mirror of
https://github.com/PaddlePaddle/FastDeploy.git
synced 2025-12-24 13:28:13 +08:00
0c8c6369ed
* [feat] simplify configuration for pd-disaggregated deployment, and refactor post-init and usage for all ports * [fix] fix some bugs * [fix] fix rdma port for cache manager/messager * [fix] temporarily cancel port availability check to see if it can pass ci test * [feat] simplify args for multi api server * [fix] fix dp * [fix] fix port for xpu * [fix] add tests for ports post processing & fix ci * [test] fix test_multi_api_server * [fix] fix rdma_comm_ports args for multi_api_server * [fix] fix test_common_engine * [fix] fix test_cache_transfer_manager * [chore] automatically setting FD_ENABLE_MULTI_API_SERVER * [fix] avoid api server from creating engine_args twice * [fix] fix test_run_batch * [fix] fix test_metrics * [fix] fix splitwise connector init * [test] add test_rdma_transfer and test_expert_service * [fix] fix code syntax * [fix] fix test_rdma_transfer and build wheel with rdma script
95 lines
2.4 KiB
Bash
95 lines
2.4 KiB
Bash
#!/bin/bash
|
|
set -e
|
|
|
|
# Test splitwise deployment
|
|
# There are two methods for splitwise deployment:
|
|
# v0: using splitwise_scheduler or dp_scheduler
|
|
# v1: using local_scheduler + router
|
|
|
|
MODEL_NAME="PaddlePaddle/ERNIE-4.5-0.3B-Paddle"
|
|
DATA_PARALLEL_SIZE=2
|
|
TENSOR_PARALLEL_SIZE=1
|
|
LOG_DATE=$(date +%Y%m%d_%H%M%S)
|
|
|
|
export FD_DEBUG=1
|
|
export FD_ENABLE_MULTI_API_SERVER=1
|
|
|
|
SCRIPT_PATH=$(readlink -f "$0")
|
|
SCRIPT_DIR=$(dirname "$SCRIPT_PATH")
|
|
source ${SCRIPT_DIR}/utils.sh
|
|
|
|
unset http_proxy && unset https_proxy
|
|
|
|
# start router
|
|
ROUTER_PORT=$(get_free_ports 1)
|
|
echo ROUTER_PORT: $ROUTER_PORT
|
|
|
|
export FD_LOG_DIR="log/$LOG_DATE/router"
|
|
rm -rf $FD_LOG_DIR
|
|
mkdir -p ${FD_LOG_DIR}
|
|
|
|
nohup python -m fastdeploy.router.launch \
|
|
--port ${ROUTER_PORT} \
|
|
--splitwise \
|
|
2>&1 >${FD_LOG_DIR}/nohup &
|
|
sleep 1
|
|
|
|
|
|
# start prefill
|
|
P_SERVER_PORTS=$(get_free_ports $DATA_PARALLEL_SIZE)
|
|
echo P_SERVER_PORTS: $P_SERVER_PORTS
|
|
|
|
export CUDA_VISIBLE_DEVICES="0,1"
|
|
export FD_LOG_DIR="log/$LOG_DATE/prefill"
|
|
rm -rf $FD_LOG_DIR
|
|
mkdir -p ${FD_LOG_DIR}
|
|
|
|
nohup python -m fastdeploy.entrypoints.openai.multi_api_server \
|
|
--num-servers ${DATA_PARALLEL_SIZE}\
|
|
--ports ${P_SERVER_PORTS} \
|
|
--args --model ${MODEL_NAME} \
|
|
--max-model-len 32768 \
|
|
--data-parallel-size ${DATA_PARALLEL_SIZE} \
|
|
--tensor-parallel-size ${TENSOR_PARALLEL_SIZE} \
|
|
--splitwise-role "prefill" \
|
|
--router "0.0.0.0:${ROUTER_PORT}" \
|
|
2>&1 >${FD_LOG_DIR}/nohup &
|
|
|
|
wait_for_health ${P_SERVER_PORTS}
|
|
|
|
|
|
# start decode
|
|
D_SERVER_PORTS=$(get_free_ports $DATA_PARALLEL_SIZE)
|
|
echo D_SERVER_PORTS: $D_SERVER_PORTS
|
|
|
|
export CUDA_VISIBLE_DEVICES="4,5"
|
|
export FD_LOG_DIR="log/$LOG_DATE/decode"
|
|
rm -rf $FD_LOG_DIR
|
|
mkdir -p ${FD_LOG_DIR}
|
|
|
|
nohup python -m fastdeploy.entrypoints.openai.multi_api_server \
|
|
--num-servers ${DATA_PARALLEL_SIZE}\
|
|
--ports ${D_SERVER_PORTS} \
|
|
--args --model ${MODEL_NAME} \
|
|
--max-model-len 32768 \
|
|
--data-parallel-size ${DATA_PARALLEL_SIZE} \
|
|
--tensor-parallel-size ${TENSOR_PARALLEL_SIZE} \
|
|
--splitwise-role "decode" \
|
|
--router "0.0.0.0:${ROUTER_PORT}" \
|
|
2>&1 >${FD_LOG_DIR}/nohup &
|
|
|
|
wait_for_health ${D_SERVER_PORTS}
|
|
|
|
|
|
# send request
|
|
sleep 10 # make sure server is registered to router
|
|
curl -X POST "http://0.0.0.0:${ROUTER_PORT}/v1/chat/completions" \
|
|
-H "Content-Type: application/json" \
|
|
-d '{
|
|
"messages": [
|
|
{"role": "user", "content": "hello"}
|
|
],
|
|
"max_tokens": 100,
|
|
"stream": false
|
|
}'
|