9acf9fdfdb
- sskj.deploy runtime 支持 NODE_HOSTS 多节点编排(ssh 分发/本地 rank/LOCAL_NODE_RANK)
与 ENGINE=vllm 启动(SERVER_CMD)，容器名按 rank 自动唯一
- scripts/common/deploy_cli.sh 新增 deploy_stop/status/multinode helper 与 node-rank 透传
- src/sskj/common/env.py 修复嵌套 ${VAR:-${OTHER}/path} 展开(平衡花括号扫描)
- deploy/profiles/pro6000/ 新增 6 个 profile: tp16/tp16_eagle/glm52(多节点)、
sglang/vllm tp_dp_matrix、qwen3(单节点)
- 6 个实验 start/stop 脚本改为 deploy 薄包装，run_bench/adaptive 的 server 启停走
deploy_render_args/deploy_start/deploy_stop，tp16 新增 matrix.json
- 首次入库 glm52_pro6000_sglang_multinode_tp16 实验目录；ops/README.md 补 pro6000 章节
- 实测通过: 单节点 dsv4 sglang/vllm 链路 + tp16 双节点启动/bench/清理
111 lines
2.9 KiB
Bash
111 lines
2.9 KiB
Bash
#!/usr/bin/env bash
|
|
# Helpers to delegate server lifecycle and argument rendering to
|
|
# `python -m sskj.deploy`. The deployment profile is the single source of
|
|
# engine launch args; experiment scripts should call these helpers instead of
|
|
# maintaining their own copies.
|
|
|
|
DEPLOY_PYTHON="${PYTHON:-python3}"
|
|
|
|
deploy_profile_abs() {
|
|
local profile="$1"
|
|
printf '%s' "${ROOT_DIR}/deploy/profiles/${profile}.env"
|
|
}
|
|
|
|
deploy_start() {
|
|
local profile="$1"
|
|
local tp="$2"
|
|
local dp="$3"
|
|
local log_dir="$4"
|
|
local port="$5"
|
|
local model_path="$6"
|
|
local container="$7"
|
|
local node_rank="${8:-}"
|
|
local -a extra=()
|
|
if [[ -n "$node_rank" ]]; then
|
|
extra+=(--node-rank "$node_rank")
|
|
fi
|
|
PYTHONPATH="${ROOT_DIR}/src" "${DEPLOY_PYTHON}" -m sskj.deploy start \
|
|
--profile "$(deploy_profile_abs "$profile")" \
|
|
--tp "$tp" \
|
|
--dp "$dp" \
|
|
--log-dir "$log_dir" \
|
|
--port "$port" \
|
|
--model-path "$model_path" \
|
|
--container-name "$container" \
|
|
"${extra[@]}"
|
|
}
|
|
|
|
deploy_render_args() {
|
|
local profile="$1"
|
|
local tp="$2"
|
|
local dp="$3"
|
|
local port="$4"
|
|
local model_path="$5"
|
|
PYTHONPATH="${ROOT_DIR}/src" "${DEPLOY_PYTHON}" -m sskj.deploy render-args \
|
|
--profile "$(deploy_profile_abs "$profile")" \
|
|
--tp "$tp" \
|
|
--dp "$dp" \
|
|
--port "$port" \
|
|
--model-path "$model_path"
|
|
}
|
|
|
|
# Multi-node convenience helpers: orchestrate all NODE_HOSTS from one host.
|
|
|
|
deploy_start_multinode() {
|
|
local profile="$1"
|
|
local tp="$2"
|
|
local dp="$3"
|
|
local log_dir="$4"
|
|
local port="$5"
|
|
local model_path="$6"
|
|
local container="$7"
|
|
PYTHONPATH="${ROOT_DIR}/src" "${DEPLOY_PYTHON}" -m sskj.deploy start \
|
|
--profile "$(deploy_profile_abs "$profile")" \
|
|
--tp "$tp" \
|
|
--dp "$dp" \
|
|
--log-dir "$log_dir" \
|
|
--port "$port" \
|
|
--model-path "$model_path" \
|
|
--container-name "$container"
|
|
}
|
|
|
|
deploy_stop() {
|
|
local profile="$1"
|
|
local tp="$2"
|
|
local dp="$3"
|
|
local port="$4"
|
|
local model_path="$5"
|
|
local container="$6"
|
|
local node_rank="${7:-}"
|
|
local -a extra=()
|
|
if [[ -n "$node_rank" ]]; then
|
|
extra+=(--node-rank "$node_rank")
|
|
fi
|
|
PYTHONPATH="${ROOT_DIR}/src" "${DEPLOY_PYTHON}" -m sskj.deploy stop \
|
|
--profile "$(deploy_profile_abs "$profile")" \
|
|
--tp "$tp" \
|
|
--dp "$dp" \
|
|
--port "$port" \
|
|
--model-path "$model_path" \
|
|
--container-name "$container" \
|
|
"${extra[@]}"
|
|
}
|
|
|
|
deploy_status() {
|
|
local profile="$1"
|
|
local tp="${2:-}"
|
|
local dp="${3:-}"
|
|
local port="${4:-}"
|
|
local model_path="${5:-}"
|
|
local container="${6:-}"
|
|
local -a cmd=(
|
|
--profile "$(deploy_profile_abs "$profile")"
|
|
)
|
|
[[ -n "$tp" ]] && cmd+=(--tp "$tp")
|
|
[[ -n "$dp" ]] && cmd+=(--dp "$dp")
|
|
[[ -n "$port" ]] && cmd+=(--port "$port")
|
|
[[ -n "$model_path" ]] && cmd+=(--model-path "$model_path")
|
|
[[ -n "$container" ]] && cmd+=(--container-name "$container")
|
|
PYTHONPATH="${ROOT_DIR}/src" "${DEPLOY_PYTHON}" -m sskj.deploy status "${cmd[@]}"
|
|
}
|