workflow: Fixed nvcc fatal : Unsupported gpu architecture 'compute_101'

Author: JamePeng

.github/workflows/build-wheels-cu130-linux-basic.yml

@@ -61,7 +61,7 @@ jobs:
           echo $LD_LIBRARY_PATH
 
           # Add project-specific and feature flags
-          CMAKE_ARGS="-DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES='75-real;80-real;86-real;87-real;89-real;90-real;100-real;101-real;120-real'"
+          CMAKE_ARGS="-DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES='75-real;80-real;86-real;87-real;89-real;90-real;100-real;120-real'"
           CMAKE_ARGS="-DGGML_CUDA_FORCE_MMQ=on ${CMAKE_ARGS}"
           CMAKE_ARGS="${CMAKE_ARGS} -DLLAMA_CURL=off -DLLAMA_OPENSSL=on -DLLAMA_HTTPLIB=on"
 

.github/workflows/build-wheels-cu130-linux.yml

@@ -61,7 +61,7 @@ jobs:
           echo $LD_LIBRARY_PATH
 
           # Add project-specific and feature flags
-          CMAKE_ARGS="-DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES='75-real;80-real;86-real;87-real;89-real;90-real;100-real;101-real;120-real'"
+          CMAKE_ARGS="-DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES='75-real;80-real;86-real;87-real;89-real;90-real;100-real;120-real'"
           CMAKE_ARGS="-DGGML_CUDA_FORCE_MMQ=on ${CMAKE_ARGS}"
           CMAKE_ARGS="${CMAKE_ARGS} -DLLAMA_CURL=off -DLLAMA_OPENSSL=on -DLLAMA_HTTPLIB=on"
 

.github/workflows/build-wheels-cu130-win-basic.yml

@@ -16,7 +16,7 @@ jobs:
         pyver: ["3.10", "3.11", "3.12", "3.13"]
         cuda: ["13.0.2"]
         releasetag: ["Basic"]
-        cudaarch: ["75-real;80-real;86-real;87-real;89-real;90-real;100-real;101-real;120-real"]
+        cudaarch: ["75-real;80-real;86-real;87-real;89-real;90-real;100-real;120-real"]
     defaults:
       run:
         shell: pwsh

.github/workflows/build-wheels-cu130-win.yml

@@ -16,7 +16,7 @@ jobs:
         pyver: ["3.10", "3.11", "3.12", "3.13"]
         cuda: ["13.0.2"]
         releasetag: ["AVX2"]
-        cudaarch: ["75-real;80-real;86-real;87-real;89-real;90-real;100-real;101-real;120-real"]
+        cudaarch: ["75-real;80-real;86-real;87-real;89-real;90-real;100-real;120-real"]
     defaults:
       run:
         shell: pwsh