Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

Restore what actually fails when NPP is missing from the torchcodec install (#10612)

Daniel Han01715683eea3
diff --git a/studio/install_python_stack.py b/studio/install_python_stack.py
index b01b59a527c..147bd8641d9 100644
--- a/studio/install_python_stack.py
+++ b/studio/install_python_stack.py
@@ -8138,7 +8138,8 @@ def install_python_stack() -> int:
)
elif _codec_index:
# torchcodec's CUDA build dlopens libnppicc and libnppc, and NPP is not in torch's
- # dependency set, so a --no-deps install from a cuNNN index imports and then fails.
+ # dependency set, so an older cuNNN wheel installs fine under --no-deps and then
+ # fails to import. 0.12+ no longer links NPP, so this only guards the older pins.
# _npp_requirement spells the name, which stops being suffixed after 12.
_npp_major = _cuda_major_for_npp(_codec_torch_ver, _codec_index)
if _codec_fellback: