Deepseek V4 (#23882)

Co-authored-by: Baizhou Zhang <sobereddiezhang@gmail.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-authored-by: fzyzcjy <ch271828n@outlook.com>
Co-authored-by: ispobock <ispobaoke@gmail.com>
Co-authored-by: Zhiqiang Xie <xiezhq@stanford.edu>
Co-authored-by: yueming-yuan <yym022502@gmail.com>
Co-authored-by: DarkSharpness <2040703891@qq.com>
Co-authored-by: Yuhao Yang <47235274+yhyang201@users.noreply.github.com>
Co-authored-by: yhyang201 <yhyang201@users.noreply.github.com>
Co-authored-by: yhyang201 <yhyang201@gmail.com>
Co-authored-by: Qiaolin Yu <90088090+qiaolin-yu@users.noreply.github.com>
Co-authored-by: Ethan (Yusheng) Su <11704492+yushengsu-thu@users.noreply.github.com>
Co-authored-by: Mingyi <27337995+wisclmy0611@users.noreply.github.com>
Co-authored-by: Cheng Wan <54331508+ch-wan@users.noreply.github.com>
Co-authored-by: Yihao Wang <42559837+againstentropy@users.noreply.github.com>

scripts/ci/cuda/ci_install_flash_mla.sh

@@ -0,0 +1,35 @@
+#!/bin/bash
+set -euxo pipefail
+
+source scripts/ci/cuda/ci_install_dependency.sh
+
+if [ -z "${PIP_CMD:-}" ]; then
+    echo "FATAL:PIP_CMD is unset after sourcing ci_install_dependency.sh"
+    exit 1
+fi
+
+export CUDA_HOME=/usr/local/cuda
+
+if [ "${FORCE_REBUILD_FLASH_MLA:-0}" = "1" ]; then
+    echo "FORCE_REBUILD_FLASH_MLA=1; uninstalling any cached flash_mla before rebuild."
+    ${PIP_UNINSTALL_CMD:-pip uninstall -y} flash_mla ${PIP_UNINSTALL_SUFFIX:-} || true
+elif python3 -c "import flash_mla" >/dev/null 2>&1; then
+    echo "flash_mla is already installed or importable. Skipping installation."
+    exit 0
+fi
+
+# CUDA 13.0 puts CCCL headers under /usr/local/cuda/include/cccl/cuda but
+# FlashMLA's build expects them at /usr/local/cuda/include/cuda. Symlink so
+# the compiler finds them. Idempotent: skip if the link/dir already exists.
+if [ ! -e /usr/local/cuda/include/cuda ] && [ -d /usr/local/cuda/include/cccl/cuda ]; then
+    ln -s /usr/local/cuda/include/cccl/cuda /usr/local/cuda/include/cuda
+fi
+
+# Install FlashMLA
+FLASH_MLA_DIR=/root/.cache/flash-mla
+rm -rf ${FLASH_MLA_DIR}
+git clone https://github.com/deepseek-ai/FlashMLA.git ${FLASH_MLA_DIR}
+pushd ${FLASH_MLA_DIR}
+git submodule update --init --recursive
+${PIP_CMD:-pip} install --no-build-isolation -v . ${PIP_INSTALL_SUFFIX:-}
+popd