From 30e9e92173990c7dc58322170a0bd5fd5c048a1f Mon Sep 17 00:00:00 2001
From: wejoncy <wejoncy@163.com>
Date: Sat, 21 Mar 2026 02:22:21 +0000
Subject: [PATCH 01/18] fix: CI build - ubuntu-22.04, MSVC setup for Windows,
 update action versions

---
 .github/workflows/mannu_build.yml    |  2 +-
 .github/workflows/python-package.yml |  2 +-
 .github/workflows/release.yml        | 15 ++++++++++-----
 3 files changed, 12 insertions(+), 7 deletions(-)

diff --git a/.github/workflows/mannu_build.yml b/.github/workflows/mannu_build.yml
index 9400fed..6be3fcc 100644
--- a/.github/workflows/mannu_build.yml
+++ b/.github/workflows/mannu_build.yml
@@ -21,7 +21,7 @@ jobs:
       
       strategy:
         matrix:
-          os: [ubuntu-20.04, windows-latest]
+          os: [ubuntu-22.04, windows-latest]
           pyver: ["3.10"]
           cuda: ["12.6.0"]
       defaults:
diff --git a/.github/workflows/python-package.yml b/.github/workflows/python-package.yml
index 84d0ae6..492f125 100644
--- a/.github/workflows/python-package.yml
+++ b/.github/workflows/python-package.yml
@@ -43,7 +43,7 @@ jobs:
     - name: Install CUDA ${{ matrix.cuda-version }}
       if: github.event.pull_request.merged == true
       run: |
-        bash -x .github/workflows/scripts/cuda-install.sh ${{ matrix.cuda-version }} ubuntu-20.04
+        bash -x .github/workflows/scripts/cuda-install.sh ${{ matrix.cuda-version }} ubuntu-22.04
 
     - name: Install PyTorch 2.2.2 with CUDA ${{ matrix.cuda-version }}
       if: github.event.pull_request.merged == true
diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index 494d004..563ff6e 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -12,8 +12,9 @@ jobs:
     # needs: release
     
     strategy:
+      fail-fast: false
       matrix:
-        os: [ubuntu-20.04, windows-latest]
+        os: [ubuntu-22.04, windows-latest]
         pyver: ["3.10", "3.11", "3.12"]
         cuda: ["12.4.1"]
     defaults:
@@ -36,14 +37,18 @@ jobs:
           docker-images: true
           swap-storage: false
         
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v4
 
-      - uses: actions/setup-python@v3
+      - uses: actions/setup-python@v5
         with:
           python-version: ${{ matrix.pyver }}
 
+      - name: Setup MSVC
+        if: runner.os == 'Windows'
+        uses: ilammy/msvc-dev-cmd@v1
+
       - name: Setup Mamba
-        uses: conda-incubator/setup-miniconda@v2.2.0
+        uses: conda-incubator/setup-miniconda@v3
         with:
           activate-environment: "build"
           python-version: ${{ matrix.pyver }}
@@ -106,7 +111,7 @@ jobs:
       upload_url: ${{ steps.create_release.outputs.upload_url }}	
     steps:	
       - name: Checkout	
-        uses: actions/checkout@v3	
+        uses: actions/checkout@v4	
 
       - name: Extract branch info	
         shell: bash	

From 3c783c3f55a6345d1583724253c16b6c98827c34 Mon Sep 17 00:00:00 2001
From: wejoncy <wejoncy@163.com>
Date: Sat, 21 Mar 2026 02:24:39 +0000
Subject: [PATCH 02/18] ci: add workflow_dispatch for manual testing

---
 .github/workflows/release.yml | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index 563ff6e..53361db 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -4,6 +4,9 @@ on:
   push:
     tags:
       - "v*"
+    branches:
+      - "fix/ci-build"
+  workflow_dispatch:
 
 jobs:
   build_wheels:
@@ -107,6 +110,7 @@ jobs:
     name: Create Release	
     runs-on: ubuntu-latest	
     needs: build_wheels
+    if: startsWith(github.ref, 'refs/tags/')
     outputs:	
       upload_url: ${{ steps.create_release.outputs.upload_url }}	
     steps:	

From 958cec97043c53680ec48d43b1b2a363e770a4db Mon Sep 17 00:00:00 2001
From: wejoncy <wejoncy@163.com>
Date: Sat, 21 Mar 2026 02:27:53 +0000
Subject: [PATCH 03/18] ci: sync MSVC fix to mannu_build.yml

---
 .github/workflows/mannu_build.yml | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/.github/workflows/mannu_build.yml b/.github/workflows/mannu_build.yml
index 6be3fcc..3e7a357 100644
--- a/.github/workflows/mannu_build.yml
+++ b/.github/workflows/mannu_build.yml
@@ -44,14 +44,18 @@ jobs:
             docker-images: true
             swap-storage: false
           
-        - uses: actions/checkout@v3
+        - uses: actions/checkout@v4
 
-        - uses: actions/setup-python@v3
+        - uses: actions/setup-python@v5
           with:
             python-version: ${{ matrix.pyver }}
 
+        - name: Setup MSVC
+          if: runner.os == 'Windows'
+          uses: ilammy/msvc-dev-cmd@v1
+
         - name: Setup Mamba
-          uses: conda-incubator/setup-miniconda@v2.2.0
+          uses: conda-incubator/setup-miniconda@v3
           with:
             activate-environment: "build"
             python-version: ${{ matrix.pyver }}

From ff276d2a84fe5cb9f1b34b82e3a6ca7b14fb89a1 Mon Sep 17 00:00:00 2001
From: wejoncy <wejoncy@163.com>
Date: Sat, 21 Mar 2026 02:29:33 +0000
Subject: [PATCH 04/18] Revert "ci: sync MSVC fix to mannu_build.yml"

This reverts commit 958cec97043c53680ec48d43b1b2a363e770a4db.
---
 .github/workflows/mannu_build.yml | 10 +++-------
 1 file changed, 3 insertions(+), 7 deletions(-)

diff --git a/.github/workflows/mannu_build.yml b/.github/workflows/mannu_build.yml
index 3e7a357..6be3fcc 100644
--- a/.github/workflows/mannu_build.yml
+++ b/.github/workflows/mannu_build.yml
@@ -44,18 +44,14 @@ jobs:
             docker-images: true
             swap-storage: false
           
-        - uses: actions/checkout@v4
+        - uses: actions/checkout@v3
 
-        - uses: actions/setup-python@v5
+        - uses: actions/setup-python@v3
           with:
             python-version: ${{ matrix.pyver }}
 
-        - name: Setup MSVC
-          if: runner.os == 'Windows'
-          uses: ilammy/msvc-dev-cmd@v1
-
         - name: Setup Mamba
-          uses: conda-incubator/setup-miniconda@v3
+          uses: conda-incubator/setup-miniconda@v2.2.0
           with:
             activate-environment: "build"
             python-version: ${{ matrix.pyver }}

From bb8a0fa2bd599031bc8b56041b5e7433083de76a Mon Sep 17 00:00:00 2001
From: wejoncy <wejoncy@163.com>
Date: Sat, 21 Mar 2026 02:32:37 +0000
Subject: [PATCH 05/18] ci: upgrade CUDA to 12.9.0, PyTorch 2.7.0

---
 .github/workflows/release.yml | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index 53361db..bb5dd3c 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -19,7 +19,7 @@ jobs:
       matrix:
         os: [ubuntu-22.04, windows-latest]
         pyver: ["3.10", "3.11", "3.12"]
-        cuda: ["12.4.1"]
+        cuda: ["12.9.0"]
     defaults:
       run:
         shell: pwsh
@@ -72,7 +72,7 @@ jobs:
           # Install torch
           $cudaVersion = $env:CUDA_VERSION.Replace('.', '')
           $cudaVersionPytorch = $cudaVersion.Substring(0, $cudaVersion.Length - 1)
-          if ([int]$cudaVersionPytorch -gt 121) { $pytorchVersion = "torch==2.5.1" } else {$pytorchVersion = "torch==2.4.1"}
+          if ([int]$cudaVersionPytorch -gt 126) { $pytorchVersion = "torch==2.7.0" } elseif ([int]$cudaVersionPytorch -gt 121) { $pytorchVersion = "torch==2.5.1" } else {$pytorchVersion = "torch==2.4.1"}
           echo "pytorchVersion=$pytorchVersion"
           echo "cudaVersion=<$cudaVersion>"
           echo "cudaVersionPytorch=$cudaVersionPytorch"

From f55825ef36c384919a49c6942bb4b0cfed4662a8 Mon Sep 17 00:00:00 2001
From: wejoncy <wejoncy@163.com>
Date: Sat, 21 Mar 2026 02:35:35 +0000
Subject: [PATCH 06/18] ci: pin mamba cuda version to prevent mismatch

---
 .github/workflows/release.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index bb5dd3c..efb308e 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -64,7 +64,7 @@ jobs:
       - name: Install Dependencies
         run: |
           # Install CUDA toolkit
-          mamba install -y 'cuda' -c "nvidia/label/cuda-${env:CUDA_VERSION}"
+          mamba install -y "cuda=${env:CUDA_VERSION}" -c "nvidia/label/cuda-${env:CUDA_VERSION}"
           # Env variables
           $env:CUDA_PATH = $env:CONDA_PREFIX
           $env:CUDA_HOME = $env:CONDA_PREFIX

From a915d5604eb3875a24bc13fbdf8ea4095483d6eb Mon Sep 17 00:00:00 2001
From: wejoncy <wejoncy@163.com>
Date: Sat, 21 Mar 2026 02:37:46 +0000
Subject: [PATCH 07/18] ci: CUDA 13.0.0 + PyTorch 2.9.0

---
 .github/workflows/release.yml | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index efb308e..d4dc1a9 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -19,7 +19,7 @@ jobs:
       matrix:
         os: [ubuntu-22.04, windows-latest]
         pyver: ["3.10", "3.11", "3.12"]
-        cuda: ["12.9.0"]
+        cuda: ["13.0.0"]
     defaults:
       run:
         shell: pwsh
@@ -72,7 +72,7 @@ jobs:
           # Install torch
           $cudaVersion = $env:CUDA_VERSION.Replace('.', '')
           $cudaVersionPytorch = $cudaVersion.Substring(0, $cudaVersion.Length - 1)
-          if ([int]$cudaVersionPytorch -gt 126) { $pytorchVersion = "torch==2.7.0" } elseif ([int]$cudaVersionPytorch -gt 121) { $pytorchVersion = "torch==2.5.1" } else {$pytorchVersion = "torch==2.4.1"}
+          if ([int]$cudaVersionPytorch -gt 129) { $pytorchVersion = "torch==2.9.0" } elseif ([int]$cudaVersionPytorch -gt 126) { $pytorchVersion = "torch==2.7.0" } elseif ([int]$cudaVersionPytorch -gt 121) { $pytorchVersion = "torch==2.5.1" } else {$pytorchVersion = "torch==2.4.1"}
           echo "pytorchVersion=$pytorchVersion"
           echo "cudaVersion=<$cudaVersion>"
           echo "cudaVersionPytorch=$cudaVersionPytorch"

From 2572422ef23e7acb0d38146467a2b35803059905 Mon Sep 17 00:00:00 2001
From: wejoncy <wejoncy@163.com>
Date: Sat, 21 Mar 2026 02:47:04 +0000
Subject: [PATCH 08/18] ci: fix cuda headers (cuda-cudart-dev) and Windows
 DISTUTILS_USE_SDK

---
 .github/workflows/release.yml | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index d4dc1a9..a5ebf15 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -64,7 +64,7 @@ jobs:
       - name: Install Dependencies
         run: |
           # Install CUDA toolkit
-          mamba install -y "cuda=${env:CUDA_VERSION}" -c "nvidia/label/cuda-${env:CUDA_VERSION}"
+          mamba install -y "cuda=${env:CUDA_VERSION}" "cuda-cudart-dev" "cuda-nvcc" -c "nvidia/label/cuda-${env:CUDA_VERSION}"
           # Env variables
           $env:CUDA_PATH = $env:CONDA_PREFIX
           $env:CUDA_HOME = $env:CONDA_PREFIX
@@ -94,6 +94,7 @@ jobs:
           }
           # echo "{CUDA_VERSION}=$env:CUDA_VERSION" >> $GITHUB_ENV
           $env:GENERAL_TORCH = 1 # OptionalCUDAGuard
+          $env:DISTUTILS_USE_SDK = 1
           python setup.py sdist bdist_wheel -k $env:PLAT_ARG.split()
           ls dist/*.whl
         env:

From 919a29049ed897bcef0011795620ee555fa11b99 Mon Sep 17 00:00:00 2001
From: wejoncy <wejoncy@163.com>
Date: Sat, 21 Mar 2026 02:54:28 +0000
Subject: [PATCH 09/18] ci: debug and fix CUDA paths for conda CUDA 13.0

---
 .github/workflows/release.yml | 28 ++++++++++++++++++++++++++++
 1 file changed, 28 insertions(+)

diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index a5ebf15..bad1a77 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -88,6 +88,34 @@ jobs:
         run: |
           $env:CUDA_PATH = $env:CONDA_PREFIX
           $env:CUDA_HOME = $env:CONDA_PREFIX
+          # Debug: find cuda_runtime_api.h and nvcc
+          echo "CONDA_PREFIX=$env:CONDA_PREFIX"
+          if ($IsLinux) {
+            bash -c "find $env:CONDA_PREFIX -name 'cuda_runtime_api.h' 2>/dev/null"
+            bash -c "find $env:CONDA_PREFIX -name 'nvcc' -type f 2>/dev/null"
+            bash -c "ls -la $env:CONDA_PREFIX/include/cuda_runtime_api.h 2>/dev/null || echo 'not in include/'"
+            bash -c "ls -la $env:CONDA_PREFIX/targets/x86_64-linux/include/cuda_runtime_api.h 2>/dev/null || echo 'not in targets/'"
+          } else {
+            Get-ChildItem -Path $env:CONDA_PREFIX -Recurse -Filter "cuda_runtime_api.h" -ErrorAction SilentlyContinue | Select-Object FullName
+            Get-ChildItem -Path $env:CONDA_PREFIX -Recurse -Filter "nvcc.exe" -ErrorAction SilentlyContinue | Select-Object FullName
+          }
+          # Set include path for Linux if needed
+          if ($IsLinux -and (Test-Path "$env:CONDA_PREFIX/targets/x86_64-linux/include/cuda_runtime_api.h")) {
+            $env:CPLUS_INCLUDE_PATH = "$env:CONDA_PREFIX/targets/x86_64-linux/include"
+            $env:C_INCLUDE_PATH = "$env:CONDA_PREFIX/targets/x86_64-linux/include"
+            $env:CUDA_HOME = "$env:CONDA_PREFIX/targets/x86_64-linux"
+            $env:CUDA_PATH = "$env:CONDA_PREFIX/targets/x86_64-linux"
+          }
+          # Set nvcc path for Windows if needed
+          if (!$IsLinux) {
+            $nvcc = Get-ChildItem -Path $env:CONDA_PREFIX -Recurse -Filter "nvcc.exe" -ErrorAction SilentlyContinue | Select-Object -First 1
+            if ($nvcc) {
+              $cudaDir = Split-Path (Split-Path $nvcc.FullName)
+              echo "Setting CUDA_HOME to $cudaDir"
+              $env:CUDA_HOME = $cudaDir
+              $env:CUDA_PATH = $cudaDir
+            }
+          }
           # Only add +cu118 to wheel if not releasing on PyPi
           if ( $env:CUDA_VERSION -eq $env:PYPI_CUDA_VERSION ){
             $env:PYPI_BUILD = 1

From d78842d061d3ea7967ad1ed32a9873fb35f0999e Mon Sep 17 00:00:00 2001
From: wejoncy <wejoncy@163.com>
Date: Sat, 21 Mar 2026 03:05:47 +0000
Subject: [PATCH 10/18] ci: add cuda-cccl for thrust/cub headers (CUDA 13)

---
 .github/workflows/release.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index bad1a77..8d6192e 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -64,7 +64,7 @@ jobs:
       - name: Install Dependencies
         run: |
           # Install CUDA toolkit
-          mamba install -y "cuda=${env:CUDA_VERSION}" "cuda-cudart-dev" "cuda-nvcc" -c "nvidia/label/cuda-${env:CUDA_VERSION}"
+          mamba install -y "cuda=${env:CUDA_VERSION}" "cuda-cudart-dev" "cuda-nvcc" "cuda-cccl" -c "nvidia/label/cuda-${env:CUDA_VERSION}"
           # Env variables
           $env:CUDA_PATH = $env:CONDA_PREFIX
           $env:CUDA_HOME = $env:CONDA_PREFIX

From f5a3ec784e9725ba14de6bea8c67c81119c2591a Mon Sep 17 00:00:00 2001
From: wejoncy <wejoncy@163.com>
Date: Sat, 21 Mar 2026 03:26:56 +0000
Subject: [PATCH 11/18] ci: symlink CUDA headers for g++, upgrade to PyTorch
 2.10.0

---
 .github/workflows/release.yml | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index 8d6192e..5fed4d1 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -72,7 +72,7 @@ jobs:
           # Install torch
           $cudaVersion = $env:CUDA_VERSION.Replace('.', '')
           $cudaVersionPytorch = $cudaVersion.Substring(0, $cudaVersion.Length - 1)
-          if ([int]$cudaVersionPytorch -gt 129) { $pytorchVersion = "torch==2.9.0" } elseif ([int]$cudaVersionPytorch -gt 126) { $pytorchVersion = "torch==2.7.0" } elseif ([int]$cudaVersionPytorch -gt 121) { $pytorchVersion = "torch==2.5.1" } else {$pytorchVersion = "torch==2.4.1"}
+          if ([int]$cudaVersionPytorch -gt 129) { $pytorchVersion = "torch==2.10.0" } elseif ([int]$cudaVersionPytorch -gt 126) { $pytorchVersion = "torch==2.7.0" } elseif ([int]$cudaVersionPytorch -gt 121) { $pytorchVersion = "torch==2.5.1" } else {$pytorchVersion = "torch==2.4.1"}
           echo "pytorchVersion=$pytorchVersion"
           echo "cudaVersion=<$cudaVersion>"
           echo "cudaVersionPytorch=$cudaVersionPytorch"
@@ -101,8 +101,9 @@ jobs:
           }
           # Set include path for Linux if needed
           if ($IsLinux -and (Test-Path "$env:CONDA_PREFIX/targets/x86_64-linux/include/cuda_runtime_api.h")) {
-            $env:CPLUS_INCLUDE_PATH = "$env:CONDA_PREFIX/targets/x86_64-linux/include"
-            $env:C_INCLUDE_PATH = "$env:CONDA_PREFIX/targets/x86_64-linux/include"
+            # Symlink CUDA headers to CONDA_PREFIX/include so g++ can find thrust/cub
+            bash -c "for d in thrust cub nv cuda; do [ -e $env:CONDA_PREFIX/targets/x86_64-linux/include/\$d ] && ln -sfn $env:CONDA_PREFIX/targets/x86_64-linux/include/\$d $env:CONDA_PREFIX/include/\$d; done"
+            bash -c "for f in $env:CONDA_PREFIX/targets/x86_64-linux/include/cuda*.h; do [ -f \$f ] && ln -sfn \$f $env:CONDA_PREFIX/include/; done"
             $env:CUDA_HOME = "$env:CONDA_PREFIX/targets/x86_64-linux"
             $env:CUDA_PATH = "$env:CONDA_PREFIX/targets/x86_64-linux"
           }

From 903ffc7e2485be91a75e136bbb1d3e2378825f85 Mon Sep 17 00:00:00 2001
From: wejoncy <wejoncy@163.com>
Date: Sat, 21 Mar 2026 03:36:06 +0000
Subject: [PATCH 12/18] ci: use CXXFLAGS -I for thrust include path, simplify
 build step

---
 .github/workflows/release.yml | 40 +++++++++++++++++------------------
 1 file changed, 20 insertions(+), 20 deletions(-)

diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index 5fed4d1..79bd55c 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -88,27 +88,28 @@ jobs:
         run: |
           $env:CUDA_PATH = $env:CONDA_PREFIX
           $env:CUDA_HOME = $env:CONDA_PREFIX
-          # Debug: find cuda_runtime_api.h and nvcc
           echo "CONDA_PREFIX=$env:CONDA_PREFIX"
+          
           if ($IsLinux) {
-            bash -c "find $env:CONDA_PREFIX -name 'cuda_runtime_api.h' 2>/dev/null"
-            bash -c "find $env:CONDA_PREFIX -name 'nvcc' -type f 2>/dev/null"
-            bash -c "ls -la $env:CONDA_PREFIX/include/cuda_runtime_api.h 2>/dev/null || echo 'not in include/'"
-            bash -c "ls -la $env:CONDA_PREFIX/targets/x86_64-linux/include/cuda_runtime_api.h 2>/dev/null || echo 'not in targets/'"
+            # Find cuda include dir containing thrust/complex.h
+            $thrustDir = bash -c "dirname `find $env:CONDA_PREFIX -path '*/thrust/complex.h' 2>/dev/null | head -1` 2>/dev/null"
+            if ($thrustDir -and $thrustDir -ne ".") {
+              $cudaInclude = Split-Path $thrustDir
+              echo "Found thrust in: $cudaInclude"
+              # Set CUDA_HOME to parent of include (for nvcc)
+              $cudaBase = Split-Path $cudaInclude
+              if (Test-Path "$cudaBase/bin/nvcc") {
+                $env:CUDA_HOME = $cudaBase
+                $env:CUDA_PATH = $cudaBase
+                echo "CUDA_HOME=$cudaBase"
+              }
+              # Add to compiler include paths
+              $env:CXXFLAGS = "-I$cudaInclude"
+              $env:CFLAGS = "-I$cudaInclude"
+              echo "CXXFLAGS=$env:CXXFLAGS"
+            }
           } else {
-            Get-ChildItem -Path $env:CONDA_PREFIX -Recurse -Filter "cuda_runtime_api.h" -ErrorAction SilentlyContinue | Select-Object FullName
-            Get-ChildItem -Path $env:CONDA_PREFIX -Recurse -Filter "nvcc.exe" -ErrorAction SilentlyContinue | Select-Object FullName
-          }
-          # Set include path for Linux if needed
-          if ($IsLinux -and (Test-Path "$env:CONDA_PREFIX/targets/x86_64-linux/include/cuda_runtime_api.h")) {
-            # Symlink CUDA headers to CONDA_PREFIX/include so g++ can find thrust/cub
-            bash -c "for d in thrust cub nv cuda; do [ -e $env:CONDA_PREFIX/targets/x86_64-linux/include/\$d ] && ln -sfn $env:CONDA_PREFIX/targets/x86_64-linux/include/\$d $env:CONDA_PREFIX/include/\$d; done"
-            bash -c "for f in $env:CONDA_PREFIX/targets/x86_64-linux/include/cuda*.h; do [ -f \$f ] && ln -sfn \$f $env:CONDA_PREFIX/include/; done"
-            $env:CUDA_HOME = "$env:CONDA_PREFIX/targets/x86_64-linux"
-            $env:CUDA_PATH = "$env:CONDA_PREFIX/targets/x86_64-linux"
-          }
-          # Set nvcc path for Windows if needed
-          if (!$IsLinux) {
+            # Windows: find nvcc.exe and set CUDA_HOME
             $nvcc = Get-ChildItem -Path $env:CONDA_PREFIX -Recurse -Filter "nvcc.exe" -ErrorAction SilentlyContinue | Select-Object -First 1
             if ($nvcc) {
               $cudaDir = Split-Path (Split-Path $nvcc.FullName)
@@ -117,11 +118,10 @@ jobs:
               $env:CUDA_PATH = $cudaDir
             }
           }
-          # Only add +cu118 to wheel if not releasing on PyPi
+          
           if ( $env:CUDA_VERSION -eq $env:PYPI_CUDA_VERSION ){
             $env:PYPI_BUILD = 1
           }
-          # echo "{CUDA_VERSION}=$env:CUDA_VERSION" >> $GITHUB_ENV
           $env:GENERAL_TORCH = 1 # OptionalCUDAGuard
           $env:DISTUTILS_USE_SDK = 1
           python setup.py sdist bdist_wheel -k $env:PLAT_ARG.split()

From 5e8960c4543fa3fed1d6fafcfcca457616e42595 Mon Sep 17 00:00:00 2001
From: wejoncy <wejoncy@163.com>
Date: Sat, 21 Mar 2026 03:45:04 +0000
Subject: [PATCH 13/18] ci: split build into Linux(bash) and Windows(pwsh) to
 fix shell escaping

---
 .github/workflows/release.yml | 72 +++++++++++++++++++----------------
 1 file changed, 39 insertions(+), 33 deletions(-)

diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index 79bd55c..57d5017 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -84,50 +84,56 @@ jobs:
           python -c "import torch; print('CUDA:', torch.version.cuda)"
           python -c "import os; print('CUDA_HOME:', os.getenv('CUDA_HOME', None))"
           python -c "from torch.utils import cpp_extension; print (cpp_extension.CUDA_HOME)"
-      - name: Build Wheel
+      - name: Build Wheel (Linux)
+        if: runner.os == 'Linux'
+        shell: bash
+        run: |
+          export CUDA_PATH=$CONDA_PREFIX
+          export CUDA_HOME=$CONDA_PREFIX
+          
+          # Find thrust include dir
+          THRUST_H=$(find $CONDA_PREFIX -path '*/thrust/complex.h' 2>/dev/null | head -1)
+          if [ -n "$THRUST_H" ]; then
+            CUDA_INCLUDE=$(dirname $(dirname "$THRUST_H"))
+            echo "Found thrust in: $CUDA_INCLUDE"
+            export CXXFLAGS="-I$CUDA_INCLUDE"
+            export CFLAGS="-I$CUDA_INCLUDE"
+            # Check if this dir has bin/nvcc
+            CUDA_BASE=$(dirname "$CUDA_INCLUDE")
+            if [ -f "$CUDA_BASE/bin/nvcc" ]; then
+              export CUDA_HOME=$CUDA_BASE
+              export CUDA_PATH=$CUDA_BASE
+            fi
+          fi
+          echo "CUDA_HOME=$CUDA_HOME"
+          echo "CXXFLAGS=$CXXFLAGS"
+          
+          export GENERAL_TORCH=1
+          python setup.py sdist bdist_wheel -k --plat-name manylinux2014_x86_64
+          ls dist/*.whl
+
+      - name: Build Wheel (Windows)
+        if: runner.os == 'Windows'
         run: |
           $env:CUDA_PATH = $env:CONDA_PREFIX
           $env:CUDA_HOME = $env:CONDA_PREFIX
-          echo "CONDA_PREFIX=$env:CONDA_PREFIX"
           
-          if ($IsLinux) {
-            # Find cuda include dir containing thrust/complex.h
-            $thrustDir = bash -c "dirname `find $env:CONDA_PREFIX -path '*/thrust/complex.h' 2>/dev/null | head -1` 2>/dev/null"
-            if ($thrustDir -and $thrustDir -ne ".") {
-              $cudaInclude = Split-Path $thrustDir
-              echo "Found thrust in: $cudaInclude"
-              # Set CUDA_HOME to parent of include (for nvcc)
-              $cudaBase = Split-Path $cudaInclude
-              if (Test-Path "$cudaBase/bin/nvcc") {
-                $env:CUDA_HOME = $cudaBase
-                $env:CUDA_PATH = $cudaBase
-                echo "CUDA_HOME=$cudaBase"
-              }
-              # Add to compiler include paths
-              $env:CXXFLAGS = "-I$cudaInclude"
-              $env:CFLAGS = "-I$cudaInclude"
-              echo "CXXFLAGS=$env:CXXFLAGS"
-            }
-          } else {
-            # Windows: find nvcc.exe and set CUDA_HOME
-            $nvcc = Get-ChildItem -Path $env:CONDA_PREFIX -Recurse -Filter "nvcc.exe" -ErrorAction SilentlyContinue | Select-Object -First 1
-            if ($nvcc) {
-              $cudaDir = Split-Path (Split-Path $nvcc.FullName)
-              echo "Setting CUDA_HOME to $cudaDir"
-              $env:CUDA_HOME = $cudaDir
-              $env:CUDA_PATH = $cudaDir
-            }
+          # Find nvcc.exe and set CUDA_HOME
+          $nvcc = Get-ChildItem -Path $env:CONDA_PREFIX -Recurse -Filter "nvcc.exe" -ErrorAction SilentlyContinue | Select-Object -First 1
+          if ($nvcc) {
+            $cudaDir = Split-Path (Split-Path $nvcc.FullName)
+            echo "Setting CUDA_HOME to $cudaDir"
+            $env:CUDA_HOME = $cudaDir
+            $env:CUDA_PATH = $cudaDir
           }
           
           if ( $env:CUDA_VERSION -eq $env:PYPI_CUDA_VERSION ){
             $env:PYPI_BUILD = 1
           }
-          $env:GENERAL_TORCH = 1 # OptionalCUDAGuard
+          $env:GENERAL_TORCH = 1
           $env:DISTUTILS_USE_SDK = 1
-          python setup.py sdist bdist_wheel -k $env:PLAT_ARG.split()
+          python setup.py sdist bdist_wheel -k --plat-name win_amd64
           ls dist/*.whl
-        env:
-          PLAT_ARG: ${{ contains(runner.os, 'Linux') && '--plat-name manylinux2014_x86_64' || '--plat-name win_amd64' }}
 
       - uses: actions/upload-artifact@v4
         with:

From c5206ffb0831281e6e7860049faa31d15fceb766 Mon Sep 17 00:00:00 2001
From: wejoncy <wejoncy@163.com>
Date: Sat, 21 Mar 2026 06:35:34 +0000
Subject: [PATCH 14/18] ci: fix Linux conda activation (bash -el), Windows use
 CUDA 12.6 (MSVC compat)

---
 .github/workflows/release.yml | 20 +++++++++++++++++---
 1 file changed, 17 insertions(+), 3 deletions(-)

diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index 57d5017..2b34c95 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -20,6 +20,19 @@ jobs:
         os: [ubuntu-22.04, windows-latest]
         pyver: ["3.10", "3.11", "3.12"]
         cuda: ["13.0.0"]
+        exclude:
+          - os: windows-latest
+            cuda: "13.0.0"
+        include:
+          - os: windows-latest
+            pyver: "3.10"
+            cuda: "12.6.3"
+          - os: windows-latest
+            pyver: "3.11"
+            cuda: "12.6.3"
+          - os: windows-latest
+            pyver: "3.12"
+            cuda: "12.6.3"
     defaults:
       run:
         shell: pwsh
@@ -72,7 +85,7 @@ jobs:
           # Install torch
           $cudaVersion = $env:CUDA_VERSION.Replace('.', '')
           $cudaVersionPytorch = $cudaVersion.Substring(0, $cudaVersion.Length - 1)
-          if ([int]$cudaVersionPytorch -gt 129) { $pytorchVersion = "torch==2.10.0" } elseif ([int]$cudaVersionPytorch -gt 126) { $pytorchVersion = "torch==2.7.0" } elseif ([int]$cudaVersionPytorch -gt 121) { $pytorchVersion = "torch==2.5.1" } else {$pytorchVersion = "torch==2.4.1"}
+          if ([int]$cudaVersionPytorch -gt 129) { $pytorchVersion = "torch==2.10.0" } elseif ([int]$cudaVersionPytorch -gt 126) { $pytorchVersion = "torch==2.7.0" } elseif ([int]$cudaVersionPytorch -gt 124) { $pytorchVersion = "torch==2.10.0" } elseif ([int]$cudaVersionPytorch -gt 121) { $pytorchVersion = "torch==2.5.1" } else {$pytorchVersion = "torch==2.4.1"}
           echo "pytorchVersion=$pytorchVersion"
           echo "cudaVersion=<$cudaVersion>"
           echo "cudaVersionPytorch=$cudaVersionPytorch"
@@ -86,11 +99,13 @@ jobs:
           python -c "from torch.utils import cpp_extension; print (cpp_extension.CUDA_HOME)"
       - name: Build Wheel (Linux)
         if: runner.os == 'Linux'
-        shell: bash
+        shell: bash -el {0}
         run: |
           export CUDA_PATH=$CONDA_PREFIX
           export CUDA_HOME=$CONDA_PREFIX
           
+          echo "CONDA_PREFIX=$CONDA_PREFIX"
+          
           # Find thrust include dir
           THRUST_H=$(find $CONDA_PREFIX -path '*/thrust/complex.h' 2>/dev/null | head -1)
           if [ -n "$THRUST_H" ]; then
@@ -98,7 +113,6 @@ jobs:
             echo "Found thrust in: $CUDA_INCLUDE"
             export CXXFLAGS="-I$CUDA_INCLUDE"
             export CFLAGS="-I$CUDA_INCLUDE"
-            # Check if this dir has bin/nvcc
             CUDA_BASE=$(dirname "$CUDA_INCLUDE")
             if [ -f "$CUDA_BASE/bin/nvcc" ]; then
               export CUDA_HOME=$CUDA_BASE

From 7f197672d0f1b5b1290c9ee004a72ff51315158d Mon Sep 17 00:00:00 2001
From: wejoncy <wejoncy@163.com>
Date: Sat, 21 Mar 2026 06:42:47 +0000
Subject: [PATCH 15/18] fix: skip compute_70 for CUDA >= 13.0 (no longer
 supported)

---
 setup.py | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

diff --git a/setup.py b/setup.py
index bbc4f55..e1c9e8a 100644
--- a/setup.py
+++ b/setup.py
@@ -8,7 +8,7 @@
 
 from packaging.version import parse, Version
 import setuptools
-from torch.utils.cpp_extension import BuildExtension, CUDA_HOME, CUDAExtension
+from torch.utils.cpp_extension import BuildExtension,  CUDA_HOME, CUDAExtension
 
 ROOT_DIR = os.path.dirname(__file__)
 
@@ -72,7 +72,8 @@ def get_nvcc_cuda_version(cuda_dir: str = "") -> Version:
 def get_compute_capabilities(compute_capabilities: Set[int], lower: int = 70):
     # Collect the compute capabilities of all available GPUs.
     if len(compute_capabilities) == 0 and (is_pypi_build() or not torch.cuda.is_available()):
-        if lower <= 70:
+        nvcc_cuda_version = get_nvcc_cuda_version()
+        if lower <= 70 and nvcc_cuda_version < Version("13.0"):
             compute_capabilities.add(70)
         if lower <= 75:
             compute_capabilities.add(75)
@@ -88,10 +89,11 @@ def get_compute_capabilities(compute_capabilities: Set[int], lower: int = 70):
             compute_capabilities.add(major * 10 + minor)
 
     if len(compute_capabilities) == 0:
-        compute_capabilities.add(70)
+        nvcc_cuda_version = get_nvcc_cuda_version()
+        if nvcc_cuda_version < Version("13.0"):
+            compute_capabilities.add(70)
         compute_capabilities.add(75)
         compute_capabilities.add(80)
-        nvcc_cuda_version = get_nvcc_cuda_version()
         if nvcc_cuda_version > Version("11.1"):
             compute_capabilities.add(86)
         if nvcc_cuda_version > Version("11.8"):

From 45a909bbc74efc81ada77d644c9e1b8d87d7cb81 Mon Sep 17 00:00:00 2001
From: wejoncy <wejoncy@163.com>
Date: Sat, 21 Mar 2026 06:50:36 +0000
Subject: [PATCH 16/18] ci: patch torch compiled_autograd.h for MSVC+CUDA13
 std:: bug, unify CUDA 13.0

---
 .github/workflows/release.yml | 23 +++++++++--------------
 1 file changed, 9 insertions(+), 14 deletions(-)

diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index 2b34c95..af4dbba 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -20,19 +20,6 @@ jobs:
         os: [ubuntu-22.04, windows-latest]
         pyver: ["3.10", "3.11", "3.12"]
         cuda: ["13.0.0"]
-        exclude:
-          - os: windows-latest
-            cuda: "13.0.0"
-        include:
-          - os: windows-latest
-            pyver: "3.10"
-            cuda: "12.6.3"
-          - os: windows-latest
-            pyver: "3.11"
-            cuda: "12.6.3"
-          - os: windows-latest
-            pyver: "3.12"
-            cuda: "12.6.3"
     defaults:
       run:
         shell: pwsh
@@ -85,7 +72,7 @@ jobs:
           # Install torch
           $cudaVersion = $env:CUDA_VERSION.Replace('.', '')
           $cudaVersionPytorch = $cudaVersion.Substring(0, $cudaVersion.Length - 1)
-          if ([int]$cudaVersionPytorch -gt 129) { $pytorchVersion = "torch==2.10.0" } elseif ([int]$cudaVersionPytorch -gt 126) { $pytorchVersion = "torch==2.7.0" } elseif ([int]$cudaVersionPytorch -gt 124) { $pytorchVersion = "torch==2.10.0" } elseif ([int]$cudaVersionPytorch -gt 121) { $pytorchVersion = "torch==2.5.1" } else {$pytorchVersion = "torch==2.4.1"}
+          if ([int]$cudaVersionPytorch -gt 129) { $pytorchVersion = "torch==2.10.0" } elseif ([int]$cudaVersionPytorch -gt 121) { $pytorchVersion = "torch==2.5.1" } else {$pytorchVersion = "torch==2.4.1"}
           echo "pytorchVersion=$pytorchVersion"
           echo "cudaVersion=<$cudaVersion>"
           echo "cudaVersionPytorch=$cudaVersionPytorch"
@@ -141,6 +128,14 @@ jobs:
             $env:CUDA_PATH = $cudaDir
           }
           
+          # Patch PyTorch compiled_autograd.h for MSVC + CUDA 13 std:: ambiguity
+          # See https://github.com/pytorch/pytorch/pull/144707#issuecomment-2692282551
+          $header = python -c "import torch; import os; print(os.path.join(os.path.dirname(torch.__file__), 'include', 'torch', 'csrc', 'dynamo', 'compiled_autograd.h'))"
+          if (Test-Path $header) {
+            (Get-Content $header) -replace '\} else if constexpr \(::std::is_same_v<T, ::std::string>\) \{', '// } else if constexpr (::std::is_same_v<T, ::std::string>) {' -replace 'return at::StringType::get\(\);', '// return at::StringType::get();' | Set-Content $header
+            echo "Patched compiled_autograd.h"
+          }
+          
           if ( $env:CUDA_VERSION -eq $env:PYPI_CUDA_VERSION ){
             $env:PYPI_BUILD = 1
           }

From 2f7237ddcab2bea027214514cbcd8738b790be34 Mon Sep 17 00:00:00 2001
From: wejoncy <wejoncy@163.com>
Date: Sat, 21 Mar 2026 07:07:27 +0000
Subject: [PATCH 17/18] feat: Python 3.11-3.13, add sm_90/sm_100/sm_120
 (H100/B200/RTX5090), update torch to 2.10.0

---
 .github/workflows/release.yml |  2 +-
 pyproject.toml                |  2 +-
 setup.py                      | 12 ++++++++----
 3 files changed, 10 insertions(+), 6 deletions(-)

diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index af4dbba..268d76f 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -18,7 +18,7 @@ jobs:
       fail-fast: false
       matrix:
         os: [ubuntu-22.04, windows-latest]
-        pyver: ["3.10", "3.11", "3.12"]
+        pyver: ["3.11", "3.12", "3.13"]
         cuda: ["13.0.0"]
     defaults:
       run:
diff --git a/pyproject.toml b/pyproject.toml
index 8a3d653..0b671c6 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -1,5 +1,5 @@
 [build-system]
-requires = ["setuptools==69.5.1", "wheel", "packaging", "ninja>=1.11.1", "torch==2.2.2"]
+requires = ["setuptools==69.5.1", "wheel", "packaging", "ninja>=1.11.1", "torch==2.10.0"]
 
 build-backend = "setuptools.build_meta"
 
diff --git a/setup.py b/setup.py
index e1c9e8a..9c0513e 100644
--- a/setup.py
+++ b/setup.py
@@ -80,8 +80,10 @@ def get_compute_capabilities(compute_capabilities: Set[int], lower: int = 70):
         compute_capabilities.add(80)
         compute_capabilities.add(86)
         compute_capabilities.add(89)
-
-    if len(compute_capabilities) == 0:
+        compute_capabilities.add(90)
+        if nvcc_cuda_version >= Version("12.8"):
+            compute_capabilities.add(100)
+            compute_capabilities.add(120)
         for i in range(torch.cuda.device_count()):
             major, minor = torch.cuda.get_device_capability(i)
             if major*10+minor < lower:
@@ -99,6 +101,9 @@ def get_compute_capabilities(compute_capabilities: Set[int], lower: int = 70):
         if nvcc_cuda_version > Version("11.8"):
             compute_capabilities.add(89)
             compute_capabilities.add(90)
+        if nvcc_cuda_version >= Version("12.8"):
+            compute_capabilities.add(100)
+            compute_capabilities.add(120)
 
     print(f"build pacakge for archs: {compute_capabilities}")
     capability_flags = []
@@ -211,7 +216,6 @@ def get_gpu_ver():
         "Documentation": "https://github.com/wejoncy/QLLM",
     },
     classifiers=[
-        "Programming Language :: Python :: 3.10",
         "Programming Language :: Python :: 3.11",
         "Programming Language :: Python :: 3.12",
         "Programming Language :: Python :: 3.13",
@@ -219,7 +223,7 @@ def get_gpu_ver():
         "Topic :: Scientific/Engineering :: Artificial Intelligence",
     ],
     packages=setuptools.find_packages(exclude=("")),
-    python_requires=">=3.10",
+    python_requires=">=3.11",
     install_requires=get_requirements(),
     ext_modules=build_cuda_extensions(),
     cmdclass={'build_ext': BuildExtension},

From f86780d4d541a545d6f7d5f69a9d98893600cba5 Mon Sep 17 00:00:00 2001
From: wejoncy <wejoncy@163.com>
Date: Sat, 21 Mar 2026 07:23:24 +0000
Subject: [PATCH 18/18] cleanup: remove test branch trigger, update Python/GPU
 arch versions

---
 .github/workflows/release.yml | 2 --
 1 file changed, 2 deletions(-)

diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index 268d76f..9e2a216 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -4,8 +4,6 @@ on:
   push:
     tags:
       - "v*"
-    branches:
-      - "fix/ci-build"
   workflow_dispatch:
 
 jobs: