diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
index 960aea6..dcdcae5 100644
--- a/.github/workflows/build.yml
+++ b/.github/workflows/build.yml
@@ -52,8 +52,6 @@ jobs:
           - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.13.3', cuda: '12.8.1', rocm:    '', torch: '2.11.0', cudaarch: '8.0 8.6 8.9 9.0 10.0 12.0+PTX' }
 
         # Python 3.14
-          - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.14', cuda: '12.8.1', rocm:    '', torch: '2.7.0', cudaarch: '8.0 8.6 8.9 9.0 10.0 12.0+PTX' }
-          - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.14', cuda: '12.8.1', rocm:    '', torch: '2.8.0', cudaarch: '8.0 8.6 8.9 9.0 10.0 12.0+PTX' }
           - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.14', cuda: '12.8.1', rocm:    '', torch: '2.9.0', cudaarch: '8.0 8.6 8.9 9.0 10.0 12.0+PTX' }
           - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.14', cuda: '12.8.1', rocm:    '', torch: '2.10.0', cudaarch: '8.0 8.6 8.9 9.0 10.0 12.0+PTX' }
           - { artname: 'wheel', os: ubuntu-22.04, pyver: '3.14', cuda: '12.8.1', rocm:    '', torch: '2.11.0', cudaarch: '8.0 8.6 8.9 9.0 10.0 12.0+PTX' }
@@ -87,8 +85,6 @@ jobs:
           - { artname: 'wheel', os: windows-2022, pyver: '3.13.3', cuda: '12.8.1', rocm:    '', torch: '2.11.0', cudaarch: '8.0 8.6 8.9 9.0 10.0 12.0+PTX' }
 
         # Python 3.14
-          - { artname: 'wheel', os: windows-2022, pyver: '3.14', cuda: '12.8.1', rocm:    '', torch: '2.7.0', cudaarch: '8.0 8.6 8.9 9.0 10.0 12.0+PTX' }
-          - { artname: 'wheel', os: windows-2022, pyver: '3.14', cuda: '12.8.1', rocm:    '', torch: '2.8.0', cudaarch: '8.0 8.6 8.9 9.0 10.0 12.0+PTX' }
           - { artname: 'wheel', os: windows-2022, pyver: '3.14', cuda: '12.8.1', rocm:    '', torch: '2.9.0', cudaarch: '8.0 8.6 8.9 9.0 10.0 12.0+PTX' }
           - { artname: 'wheel', os: windows-2022, pyver: '3.14', cuda: '12.8.1', rocm:    '', torch: '2.10.0', cudaarch: '8.0 8.6 8.9 9.0 10.0 12.0+PTX' }
           - { artname: 'wheel', os: windows-2022, pyver: '3.14', cuda: '12.8.1', rocm:    '', torch: '2.11.0', cudaarch: '8.0 8.6 8.9 9.0 10.0 12.0+PTX' }
diff --git a/exllamav3/exllamav3_ext/softcap.cu b/exllamav3/exllamav3_ext/softcap.cu
index 2c6a2ef..c54c81b 100644
--- a/exllamav3/exllamav3_ext/softcap.cu
+++ b/exllamav3/exllamav3_ext/softcap.cu
@@ -16,7 +16,7 @@ __global__ void softcap_kernel
     const float scale
 )
 {
-    uint64_t idx = (uint64_t)blockIdx.x * NUM_THREADS + (uint64_t)threadIdx.x;
+    uint64_t idx = (uint64_t)blockIdx.x * blockDim.x + (uint64_t)threadIdx.x;
     if (idx >= numel) return;
 
     float v = x[idx];