
+  - handle invalid `triton` with `torch==2.6.0`  
+  - correct library import order  
+  - update requirements  
 
 ## Update for 2025-01-29
 

installer.py

@@ -524,7 +524,13 @@ def install_cuda():
     t_start = time.time()
     log.info('CUDA: nVidia toolkit detected')
     ts('cuda', t_start)
-    return os.environ.get('TORCH_COMMAND', 'torch==2.5.1+cu124 torchvision==0.20.1+cu124 --index-url https://download.pytorch.org/whl/cu124')
+    if args.use_nightly:
+        cmd = os.environ.get('TORCH_COMMAND', '--pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/cu126')
+    else:
+        cmd = os.environ.get('TORCH_COMMAND', 'torch==2.5.1+cu124 torchvision==0.20.1+cu124 --index-url https://download.pytorch.org/whl/cu124')
+        # TODO torch no triton for torch==2.6
+        # TODO blackwell requires cuda==12.8
+    return cmd
 
 
 def install_rocm_zluda():
@@ -1429,6 +1435,7 @@ def add_args(parser):
     group_compute.add_argument("--use-openvino", default=os.environ.get("SD_USEOPENVINO",False), action='store_true', help="Use Intel OpenVINO backend, default: %(default)s")
     group_compute.add_argument("--use-ipex", default=os.environ.get("SD_USEIPEX",False), action='store_true', help="Force use Intel OneAPI XPU backend, default: %(default)s")
     group_compute.add_argument("--use-cuda", default=os.environ.get("SD_USECUDA",False), action='store_true', help="Force use nVidia CUDA backend, default: %(default)s")
+    group_compute.add_argument("--use-nightly", default=os.environ.get("SD_USENIGHLY",False), action='store_true', help="Force use nightly torch builds, default: %(default)s")
     group_compute.add_argument("--use-rocm", default=os.environ.get("SD_USEROCM",False), action='store_true', help="Force use AMD ROCm backend, default: %(default)s")
     group_compute.add_argument('--use-zluda', default=os.environ.get("SD_USEZLUDA", False), action='store_true', help="Force use ZLUDA, AMD GPUs only, default: %(default)s")
     group_compute.add_argument("--use-xformers", default=os.environ.get("SD_USEXFORMERS",False), action='store_true', help="Force use xFormers cross-optimization, default: %(default)s")

modules/loader.py

@@ -42,6 +42,11 @@ if ".dev" in torch.__version__ or "+git" in torch.__version__:
     torch.__long_version__ = torch.__version__
     torch.__version__ = re.search(r'[\d.]+[\d]', torch.__version__).group(0)
 timer.startup.record("torch")
+try:
+    import bitsandbytes # pylint: disable=W0611,C0411
+except Exception:
+    from diffusers.utils import import_utils
+    import_utils._bitsandbytes_available = False # pylint: disable=protected-access
 
 import transformers # pylint: disable=W0611,C0411
 from transformers import logging as transformers_logging # pylint: disable=W0611,C0411

requirements.txt

@@ -32,7 +32,7 @@ invisible-watermark
 pi-heif
 
 # versioned
-safetensors==0.5.0
+safetensors==0.5.2
 tensordict==0.1.2
 peft==0.14.0
 httpx==0.24.1
@@ -41,20 +41,20 @@ torchsde==0.2.6
 antlr4-python3-runtime==4.9.3
 requests==2.32.3
 tqdm==4.66.5
-accelerate==1.2.1
+accelerate==1.3.0
 opencv-contrib-python-headless==4.9.0.80
 einops==0.4.1
 gradio==3.43.2
-huggingface_hub==0.27.0
+huggingface_hub==0.28.1
 numexpr==2.8.8
 numpy==1.26.4
 numba==0.59.1
 protobuf==4.25.3
 pytorch_lightning==1.9.4
 tokenizers==0.21.0
-transformers==4.47.1
+transformers==4.48.1
 urllib3==1.26.19
-Pillow==10.4.0
+Pillow==11.1.0
 timm==0.9.16
 pydantic==1.10.15
 pyparsing==3.1.4

webui.py

@@ -9,8 +9,8 @@ import logging
 import importlib
 import contextlib
 from threading import Thread
-import modules.hashes
 import modules.loader
+import modules.hashes
 
 from installer import log, git_commit, custom_excepthook
 from modules import timer, paths, shared, extensions, gr_tempdir, modelloader