use experimental features; runtime fix

run_experimental_gpu.sh

@@ -3,6 +3,7 @@
 # This tells ROCm to treat gfx1151 as gfx1100
 export HSA_OVERRIDE_GFX_VERSION=11.0.0
 export AMD_SERIALIZE_KERNEL=3  # Enable debugging
+export TORCH_ROCM_AOTRITON_ENABLE_EXPERIMENTAL=1  # Enable Flash Efficient attention
 cd /mnt/shared/DEV/repos/d-popov.com/gogo2
 source venv/bin/activate
 python ANNOTATE/web/app.py "$@"