gahow/aituner
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Extend Qwen30 JIT readiness for latency surface

Browse Source
This commit is contained in:
Gahow Wang
2026-07-18 11:09:22 +08:00
parent 686b050517
commit ecb45cf762
2 changed files with 7 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
runs/frontier-fidelity-envelope-v1/run_qwen30_exact_trace_real_anchor.sh
View File

@@ -12,6 +12,7 @@ PREFIX_CACHING="${PREFIX_CACHING:-true}"
VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}"
MODEL_ROOT="${MODEL_ROOT:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}"
FLASHINFER_WORKSPACE_BASE="${FLASHINFER_WORKSPACE_BASE:-/tmp/wjh/flashinfer-workspace-vllm020-profiler-v1}"
SERVER_READY_ATTEMPTS="${SERVER_READY_ATTEMPTS:-120}"
SERVED_MODEL="qwen3-30b-exact-trace"
EXACT_TRACE_CLIENT="${EXACT_TRACE_CLIENT:-qwen30_exact_trace_client.py}"
SERVER_PID=""
@@ -53,7 +54,7 @@ case "${PREFIX_CACHING}" in
esac
REQUEST_COUNT="$(wc -l < "${REQUESTS_FILE}")"
echo "QWEN30_EXACT_TRACE_REAL_LAUNCH_ECHO host=$(hostname) gpus=${CUDA_VISIBLE_DEVICES} model=${MODEL_ROOT} runtime=vLLM-0.20.0+cu129 dtype=BF16 config=TP${TP}_MNS${MNS}_MBT8192 trace=${TRACE_LABEL} requests=${REQUEST_COUNT} source=${REQUESTS_FILE} arrivals=manifest prefix=${PREFIX_CACHING} block=16 metrics=TTFT,TPOT-if-OSL-gt-1,E2E flashinfer_workspace=${FLASHINFER_WORKSPACE_BASE} output=${OUTPUT_ROOT} expected_wall=12-35m hard_wall=3600s"
echo "QWEN30_EXACT_TRACE_REAL_LAUNCH_ECHO host=$(hostname) gpus=${CUDA_VISIBLE_DEVICES} model=${MODEL_ROOT} runtime=vLLM-0.20.0+cu129 dtype=BF16 config=TP${TP}_MNS${MNS}_MBT8192 trace=${TRACE_LABEL} requests=${REQUEST_COUNT} source=${REQUESTS_FILE} arrivals=manifest prefix=${PREFIX_CACHING} block=16 metrics=TTFT,TPOT-if-OSL-gt-1,E2E flashinfer_workspace=${FLASHINFER_WORKSPACE_BASE} output=${OUTPUT_ROOT} ready_budget_s=$((SERVER_READY_ATTEMPTS * 3)) hard_wall=3600s"
date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ"
sha256sum qwen30_exact_trace_client.py run_qwen30_exact_trace_real_anchor.sh \
../frontier-phase-factorial-v0/qwen30_prefill_client.py \
@@ -79,7 +80,7 @@ setsid "${VENV_ROOT}/bin/vllm" serve "${MODEL_ROOT}" \
> "${OUTPUT_ROOT}/logs/server.log" 2>&1 &
SERVER_PID=$!
READY=0
for _ in $(seq 1 120); do
for _ in $(seq 1 "${SERVER_READY_ATTEMPTS}"); do
if curl -fsS --max-time 2 "http://127.0.0.1:${SERVER_PORT}/v1/models" \
> "${OUTPUT_ROOT}/results/models.json" 2>/dev/null; then
READY=1
@@ -92,6 +93,7 @@ for _ in $(seq 1 120); do
sleep 3
done
if [[ "${READY}" -ne 1 ]]; then
echo "ERROR server did not become ready within $((SERVER_READY_ATTEMPTS * 3)) seconds" >&2
tail -200 "${OUTPUT_ROOT}/logs/server.log"
exit 1
fi

4
runs/frontier-fidelity-envelope-v1/run_qwen30_latency_case_real_surface.sh
View File

@@ -16,7 +16,8 @@ CLIENT="${RUNNER_DIR}/qwen30_exact_trace_client.py"
PREFILL_CLIENT="${RUNNER_DIR}/../frontier-phase-factorial-v0/qwen30_prefill_client.py"
VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}"
MODEL_ROOT="${MODEL_ROOT:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}"
FLASHINFER_SHARED_WORKSPACE="${FLASHINFER_SHARED_WORKSPACE:-${OUT}/flashinfer-shared-workspace}"
FLASHINFER_SHARED_WORKSPACE="${FLASHINFER_SHARED_WORKSPACE:-/tmp/wjh/flashinfer-workspace-vllm020-profiler-v1}"
SERVER_READY_ATTEMPTS="${SERVER_READY_ATTEMPTS:-300}"
PORT="${BASE_PORT:-8300}"
case "${PREFIX_CACHING}" in
@@ -97,6 +98,7 @@ launch_config() {
VENV_ROOT="${VENV_ROOT}" \
MODEL_ROOT="${MODEL_ROOT}" \
FLASHINFER_WORKSPACE_BASE="${FLASHINFER_SHARED_WORKSPACE}" \
SERVER_READY_ATTEMPTS="${SERVER_READY_ATTEMPTS}" \
EXACT_TRACE_CLIENT="${CLIENT}" \
timeout --signal=TERM --kill-after=60s 4200 bash "${RUNNER}" \
> "${run_out}/launcher.stdout.log" 2> "${run_out}/launcher.stderr.log"