Skip to content

GitLab

Commit b3233ef9 authored by silencealiang's avatar silencealiang
Browse files

fix bug

parent 0ead9141
Pipeline #2550 passed with stage
Hide whitespace changes
Inline Side-by-side
Showing with 343 additions and 19 deletions
README.md
View file @ b3233ef9
......@@ -16,7 +16,9 @@
# 更新日志
2025.3.14适配最新代码，shell启动脚本在examples对应模型目录下
2025.3.14适配最新代码，shell启动脚本在examples对应模型目录下，模型相关数据集
[下载]: https://r0ddbu55vzx.feishu.cn/drive/folder/ZxHHfCoX4lg75td2hTqcmiAin3g
2024.12.16适配了torch prof
......
examples/gpt3/run_gpt_567B_multinodes.sh
View file @ b3233ef9
......@@ -16,4 +16,3 @@ mpirun -np 512 --hostfile hostfile_gpt_567B \
wait
rm -rf CKPT
#rm -rf mixtral_dataset/my-mixtral_text_document
examples/gpt3/topo-input.xml 0 → 100644
View file @ b3233ef9
<system version="2">
<cpu numaid="3" affinity="00000000,00000000,ffff0000,00000000,00000000,00000000,ffff0000,00000000" arch="x86_64" vendor="HygonGenuine" familyid="159" modelid="4">
<pci busid="0000:99:00.0" class="0x060400" vendor="0x1000" device="0xc030" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:9d:00.0" class="0x060400" vendor="0x1d94" device="0x23b7" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:9f:00.0" class="0x0b4000" vendor="0x1d94" device="0x6320" subsystem_vendor="0x1d94" subsystem_device="0x6310" link_speed="32.0 GT/s PCIe" link_width="16">
<gpu dev="0" sm="93" gcn="gfx936" arch="169983" rank="0" gdr="1">
<xgmi target="0000:56:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:5d:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:05:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:e5:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:ca:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:b1:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:c1:00.0" count="7" tclass="0x0b4000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:51:00.0" class="0x060400" vendor="0x1000" device="0xc030" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:54:00.0" class="0x060400" vendor="0x1d94" device="0x23b7" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:56:00.0" class="0x0b4000" vendor="0x1d94" device="0x6320" subsystem_vendor="0x1d94" subsystem_device="0x6310" link_speed="32.0 GT/s PCIe" link_width="16">
<gpu dev="1" sm="93" gcn="gfx936" arch="169983" rank="1" gdr="1">
<xgmi target="0000:9f:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:5d:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:05:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:e5:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:ca:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:b1:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:c1:00.0" count="7" tclass="0x0b4000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:9b:00.0" class="0x020000" vendor="0x15b3" device="0x1021" subsystem_vendor="0x15b3" subsystem_device="0x0022" link_speed="32.0 GT/s PCIe" link_width="16">
<nic>
<net name="mlx5_2" dev="2" speed="200000" port="1" latency="0.000000" guid="0x2227a1000373255c" maxconn="131072" gdr="1"/>
<net name="mlx5_3" dev="3" speed="200000" port="2" latency="0.000000" guid="0x2227a1000373255c" maxconn="131072" gdr="1"/>
</nic>
</pci>
</pci>
</cpu>
<cpu numaid="0" affinity="00000000,00000000,00000000,0000ffff,00000000,00000000,00000000,0000ffff" arch="x86_64" vendor="HygonGenuine" familyid="159" modelid="4">
<pci busid="0000:01:00.0" class="0x060400" vendor="0x1000" device="0xc030" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:03:00.0" class="0x060400" vendor="0x1d94" device="0x23b7" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:05:00.0" class="0x0b4000" vendor="0x1d94" device="0x6320" subsystem_vendor="0x1d94" subsystem_device="0x6310" link_speed="32.0 GT/s PCIe" link_width="16">
<gpu dev="3" sm="93" gcn="gfx936" arch="169983" rank="3" gdr="1">
<xgmi target="0000:9f:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:56:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:5d:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:e5:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:ca:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:b1:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:c1:00.0" count="7" tclass="0x0b4000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:59:00.0" class="0x060400" vendor="0x1000" device="0xc030" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:5b:00.0" class="0x060400" vendor="0x1d94" device="0x23b7" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:5d:00.0" class="0x0b4000" vendor="0x1d94" device="0x6320" subsystem_vendor="0x1d94" subsystem_device="0x6310" link_speed="32.0 GT/s PCIe" link_width="16">
<gpu dev="2" sm="93" gcn="gfx936" arch="169983" rank="2" gdr="1">
<xgmi target="0000:9f:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:56:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:05:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:e5:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:ca:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:b1:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:c1:00.0" count="7" tclass="0x0b4000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:06:00.0" class="0x020000" vendor="0x15b3" device="0x1021" subsystem_vendor="0x15b3" subsystem_device="0x0022" link_speed="32.0 GT/s PCIe" link_width="16">
<nic>
<net name="mlx5_4" dev="4" speed="200000" port="1" latency="0.000000" guid="0x8228a1000373255c" maxconn="131072" gdr="1"/>
<net name="mlx5_5" dev="5" speed="200000" port="2" latency="0.000000" guid="0x8228a1000373255c" maxconn="131072" gdr="1"/>
</nic>
</pci>
</pci>
</cpu>
<cpu numaid="7" affinity="7fff0000,00000000,00000000,00000000,ffff0000,00000000,00000000,00000000" arch="x86_64" vendor="HygonGenuine" familyid="159" modelid="4">
<pci busid="0000:e1:00.0" class="0x060400" vendor="0x1000" device="0xc030" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:e3:00.0" class="0x060400" vendor="0x1d94" device="0x23b7" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:e5:00.0" class="0x0b4000" vendor="0x1d94" device="0x6320" subsystem_vendor="0x1d94" subsystem_device="0x6310" link_speed="32.0 GT/s PCIe" link_width="16">
<gpu dev="4" sm="93" gcn="gfx936" arch="169983" rank="4" gdr="1">
<xgmi target="0000:9f:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:56:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:5d:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:05:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:ca:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:b1:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:c1:00.0" count="7" tclass="0x0b4000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:bd:00.0" class="0x060400" vendor="0x1000" device="0xc030" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:bf:00.0" class="0x060400" vendor="0x1d94" device="0x23b7" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:c1:00.0" class="0x0b4000" vendor="0x1d94" device="0x6320" subsystem_vendor="0x1d94" subsystem_device="0x6310" link_speed="32.0 GT/s PCIe" link_width="16">
<gpu dev="5" sm="93" gcn="gfx936" arch="169983" rank="5" gdr="1">
<xgmi target="0000:9f:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:56:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:5d:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:05:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:e5:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:ca:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:b1:00.0" count="7" tclass="0x0b4000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:e6:00.0" class="0x020000" vendor="0x15b3" device="0x1021" subsystem_vendor="0x15b3" subsystem_device="0x0022" link_speed="32.0 GT/s PCIe" link_width="16">
<nic>
<net name="mlx5_6" dev="6" speed="200000" port="1" latency="0.000000" guid="0x6227a1000373255c" maxconn="131072" gdr="1"/>
<net name="mlx5_7" dev="7" speed="200000" port="2" latency="0.000000" guid="0x6227a1000373255c" maxconn="131072" gdr="1"/>
</nic>
</pci>
</pci>
</cpu>
<cpu numaid="4" affinity="00000000,0000ffff,00000000,00000000,00000000,0000ffff,00000000,00000000" arch="x86_64" vendor="HygonGenuine" familyid="159" modelid="4">
<pci busid="0000:ab:00.0" class="0x060400" vendor="0x1000" device="0xc030" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:af:00.0" class="0x060400" vendor="0x1d94" device="0x23b7" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:b1:00.0" class="0x0b4000" vendor="0x1d94" device="0x6320" subsystem_vendor="0x1d94" subsystem_device="0x6310" link_speed="32.0 GT/s PCIe" link_width="16">
<gpu dev="7" sm="93" gcn="gfx936" arch="169983" rank="7" gdr="1">
<xgmi target="0000:9f:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:56:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:5d:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:05:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:e5:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:ca:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:c1:00.0" count="7" tclass="0x0b4000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:c5:00.0" class="0x060400" vendor="0x1000" device="0xc030" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:c8:00.0" class="0x060400" vendor="0x1d94" device="0x23b7" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:ca:00.0" class="0x0b4000" vendor="0x1d94" device="0x6320" subsystem_vendor="0x1d94" subsystem_device="0x6310" link_speed="32.0 GT/s PCIe" link_width="16">
<gpu dev="6" sm="93" gcn="gfx936" arch="169983" rank="6" gdr="1">
<xgmi target="0000:9f:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:56:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:5d:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:05:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:e5:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:b1:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:c1:00.0" count="7" tclass="0x0b4000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:ad:00.0" class="0x020000" vendor="0x15b3" device="0x1021" subsystem_vendor="0x15b3" subsystem_device="0x0022" link_speed="32.0 GT/s PCIe" link_width="16">
<nic>
<net name="mlx5_8" dev="8" speed="200000" port="1" latency="0.000000" guid="0xd226a1000373255c" maxconn="131072" gdr="1"/>
<net name="mlx5_9" dev="9" speed="200000" port="2" latency="0.000000" guid="0xd226a1000373255c" maxconn="131072" gdr="1"/>
</nic>
</pci>
</pci>
</cpu>
<cpu numaid="2" affinity="00000000,00000000,0000ffff,00000000,00000000,00000000,0000ffff,00000000" arch="x86_64" vendor="HygonGenuine" familyid="159" modelid="4">
<pci busid="0000:71:00.0" class="0x020000" vendor="0x15b3" device="0xa2dc" subsystem_vendor="0x15b3" subsystem_device="0x0009" link_speed="32.0 GT/s PCIe" link_width="16">
<nic>
<net name="mlx5_0" dev="0" speed="200000" port="1" latency="0.000000" guid="0xc0d00a000324e9b8" maxconn="131072" gdr="1"/>
<net name="mlx5_1" dev="1" speed="40000" port="2" latency="0.000000" guid="0xc0d00a000324e9b8" maxconn="131072" gdr="1"/>
</nic>
</pci>
</cpu>
</system>
examples/gpt3/train_gpt_567B_1nodes.sh
View file @ b3233ef9
......@@ -10,7 +10,7 @@ done
# Runs GPT 567B model
source /opt/dtk/env.sh
# default env
# defauat env
CURRENT_DIR="$( cd "$( dirname "$0" )" && pwd )"
MEGATRON_PATH=$( dirname $( dirname ${CURRENT_DIR}))
export PYTHONPATH=${MEGATRON_PATH}:$PYTHONPATH
......@@ -28,7 +28,7 @@ export NCCL_NET_GDR_LEVEL=7
export NCCL_NET_GDR_READ=1
export RCCL_SDMA_COPY_ENABLE=0
export NCCL_IB_HCA=mlx5_2:1,mlx5_3:1,mlx5_4:1,mlx5_5:1,mlx5_6:1,mlx5_7:1,mlx5_8:1,mlx5_9:1
export NCCL_TOPO_FILE="/public/home/xingjl/dependency/rccl-tests-0204/topo-input.xml"
export NCCL_TOPO_FILE="./topo-input.xml"
# enable BatchLinear
export GROUPED_GEMM_BatchLinear=1
......@@ -40,8 +40,8 @@ DIST_URL=${1}
DIST_PORT=25900
CHECKPOINT_PATH=./CKPT
TOKENIZER_MODEL=./mixtral_dataset/tokenizer.model
DATA_PATH=./mixtral_dataset/my-mixtral_text_document
TOKENIZER_MODEL="path to tokenizer.model"
DATA_PATH="path to my-mixtral_text_document"
DISTRIBUTED_ARGS=(
--rank ${RANK}
......
examples/gpt3/train_gpt_567B_multinodes.sh
View file @ b3233ef9
......@@ -10,7 +10,7 @@ done
# Runs GPT 567B model
source /opt/dtk/env.sh
# default env
# defauat env
CURRENT_DIR="$( cd "$( dirname "$0" )" && pwd )"
MEGATRON_PATH=$( dirname $( dirname ${CURRENT_DIR}))
export PYTHONPATH=${MEGATRON_PATH}:$PYTHONPATH
......@@ -28,7 +28,7 @@ export NCCL_NET_GDR_LEVEL=7
export NCCL_NET_GDR_READ=1
export RCCL_SDMA_COPY_ENABLE=0
export NCCL_IB_HCA=mlx5_2:1,mlx5_3:1,mlx5_4:1,mlx5_5:1,mlx5_6:1,mlx5_7:1,mlx5_8:1,mlx5_9:1
export NCCL_TOPO_FILE="/public/home/xingjl/dependency/rccl-tests-0204/topo-input.xml"
export NCCL_TOPO_FILE="./topo-input.xml"
# enable BatchLinear
export GROUPED_GEMM_BatchLinear=1
......@@ -40,8 +40,8 @@ DIST_URL=${1}
DIST_PORT=25900
CHECKPOINT_PATH=./CKPT
TOKENIZER_MODEL=./mixtral_dataset/tokenizer.model
DATA_PATH=./mixtral_dataset/my-mixtral_text_document
TOKENIZER_MODEL="path to tokenizer.model"
DATA_PATH="path to my-mixtral_text_document"
DISTRIBUTED_ARGS=(
--rank ${RANK}
......
examples/mixtral/run_mixtral_8x7B_multinodes.sh
View file @ b3233ef9
......@@ -16,4 +16,3 @@ mpirun -np 32 --hostfile hostfile_mixtral_8x7B \
wait
rm -rf CKPT
#rm -rf mixtral_dataset/my-mixtral_text_document
examples/mixtral/topo-input.xml 0 → 100644
View file @ b3233ef9
<system version="2">
<cpu numaid="3" affinity="00000000,00000000,ffff0000,00000000,00000000,00000000,ffff0000,00000000" arch="x86_64" vendor="HygonGenuine" familyid="159" modelid="4">
<pci busid="0000:99:00.0" class="0x060400" vendor="0x1000" device="0xc030" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:9d:00.0" class="0x060400" vendor="0x1d94" device="0x23b7" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:9f:00.0" class="0x0b4000" vendor="0x1d94" device="0x6320" subsystem_vendor="0x1d94" subsystem_device="0x6310" link_speed="32.0 GT/s PCIe" link_width="16">
<gpu dev="0" sm="93" gcn="gfx936" arch="169983" rank="0" gdr="1">
<xgmi target="0000:56:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:5d:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:05:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:e5:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:ca:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:b1:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:c1:00.0" count="7" tclass="0x0b4000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:51:00.0" class="0x060400" vendor="0x1000" device="0xc030" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:54:00.0" class="0x060400" vendor="0x1d94" device="0x23b7" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:56:00.0" class="0x0b4000" vendor="0x1d94" device="0x6320" subsystem_vendor="0x1d94" subsystem_device="0x6310" link_speed="32.0 GT/s PCIe" link_width="16">
<gpu dev="1" sm="93" gcn="gfx936" arch="169983" rank="1" gdr="1">
<xgmi target="0000:9f:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:5d:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:05:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:e5:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:ca:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:b1:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:c1:00.0" count="7" tclass="0x0b4000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:9b:00.0" class="0x020000" vendor="0x15b3" device="0x1021" subsystem_vendor="0x15b3" subsystem_device="0x0022" link_speed="32.0 GT/s PCIe" link_width="16">
<nic>
<net name="mlx5_2" dev="2" speed="200000" port="1" latency="0.000000" guid="0x2227a1000373255c" maxconn="131072" gdr="1"/>
<net name="mlx5_3" dev="3" speed="200000" port="2" latency="0.000000" guid="0x2227a1000373255c" maxconn="131072" gdr="1"/>
</nic>
</pci>
</pci>
</cpu>
<cpu numaid="0" affinity="00000000,00000000,00000000,0000ffff,00000000,00000000,00000000,0000ffff" arch="x86_64" vendor="HygonGenuine" familyid="159" modelid="4">
<pci busid="0000:01:00.0" class="0x060400" vendor="0x1000" device="0xc030" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:03:00.0" class="0x060400" vendor="0x1d94" device="0x23b7" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:05:00.0" class="0x0b4000" vendor="0x1d94" device="0x6320" subsystem_vendor="0x1d94" subsystem_device="0x6310" link_speed="32.0 GT/s PCIe" link_width="16">
<gpu dev="3" sm="93" gcn="gfx936" arch="169983" rank="3" gdr="1">
<xgmi target="0000:9f:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:56:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:5d:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:e5:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:ca:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:b1:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:c1:00.0" count="7" tclass="0x0b4000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:59:00.0" class="0x060400" vendor="0x1000" device="0xc030" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:5b:00.0" class="0x060400" vendor="0x1d94" device="0x23b7" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:5d:00.0" class="0x0b4000" vendor="0x1d94" device="0x6320" subsystem_vendor="0x1d94" subsystem_device="0x6310" link_speed="32.0 GT/s PCIe" link_width="16">
<gpu dev="2" sm="93" gcn="gfx936" arch="169983" rank="2" gdr="1">
<xgmi target="0000:9f:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:56:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:05:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:e5:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:ca:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:b1:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:c1:00.0" count="7" tclass="0x0b4000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:06:00.0" class="0x020000" vendor="0x15b3" device="0x1021" subsystem_vendor="0x15b3" subsystem_device="0x0022" link_speed="32.0 GT/s PCIe" link_width="16">
<nic>
<net name="mlx5_4" dev="4" speed="200000" port="1" latency="0.000000" guid="0x8228a1000373255c" maxconn="131072" gdr="1"/>
<net name="mlx5_5" dev="5" speed="200000" port="2" latency="0.000000" guid="0x8228a1000373255c" maxconn="131072" gdr="1"/>
</nic>
</pci>
</pci>
</cpu>
<cpu numaid="7" affinity="7fff0000,00000000,00000000,00000000,ffff0000,00000000,00000000,00000000" arch="x86_64" vendor="HygonGenuine" familyid="159" modelid="4">
<pci busid="0000:e1:00.0" class="0x060400" vendor="0x1000" device="0xc030" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:e3:00.0" class="0x060400" vendor="0x1d94" device="0x23b7" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:e5:00.0" class="0x0b4000" vendor="0x1d94" device="0x6320" subsystem_vendor="0x1d94" subsystem_device="0x6310" link_speed="32.0 GT/s PCIe" link_width="16">
<gpu dev="4" sm="93" gcn="gfx936" arch="169983" rank="4" gdr="1">
<xgmi target="0000:9f:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:56:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:5d:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:05:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:ca:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:b1:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:c1:00.0" count="7" tclass="0x0b4000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:bd:00.0" class="0x060400" vendor="0x1000" device="0xc030" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:bf:00.0" class="0x060400" vendor="0x1d94" device="0x23b7" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:c1:00.0" class="0x0b4000" vendor="0x1d94" device="0x6320" subsystem_vendor="0x1d94" subsystem_device="0x6310" link_speed="32.0 GT/s PCIe" link_width="16">
<gpu dev="5" sm="93" gcn="gfx936" arch="169983" rank="5" gdr="1">
<xgmi target="0000:9f:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:56:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:5d:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:05:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:e5:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:ca:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:b1:00.0" count="7" tclass="0x0b4000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:e6:00.0" class="0x020000" vendor="0x15b3" device="0x1021" subsystem_vendor="0x15b3" subsystem_device="0x0022" link_speed="32.0 GT/s PCIe" link_width="16">
<nic>
<net name="mlx5_6" dev="6" speed="200000" port="1" latency="0.000000" guid="0x6227a1000373255c" maxconn="131072" gdr="1"/>
<net name="mlx5_7" dev="7" speed="200000" port="2" latency="0.000000" guid="0x6227a1000373255c" maxconn="131072" gdr="1"/>
</nic>
</pci>
</pci>
</cpu>
<cpu numaid="4" affinity="00000000,0000ffff,00000000,00000000,00000000,0000ffff,00000000,00000000" arch="x86_64" vendor="HygonGenuine" familyid="159" modelid="4">
<pci busid="0000:ab:00.0" class="0x060400" vendor="0x1000" device="0xc030" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:af:00.0" class="0x060400" vendor="0x1d94" device="0x23b7" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:b1:00.0" class="0x0b4000" vendor="0x1d94" device="0x6320" subsystem_vendor="0x1d94" subsystem_device="0x6310" link_speed="32.0 GT/s PCIe" link_width="16">
<gpu dev="7" sm="93" gcn="gfx936" arch="169983" rank="7" gdr="1">
<xgmi target="0000:9f:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:56:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:5d:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:05:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:e5:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:ca:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:c1:00.0" count="7" tclass="0x0b4000"/>
</gpu>
</pci>
</pci>
<pci busid="0000:c5:00.0" class="0x060400" vendor="0x1000" device="0xc030" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:c8:00.0" class="0x060400" vendor="0x1d94" device="0x23b7" subsystem_vendor="0x1000" subsystem_device="0x100b" link_speed="32.0 GT/s PCIe" link_width="16">
<pci busid="0000:ca:00.0" class="0x0b4000" vendor="0x1d94" device="0x6320" subsystem_vendor="0x1d94" subsystem_device="0x6310" link_speed="32.0 GT/s PCIe" link_width="16">
<gpu dev="6" sm="93" gcn="gfx936" arch="169983" rank="6" gdr="1">
<xgmi target="0000:9f:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:56:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:5d:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:05:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:e5:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:b1:00.0" count="7" tclass="0x0b4000"/>
<xgmi target="0000:c1:00.0" count="7" tclass="0x0b4000"/>
</gpu>
</pci>
</pci>
</pci>
<pci busid="0000:ad:00.0" class="0x020000" vendor="0x15b3" device="0x1021" subsystem_vendor="0x15b3" subsystem_device="0x0022" link_speed="32.0 GT/s PCIe" link_width="16">
<nic>
<net name="mlx5_8" dev="8" speed="200000" port="1" latency="0.000000" guid="0xd226a1000373255c" maxconn="131072" gdr="1"/>
<net name="mlx5_9" dev="9" speed="200000" port="2" latency="0.000000" guid="0xd226a1000373255c" maxconn="131072" gdr="1"/>
</nic>
</pci>
</pci>
</cpu>
<cpu numaid="2" affinity="00000000,00000000,0000ffff,00000000,00000000,00000000,0000ffff,00000000" arch="x86_64" vendor="HygonGenuine" familyid="159" modelid="4">
<pci busid="0000:71:00.0" class="0x020000" vendor="0x15b3" device="0xa2dc" subsystem_vendor="0x15b3" subsystem_device="0x0009" link_speed="32.0 GT/s PCIe" link_width="16">
<nic>
<net name="mlx5_0" dev="0" speed="200000" port="1" latency="0.000000" guid="0xc0d00a000324e9b8" maxconn="131072" gdr="1"/>
<net name="mlx5_1" dev="1" speed="40000" port="2" latency="0.000000" guid="0xc0d00a000324e9b8" maxconn="131072" gdr="1"/>
</nic>
</pci>
</cpu>
</system>
examples/mixtral/train_mixtral_8x7B_1nodes.sh
View file @ b3233ef9
......@@ -10,7 +10,7 @@ done
# Runs Mixtral 8x7B model
source /opt/dtk/env.sh
# default env
# defauat env
CURRENT_DIR="$( cd "$( dirname "$0" )" && pwd )"
MEGATRON_PATH=$( dirname $( dirname ${CURRENT_DIR}))
export PYTHONPATH=${MEGATRON_PATH}:$PYTHONPATH
......@@ -28,7 +28,7 @@ export NCCL_NET_GDR_LEVEL=7
export NCCL_NET_GDR_READ=1
export RCCL_SDMA_COPY_ENABLE=0
export NCCL_IB_HCA=mlx5_2:1,mlx5_3:1,mlx5_4:1,mlx5_5:1,mlx5_6:1,mlx5_7:1,mlx5_8:1,mlx5_9:1
export NCCL_TOPO_FILE="/public/home/xingjl/dependency/rccl-tests-0204/topo-input.xml"
export NCCL_TOPO_FILE="./topo-input.xml"
# enable BatchLinear
export GROUPED_GEMM_BatchLinear=1
......@@ -40,8 +40,8 @@ DIST_URL=${1}
DIST_PORT=25900
CHECKPOINT_PATH=./CKPT
TOKENIZER_MODEL=./mixtral_dataset/tokenizer.model
DATA_PATH=./mixtral_dataset/my-mixtral_text_document
TOKENIZER_MODEL="path to tokenizer.model"
DATA_PATH="path to my-mixtral_text_document"
DISTRIBUTED_ARGS=(
--rank ${RANK}
......
examples/mixtral/train_mixtral_8x7B_multinodes.sh
View file @ b3233ef9
......@@ -10,7 +10,7 @@ done
# Runs Mixtral 8x7B model
source /opt/dtk/env.sh
# default env
# defauat env
CURRENT_DIR="$( cd "$( dirname "$0" )" && pwd )"
MEGATRON_PATH=$( dirname $( dirname ${CURRENT_DIR}))
export PYTHONPATH=${MEGATRON_PATH}:$PYTHONPATH
......@@ -28,7 +28,7 @@ export NCCL_NET_GDR_LEVEL=7
export NCCL_NET_GDR_READ=1
export RCCL_SDMA_COPY_ENABLE=0
export NCCL_IB_HCA=mlx5_2:1,mlx5_3:1,mlx5_4:1,mlx5_5:1,mlx5_6:1,mlx5_7:1,mlx5_8:1,mlx5_9:1
export NCCL_TOPO_FILE="/public/home/xingjl/dependency/rccl-tests-0204/topo-input.xml"
export NCCL_TOPO_FILE="./topo-input.xml"
# enable BatchLinear
export GROUPED_GEMM_BatchLinear=1
......@@ -40,8 +40,8 @@ DIST_URL=${1}
DIST_PORT=25900
CHECKPOINT_PATH=./CKPT
TOKENIZER_MODEL=./mixtral_dataset/tokenizer.model
DATA_PATH=./mixtral_dataset/my-mixtral_text_document
TOKENIZER_MODEL="path to tokenizer.model"
DATA_PATH="path to my-mixtral_text_document"
DISTRIBUTED_ARGS=(
--rank ${RANK}
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment