[Windows] RTX 5070 Ti (Blackwell sm_120) - setup and deployment notes

[loongmiaow-pixel]

Environment

• GPU: NVIDIA GeForce RTX 5070 Ti Laptop GPU (Blackwell, compute capability 12.0)
• Driver: 595.79 (CUDA 13.2)
• OS: Windows 11
• Python: 3.14
• transformers: [latest]

Problem

transformers on RTX 5070 Ti requires workarounds:

• TORCH_CUDA_ARCH_LIST=12.0 required for Blackwell
• Model loading may fail or fall back to CPU without explicit arch
• CUDA_VISIBLE_DEVICES=-1 system env var blocks GPU detection

Solution

$env:TORCH_CUDA_ARCH_LIST = "12.0"
$env:CUDA_VISIBLE_DEVICES = "0"
$env:CUDA_MODULE_LOADING = "LAZY"

Question

Would it be helpful to add RTX 50 series notes to the GPU documentation? I can contribute a PR with Windows troubleshooting.