50 lines
2.3 KiB
Bash
50 lines
2.3 KiB
Bash
#!/usr/bin/env bash
|
|
|
|
set -euo pipefail
|
|
|
|
OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT must be set}"
|
|
VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}"
|
|
VLLM_SOURCE="${VLLM_SOURCE:-/home/admin/cpfs/wjh/agentic-kv/third_party/vllm_v20_build}"
|
|
MODEL="${MODEL:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}"
|
|
TOKENS=(1 8 16 32 64 128 256 512 1024 2048 4096 8192)
|
|
mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" "${OUTPUT_ROOT}/raw"
|
|
exec > >(tee -a "${OUTPUT_ROOT}/logs/full.log") 2>&1
|
|
|
|
IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES:?fleet GPU is required}"
|
|
if [[ "${#GPU_IDS[@]}" -ne 1 ]]; then
|
|
echo "ERROR: expected exactly one GPU, got ${CUDA_VISIBLE_DEVICES}" >&2
|
|
exit 1
|
|
fi
|
|
|
|
echo "PROFILE_LAUNCH_ECHO host=$(hostname) gpu=${CUDA_VISIBLE_DEVICES} model=${MODEL} runtime=vLLM-0.20.0+cu129 operators=ReplicatedLinear,fused_topk tokens=${TOKENS[*]} dtype=BF16 output=${OUTPUT_ROOT} expected_wall=2-6m hard_wall=900s hard_gpu_cap=0.25_H20h"
|
|
date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ"
|
|
nvidia-smi --query-gpu=index,name,driver_version,memory.used,utilization.gpu --format=csv,noheader
|
|
|
|
test "$(git -C "${VLLM_SOURCE}" rev-parse HEAD)" = "88d34c6409e9fb3c7b8ca0c04756f061d2099eb1"
|
|
test -s "${MODEL}/config.json"
|
|
git rev-parse HEAD > "${OUTPUT_ROOT}/provenance/aituner.commit"
|
|
git -C "${VLLM_SOURCE}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/vllm-source.commit"
|
|
sha256sum profile_vllm020_router.py run_router_full.sh \
|
|
> "${OUTPUT_ROOT}/provenance/source.sha256"
|
|
sha256sum "${MODEL}/config.json" > "${OUTPUT_ROOT}/provenance/model-config.sha256"
|
|
uv pip freeze --python "${VENV_ROOT}/bin/python" \
|
|
> "${OUTPUT_ROOT}/provenance/pip-freeze.txt"
|
|
nvidia-smi --query-gpu=index,uuid,name,driver_version,memory.total \
|
|
--format=csv,noheader > "${OUTPUT_ROOT}/provenance/gpus.csv"
|
|
printf '%s\n' "${TOKENS[@]}" > "${OUTPUT_ROOT}/provenance/tokens.txt"
|
|
|
|
timeout --signal=TERM --kill-after=30s 780 \
|
|
"${VENV_ROOT}/bin/python" profile_vllm020_router.py \
|
|
--vllm-source "${VLLM_SOURCE}" \
|
|
--model "${MODEL}" \
|
|
--output "${OUTPUT_ROOT}/raw/router.json" \
|
|
--num-tokens "${TOKENS[@]}" \
|
|
--warmup-iters 5 \
|
|
--repeats 20
|
|
|
|
test -s "${OUTPUT_ROOT}/raw/router.json"
|
|
sha256sum "${OUTPUT_ROOT}/raw/router.json" "${OUTPUT_ROOT}/provenance"/* \
|
|
> "${OUTPUT_ROOT}/artifacts.sha256"
|
|
date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ"
|
|
echo "ROUTER_FULL_COMPLETE cases=${#TOKENS[@]}"
|