From cd362c5ca07bcc9af0816b91b5f9c822d41547d0 Mon Sep 17 00:00:00 2001
From: Zhiyi Hong <2497491955@qq.com>
Date: Tue, 28 Jul 2026 17:20:25 +0800
Subject: [PATCH] fix: max_running_requests=256, cuda_graph_max_bs=256 to match
 max concurrency 128

---
 .../pro6000/dsv4_pro6000_sglang_tp16/config.env        | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)

diff --git a/experiments/pro6000/dsv4_pro6000_sglang_tp16/config.env b/experiments/pro6000/dsv4_pro6000_sglang_tp16/config.env
index 3c74275..f6c29da 100644
--- a/experiments/pro6000/dsv4_pro6000_sglang_tp16/config.env
+++ b/experiments/pro6000/dsv4_pro6000_sglang_tp16/config.env
@@ -32,9 +32,11 @@ NCCL_SOCKET_IFNAME="${NCCL_SOCKET_IFNAME:-eth0}"
 NCCL_DEBUG="${NCCL_DEBUG:-WARN}"
 
 # ---- SGLang memory / perf ----
+# max_running_requests must >= max concurrency tested (128) to avoid queueing
+# cuda_graph_max_bs_decode must >= max concurrent batch size for accurate decode perf
 MEM_FRACTION_STATIC="${MEM_FRACTION_STATIC:-0.9}"
-CUDA_GRAPH_MAX_BS_DECODE="${CUDA_GRAPH_MAX_BS_DECODE:-8}"
-MAX_RUNNING_REQUESTS="${MAX_RUNNING_REQUESTS:-32}"
+CUDA_GRAPH_MAX_BS_DECODE="${CUDA_GRAPH_MAX_BS_DECODE:-256}"
+MAX_RUNNING_REQUESTS="${MAX_RUNNING_REQUESTS:-256}"
 
 # ---- Benchmark ----
 BENCH_DATASET_NAME="${BENCH_DATASET_NAME:-random}"
@@ -44,10 +46,6 @@ RANDOM_RANGE_RATIO="${RANDOM_RANGE_RATIO:-1.0}"
 WARMUP_REQUESTS="${WARMUP_REQUESTS:-16}"
 NUM_REPETITIONS="${NUM_REPETITIONS:-3}"
 
-# ---- SLO ----
-TTFT_SLO_MS="${TTFT_SLO_MS:-4000}"
-ENABLE_TTFT_SLO_STOP="${ENABLE_TTFT_SLO_STOP:-1}"
-
 # ---- Timeouts ----
 SCENARIO_TIMEOUT_S="${SCENARIO_TIMEOUT_S:-1800}"
 HEALTH_CHECK_RETRIES="${HEALTH_CHECK_RETRIES:-600}"