Use CUDA 12.8 PyTorch for ZeroGPU Blackwell GPUs

ZeroGPU now runs on NVIDIA RTX PRO 6000 Blackwell (sm_120), which the cu126
torch wheels don't support -> "no kernel image is available for execution on
the device" at GPU call time.

- install.sh: install torch 2.8.0 + torchvision 0.23.0 from the cu128 index
- my_req.txt: drop torch/torchvision/triton and all nvidia-*-cu12 (12.6.x)
pins so the cu128 torch owns the CUDA stack and mmcv/mmgen compile against it

install.sh

@@ -7,11 +7,15 @@ set -e
 # it). Keep a pre-82 setuptools, plus the tools needed to build native exts.
 pip install "setuptools<82" wheel ninja
 
-# Install torch first, at the SAME versions pinned in my_req.txt, so mmcv/mmgen
-# are compiled against the exact torch used at runtime (otherwise their compiled
-# extensions raise ABI / "undefined symbol" errors), and torch isn't downloaded
-# twice.
-pip install torch==2.7.1 torchvision==0.22.1
+# Install torch first. ZeroGPU now runs on NVIDIA RTX PRO 6000 Blackwell
+# (compute capability sm_120), which is ONLY supported by CUDA 12.8+ builds of
+# PyTorch -- the default cu126 wheels raise "no kernel image is available for
+# execution on the device". Use the cu128 build (torch 2.8.x is the minimum
+# version supported by HF ZeroGPU). torch/torchvision/triton/nvidia-* are
+# intentionally NOT pinned in my_req.txt so this install owns the CUDA stack,
+# and mmcv/mmgen below compile against this exact torch.
+pip install --index-url https://download.pytorch.org/whl/cu128 \
+  torch==2.8.0 torchvision==0.23.0
 
 # --- piFlow (lakonlab) -------------------------------------------------------
 # Install ONLY the `lakonlab` package, not piFlow's own dependencies: piFlow's

my_req.txt

@@ -1,4 +1,3 @@
-torchvision==0.22.1
 numpy==2.2.6
 accelerate==1.9.0
 diffusers @ git+https://github.com/huggingface/diffusers@7a2b78bf0f788d311cc96b61e660a8e13e3b1e63
@@ -72,20 +71,6 @@ multidict==6.7.0
 multiprocess==0.70.16
 networkx==3.4.2
 ninja==1.13.0
-nvidia-cublas-cu12==12.6.4.1
-nvidia-cuda-cupti-cu12==12.6.80
-nvidia-cuda-nvrtc-cu12==12.6.77
-nvidia-cuda-runtime-cu12==12.6.77
-nvidia-cudnn-cu12==9.5.1.17
-nvidia-cufft-cu12==11.3.0.4
-nvidia-cufile-cu12==1.11.1.6
-nvidia-curand-cu12==10.3.7.77
-nvidia-cusolver-cu12==11.7.1.2
-nvidia-cusparse-cu12==12.5.4.2
-nvidia-cusparselt-cu12==0.6.3
-nvidia-nccl-cu12==2.26.2
-nvidia-nvjitlink-cu12==12.6.85
-nvidia-nvtx-cu12==12.6.77
 omegaconf==2.3.0
 onnxruntime==1.22.1
 opencv-python==4.12.0.88
@@ -123,10 +108,7 @@ sympy==1.14.0
 timm==1.0.19
 tokenizers==0.21.4
 tomlkit==0.13.3
-torch==2.7.1
-torchvision==0.22.1
 tqdm==4.67.1
-triton==3.3.1
 typer==0.19.2
 typing-inspection==0.4.2
 typing_extensions==4.15.0
@@ -136,4 +118,4 @@ uvicorn==0.37.0
 websockets==15.0.1
 xxhash==3.6.0
 yarl==1.22.0
-zipp==3.23.0
+zipp==3.23.0