Switch torch wheels from cu126 to cu128 for Blackwell sm_120

Runtime traceback showed the allocated GPU is an NVIDIA RTX PRO 6000
Blackwell with CUDA capability sm_120, but cu126 wheels only support
sm_50..sm_90 — every kernel launch failed with cudaErrorNoKernelImage,
surfaced as torch.AcceleratorError. cu128 wheels include sm_120 SASS,
which matches the hardware ZeroGPU is allocating.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

requirements.txt

@@ -1,5 +1,5 @@
 --index-url https://pypi.org/simple
---extra-index-url https://download.pytorch.org/whl/cu126
+--extra-index-url https://download.pytorch.org/whl/cu128
 
 #gradio==4.44.1
 #gradio_image_prompter
@@ -28,9 +28,9 @@ cvcuda_cu12
 gltflib
 https://huggingface.co/spaces/VAST-AI/TripoSG/resolve/main/diso-0.1.4-cp310-cp310-linux_x86_64.whl?download=true
 
-https://download.pytorch.org/whl/cu126/torch-2.11.0%2Bcu126-cp310-cp310-manylinux_2_28_x86_64.whl
-https://download.pytorch.org/whl/cu126/torchvision-0.26.0%2Bcu126-cp310-cp310-manylinux_2_28_x86_64.whl
-https://download.pytorch.org/whl/cu126/torchaudio-2.11.0%2Bcu126-cp310-cp310-manylinux_2_28_x86_64.whl
+https://download.pytorch.org/whl/cu128/torch-2.11.0%2Bcu128-cp310-cp310-manylinux_2_28_x86_64.whl
+https://download.pytorch.org/whl/cu128/torchvision-0.26.0%2Bcu128-cp310-cp310-manylinux_2_28_x86_64.whl
+https://download.pytorch.org/whl/cu128/torchaudio-2.11.0%2Bcu128-cp310-cp310-manylinux_2_28_x86_64.whl
 
---find-links https://data.pyg.org/whl/torch-2.11.0+cu126.html
+--find-links https://data.pyg.org/whl/torch-2.11.0+cu128.html
 torch-cluster