[AMD][MI35X]Update qwen3.5 perf

.github/configs/amd-master.yaml

@@ -114,7 +114,7 @@ dsr1-fp8-mi355x-sglang:
     - { tp: 8, conc-start: 4, conc-end: 64 }
 
 qwen3.5-bf16-mi355x-sglang:
-  image: rocm/sgl-dev:v0.5.8.post1-rocm720-mi35x-20260215
+  image: lmsysorg/sglang-rocm:v0.5.10rc0-rocm720-mi35x-20260415
   model: Qwen/Qwen3.5-397B-A17B
   model-prefix: qwen3.5
   runner: mi355x
@@ -125,11 +125,11 @@ qwen3.5-bf16-mi355x-sglang:
   - isl: 1024
     osl: 1024
     search-space:
-    - { tp: 8, conc-start: 4, conc-end: 64 }
+    - { tp: 8, ep: 1, conc-start: 4, conc-end: 256 }
   - isl: 8192
     osl: 1024
     search-space:
-    - { tp: 8, conc-start: 4, conc-end: 64 }
+    - { tp: 8, ep: 1, conc-start: 4, conc-end: 256 }
 
 qwen3.5-bf16-mi300x-sglang:
   image: lmsysorg/sglang:v0.5.10-rocm720-mi30x
@@ -186,7 +186,7 @@ qwen3.5-fp8-mi325x-sglang:
     - { tp: 8, conc-start: 4, conc-end: 64 }
 
 qwen3.5-fp8-mi355x-sglang:
-  image: rocm/sgl-dev:v0.5.8.post1-rocm720-mi35x-20260218
+  image: lmsysorg/sglang-rocm:v0.5.10rc0-rocm720-mi35x-20260415
   model: Qwen/Qwen3.5-397B-A17B-FP8
   model-prefix: qwen3.5
   runner: mi355x
@@ -197,11 +197,14 @@ qwen3.5-fp8-mi355x-sglang:
   - isl: 1024
     osl: 1024
     search-space:
-    - { tp: 8, conc-start: 4, conc-end: 64 }
+    - { tp: 8, ep: 1, conc-start: 4, conc-end: 32 }
+    - { tp: 8, ep: 8, conc-start: 64, conc-end: 256 }
+    - { tp: 2, ep: 2, conc-start: 128, conc-end: 256 }
   - isl: 8192
     osl: 1024
     search-space:
-    - { tp: 8, conc-start: 4, conc-end: 64 }
+    - { tp: 2, ep: 2, conc-start: 4, conc-end: 32 }
+    - { tp: 4, ep: 1, conc-start: 32, conc-end: 256 }
 
 qwen3.5-fp4-mi355x-sglang:
   image: lmsysorg/sglang:v0.5.10-rocm720-mi35x

benchmarks/single_node/qwen3.5_bf16_mi355x.sh

@@ -39,15 +39,14 @@ python3 -m sglang.launch_server \
     --port $PORT \
     --tensor-parallel-size $TP \
     --ep-size $EP_SIZE \
-    --data-parallel-size 1 \
     --trust-remote-code \
     --tokenizer-worker-num 6 \
     --enable-aiter-allreduce-fusion \
     --cuda-graph-max-bs $CONC \
     --disable-radix-cache \
     --max-prefill-tokens $MAX_PREFILL_TOKENS \
     --scheduler-recv-interval 30 \
-    --mem-fraction-static 0.75 $EVAL_CONTEXT_ARGS > $SERVER_LOG 2>&1 &
+    --mem-fraction-static 0.8 $EVAL_CONTEXT_ARGS > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
 

benchmarks/single_node/qwen3.5_fp8_mi355x.sh

@@ -39,15 +39,14 @@ python3 -m sglang.launch_server \
     --port $PORT \
     --tensor-parallel-size $TP \
     --ep-size $EP_SIZE \
-    --data-parallel-size 1 \
     --trust-remote-code \
     --tokenizer-worker-num 6 \
     --enable-aiter-allreduce-fusion \
     --cuda-graph-max-bs $CONC \
     --disable-radix-cache \
     --max-prefill-tokens $MAX_PREFILL_TOKENS \
     --scheduler-recv-interval 30 \
-    --mem-fraction-static 0.75 $EVAL_CONTEXT_ARGS > $SERVER_LOG 2>&1 &
+    --mem-fraction-static 0.8 $EVAL_CONTEXT_ARGS > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
 

perf-changelog.yaml

@@ -1358,3 +1358,11 @@
   description:
     - "Enable SGLANG_ENABLE_SPEC_V2=1 for Qwen3.5 FP8 H200 SGLang MTP"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1017
+
+- config-keys:
+    - qwen3.5-fp8-mi355x-sglang
+    - qwen3.5-bf16-mi355x-sglang
+  description:
+    - "Update cli args of Qwen3.5 FP8 and BF16 SGLang benchmarks for MI355X to achieve better performance"
+    - "Use lmsysorg/sglang-rocm:v0.5.10rc0-rocm720-mi35x-20260415"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1036