From b4bd4005fe19e5cfb5d8ad6b0ca92cc8bc7f6035 Mon Sep 17 00:00:00 2001
From: "claude[bot]" <41898282+claude[bot]@users.noreply.github.com>
Date: Sat, 28 Mar 2026 05:32:42 +0000
Subject: [PATCH] Disable prefix caching for qwen3.5 & glm5 AMD benchmark
 scripts

Add --disable-radix-cache to SGLang server launch commands for:
- qwen3.5_bf16_mi300x.sh
- qwen3.5_bf16_mi325x.sh
- qwen3.5_bf16_mi355x.sh
- qwen3.5_fp8_mi300x.sh
- qwen3.5_fp8_mi325x.sh
- qwen3.5_fp8_mi355x.sh
- glm5_fp8_mi355x.sh

Closes #968

Co-authored-by: functionstackx <functionstackx@users.noreply.github.com>
---
 benchmarks/single_node/glm5_fp8_mi355x.sh     | 3 ++-
 benchmarks/single_node/qwen3.5_bf16_mi300x.sh | 3 ++-
 benchmarks/single_node/qwen3.5_bf16_mi325x.sh | 3 ++-
 benchmarks/single_node/qwen3.5_bf16_mi355x.sh | 3 ++-
 benchmarks/single_node/qwen3.5_fp8_mi300x.sh  | 3 ++-
 benchmarks/single_node/qwen3.5_fp8_mi325x.sh  | 3 ++-
 benchmarks/single_node/qwen3.5_fp8_mi355x.sh  | 3 ++-
 7 files changed, 14 insertions(+), 7 deletions(-)

diff --git a/benchmarks/single_node/glm5_fp8_mi355x.sh b/benchmarks/single_node/glm5_fp8_mi355x.sh
index ee11463ce..690298986 100755
--- a/benchmarks/single_node/glm5_fp8_mi355x.sh
+++ b/benchmarks/single_node/glm5_fp8_mi355x.sh
@@ -44,7 +44,8 @@ python3 -m sglang.launch_server \
     --mem-fraction-static 0.85 \
     --model-loader-extra-config '{"enable_multithread_load": true, "num_threads": 8}' \
     --nsa-prefill-backend tilelang \
-    --nsa-decode-backend tilelang > $SERVER_LOG 2>&1 &
+    --nsa-decode-backend tilelang \
+    --disable-radix-cache > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
 
diff --git a/benchmarks/single_node/qwen3.5_bf16_mi300x.sh b/benchmarks/single_node/qwen3.5_bf16_mi300x.sh
index ea10647d6..1c14ab021 100755
--- a/benchmarks/single_node/qwen3.5_bf16_mi300x.sh
+++ b/benchmarks/single_node/qwen3.5_bf16_mi300x.sh
@@ -31,7 +31,8 @@ python3 -m sglang.launch_server \
     --port $PORT \
     --tensor-parallel-size $TP \
     --trust-remote-code \
-    --mem-fraction-static 0.8 > $SERVER_LOG 2>&1 &
+    --mem-fraction-static 0.8 \
+    --disable-radix-cache > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
 
diff --git a/benchmarks/single_node/qwen3.5_bf16_mi325x.sh b/benchmarks/single_node/qwen3.5_bf16_mi325x.sh
index ea10647d6..1c14ab021 100644
--- a/benchmarks/single_node/qwen3.5_bf16_mi325x.sh
+++ b/benchmarks/single_node/qwen3.5_bf16_mi325x.sh
@@ -31,7 +31,8 @@ python3 -m sglang.launch_server \
     --port $PORT \
     --tensor-parallel-size $TP \
     --trust-remote-code \
-    --mem-fraction-static 0.8 > $SERVER_LOG 2>&1 &
+    --mem-fraction-static 0.8 \
+    --disable-radix-cache > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
 
diff --git a/benchmarks/single_node/qwen3.5_bf16_mi355x.sh b/benchmarks/single_node/qwen3.5_bf16_mi355x.sh
index f77390707..90163ccfd 100755
--- a/benchmarks/single_node/qwen3.5_bf16_mi355x.sh
+++ b/benchmarks/single_node/qwen3.5_bf16_mi355x.sh
@@ -30,7 +30,8 @@ python3 -m sglang.launch_server \
     --port $PORT \
     --tensor-parallel-size $TP \
     --trust-remote-code \
-    --mem-fraction-static 0.8 > $SERVER_LOG 2>&1 &
+    --mem-fraction-static 0.8 \
+    --disable-radix-cache > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
 
diff --git a/benchmarks/single_node/qwen3.5_fp8_mi300x.sh b/benchmarks/single_node/qwen3.5_fp8_mi300x.sh
index 0640a20ab..83006ca95 100755
--- a/benchmarks/single_node/qwen3.5_fp8_mi300x.sh
+++ b/benchmarks/single_node/qwen3.5_fp8_mi300x.sh
@@ -32,7 +32,8 @@ python3 -m sglang.launch_server \
     --port $PORT \
     --tensor-parallel-size $TP \
     --trust-remote-code \
-    --mem-fraction-static 0.8 > $SERVER_LOG 2>&1 &
+    --mem-fraction-static 0.8 \
+    --disable-radix-cache > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
 
diff --git a/benchmarks/single_node/qwen3.5_fp8_mi325x.sh b/benchmarks/single_node/qwen3.5_fp8_mi325x.sh
index 0640a20ab..83006ca95 100755
--- a/benchmarks/single_node/qwen3.5_fp8_mi325x.sh
+++ b/benchmarks/single_node/qwen3.5_fp8_mi325x.sh
@@ -32,7 +32,8 @@ python3 -m sglang.launch_server \
     --port $PORT \
     --tensor-parallel-size $TP \
     --trust-remote-code \
-    --mem-fraction-static 0.8 > $SERVER_LOG 2>&1 &
+    --mem-fraction-static 0.8 \
+    --disable-radix-cache > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
 
diff --git a/benchmarks/single_node/qwen3.5_fp8_mi355x.sh b/benchmarks/single_node/qwen3.5_fp8_mi355x.sh
index f77390707..90163ccfd 100644
--- a/benchmarks/single_node/qwen3.5_fp8_mi355x.sh
+++ b/benchmarks/single_node/qwen3.5_fp8_mi355x.sh
@@ -30,7 +30,8 @@ python3 -m sglang.launch_server \
     --port $PORT \
     --tensor-parallel-size $TP \
     --trust-remote-code \
-    --mem-fraction-static 0.8 > $SERVER_LOG 2>&1 &
+    --mem-fraction-static 0.8 \
+    --disable-radix-cache > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!