sskj-h3/throughput/orchestrator_3runs.sh

22 lines
1.1 KiB
Bash

#!/usr/bin/env bash
# Sequentially run: base -> cache-dit(3tier, plain model) -> cache+lora(TP4x2, 3tier)
set -Eeuo pipefail
LOGP=/data/wxy/sskj-h3/throughput
REC=/data/wxy/sskj-h3/throughput/sglang-lora/records_rewritten.json
cd /data/wxy/sskj-h3/throughput
echo "========== [1/3] BASE (new prompt, TP2x4, 20 steps) =========="
cd /data/wxy/sskj-h3/throughput/sglang-base
RECORDS=$REC RUN_ID="ref2va-feishu-base-tp2x4-newprompt-768p-15s-20steps-$(date +%Y%m%d-%H%M%S)" \
bash scripts/run_sglang_h3_ref2va_feishu_base_tp2x4.sh 2>&1 | tee $LOGP/base_newprompt_run.log
echo "========== [2/3] CACHE-DIT plain (new prompt, TP2x4, 3 tiers) =========="
cd /data/wxy/sskj-h3/throughput/sglang-cache-dit
RECORDS=$REC RUN_ID="ref2va-feishu-cachedit-3tier-newprompt-$(date +%Y%m%d-%H%M%S)" \
bash scripts/run_sglang_h3_ref2va_feishu_cachedit_3tier_tp2x4.sh 2>&1 | tee $LOGP/cachedit_newprompt_run.log
echo "========== [3/3] CACHE+LORA (new prompt, TP4x2, 3 tiers) =========="
bash /data/wxy/sskj-h3/throughput/run_cache_lora_tp4_3tier.sh 2>&1 | tee $LOGP/cache_lora_tp4_run.log
echo "========== ALL THREE COMPLETE =========="