GPU: NVIDIA GeForce RTX 5090, 32607 MiB, compute capability 12.0 (sm_120), driver 610.43.02
OS: Linux 6.12.95 (Debian 13), 91 GiB RAM
Python 3.12.13 (author: 3.10)
torch 2.13.0+cu130, CUDA 13.0 (author: torch 2.6.0+cu124)
torchvision 0.28.0 (author: 0.21.0)
triton 3.7.1
NATTEN 0.17.5 (author pin, rebuilt from source for sm_120, see natten-patch-0311.diff)
flex_gemm 1.0.0, cumesh 0.0.1, o_voxel 0.0.1, nvdiffrast 0.4.0 (CUDA extensions built on the box for sm_120)
numba 0.65.1 (author pin), llvmlite 0.47.0
transformers 4.57.3, diffusers 0.37.1, accelerate 1.13.0, safetensors 0.7.0
Attention: PyTorch SDPA (log: attention=sdpa; sparse_attention=sdpa; low_vram=False)
Sparse conv backend: flex_gemm (log: [SPARSE] Conv backend: flex_gemm)
Source: pip list and environment capture of 3 Nov 2026 18:27 (torch/python/driver), logs/*.log