Add exact vLLM 0.20 MoE profile smoke
This commit is contained in:
47
runs/frontier-qwen30-vllm020-profile-v1/run_moe_smoke.sh
Normal file
47
runs/frontier-qwen30-vllm020-profile-v1/run_moe_smoke.sh
Normal file
@@ -0,0 +1,47 @@
|
||||
#!/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}"
|
||||
mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" "${OUTPUT_ROOT}/raw"
|
||||
exec > >(tee -a "${OUTPUT_ROOT}/logs/smoke.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 operator=FlashInfer-CUTLASS-unquantized-MoE tp_local_shapes=1,2,4 tokens=8 routing=uniform_random_logits reference_check=on dtype=BF16 output=${OUTPUT_ROOT} expected_wall=3-8m hard_wall=900s hard_gpu_cap=0.25_H20h"
|
||||
date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ"
|
||||
nvidia-smi --query-gpu=index,name,driver_version,memory.used,utilization.gpu --format=csv,noheader
|
||||
|
||||
git rev-parse HEAD > "${OUTPUT_ROOT}/provenance/aituner.commit"
|
||||
git -C "${VLLM_SOURCE}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/vllm-source.commit"
|
||||
sha256sum profile_vllm020_moe.py run_moe_smoke.sh \
|
||||
> "${OUTPUT_ROOT}/provenance/source.sha256"
|
||||
uv pip freeze --python "${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"
|
||||
|
||||
timeout --signal=TERM --kill-after=30s 780 \
|
||||
"${VENV_ROOT}/bin/python" profile_vllm020_moe.py \
|
||||
--vllm-source "${VLLM_SOURCE}" \
|
||||
--model "${MODEL}" \
|
||||
--output "${OUTPUT_ROOT}/raw/moe-smoke.json" \
|
||||
--tp 1 2 4 \
|
||||
--num-tokens 8 \
|
||||
--routing-modes uniform_random_logits \
|
||||
--warmup-iters 3 \
|
||||
--repeats 5 \
|
||||
--check-reference
|
||||
|
||||
test -s "${OUTPUT_ROOT}/raw/moe-smoke.json"
|
||||
sha256sum "${OUTPUT_ROOT}/raw/moe-smoke.json" "${OUTPUT_ROOT}/provenance"/* \
|
||||
> "${OUTPUT_ROOT}/artifacts.sha256"
|
||||
date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ"
|
||||
echo "MOE_SMOKE_COMPLETE"
|
||||
Reference in New Issue
Block a user