From 271916d1964184df53ac2be7e9366f24f86eba52 Mon Sep 17 00:00:00 2001
From: Xiwen Yu <13230610+VALLIS-NERIA@users.noreply.github.com>
Date: Tue, 5 Aug 2025 13:18:33 +0800
Subject: [PATCH] fix deep_gemm & CUDA13

Signed-off-by: Xiwen Yu <13230610+VALLIS-NERIA@users.noreply.github.com>
---
 .gitmodules      | 2 +-
 bringup_fix.sh   | 9 ++++++++-
 requirements.txt | 2 +-
 3 files changed, 10 insertions(+), 3 deletions(-)

diff --git a/.gitmodules b/.gitmodules
index 067d732783..00ff73d136 100644
--- a/.gitmodules
+++ b/.gitmodules
@@ -25,4 +25,4 @@
 	url = https://github.com/wjakob/nanobind
 [submodule "3rdparty/cppzmq"]
 	path = 3rdparty/cppzmq
-	url = https://github.com/zeromq/cppzmq.github
+	url = https://github.com/zeromq/cppzmq.git
diff --git a/bringup_fix.sh b/bringup_fix.sh
index 9d52d15748..7c18044182 100644
--- a/bringup_fix.sh
+++ b/bringup_fix.sh
@@ -15,6 +15,7 @@ apt install -y libstdc++-14-dev
 
 elif [ $ARCH == "aarch64" ]; then
 
+# to be moved to docker/common/ scripts
 wget https://urm.nvidia.com/artifactory/sw-gpu-cuda-installer-generic-local/packaging/r13.0/cuda_nvrtc/linux-sbsa/13.0.48/cuda-nvrtc-dev-13-0_13.0.48-1_arm64.deb && \
     dpkg -i cuda-nvrtc-dev-13-0_13.0.48-1_arm64.deb && \
     rm cuda-nvrtc-dev-13-0_13.0.48-1_arm64.deb
@@ -23,15 +24,21 @@ wget https://github.com/Kitware/CMake/releases/download/v4.0.3/cmake-4.0.3-linux
     bash cmake-4.0.3-linux-aarch64.sh --skip-license --prefix=/usr/local/cmake --exclude-subdir
 
 apt update
-apt remove -y ibverbs-providers libibverbs1
+# fix NXIL
+apt remove -y ibverbs-providers libibverbs1 # package version conflict with libibverbs-dev
 apt install -y libibverbs-dev
+# fix LLVM build
 apt install -y libstdc++-14-dev
 
+# wait for https://github.com/NVIDIA/TensorRT-LLM/pull/6588
+pip install deep_gemm@git+https://github.com/VALLIS-NERIA/DeepGEMM.git@97d97a20c2ecd53a248ab64242219d780cf822b8 --no-build-isolation
+
 else
     echo "Unsupported architecture: $ARCH"
     exit 1
 fi
 
+# wait for new triton to be published
 cd /usr/local/lib/python3.12/dist-packages/ && \
     ls -la | grep pytorch_triton && \
     mv pytorch_triton-3.3.1+gitc8757738.dist-info triton-3.3.1+gitc8757738.dist-info && \
diff --git a/requirements.txt b/requirements.txt
index 9612f7d1f6..b02f8abd93 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -61,6 +61,6 @@ etcd3
 blake3
 llguidance==0.7.29
 soundfile
-deep_gemm @ git+https://github.com/zongfeijing/DeepGEMM.git@a9d538ef4dff0326fe521c6ca0bfde115703b56a
+deep_gemm@git+https://github.com/VALLIS-NERIA/DeepGEMM.git@97d97a20c2ecd53a248ab64242219d780cf822b8
 triton==3.3.1; platform_machine == "x86_64"
 blobfile