[AMD][MI35X] Bump Qwen3.5 MXFP4 MI355X SGLang single-node image to v0.5.18

benchmarks/single_node/fixed_seq_len/qwen3.5_fp4_mi355x.sh

@@ -43,6 +43,7 @@ python3 -m sglang.launch_server --model-path=$MODEL --trust-remote-code \
 --disable-radix-cache \
 --enable-aiter-allreduce-fusion --max-running-requests $CONC \
 --page-size 16 \
+--kv-cache-dtype fp8_e4m3 \
 > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!

benchmarks/single_node/fixed_seq_len/qwen3.5_fp4_mi355x_mtp.sh

@@ -47,6 +47,7 @@ python3 -m sglang.launch_server --model-path=$MODEL --trust-remote-code \
 --speculative-num-steps 3 \
 --speculative-eagle-topk 1 \
 --speculative-num-draft-tokens 4 \
+--kv-cache-dtype fp8_e4m3 \
 > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!

configs/amd-master.yaml

@@ -313,7 +313,7 @@ qwen3.5-fp8-mi355x-sglang-disagg:
           - "DECODE_MTP_SIZE=0"
 
 qwen3.5-fp4-mi355x-sglang:
-  image: lmsysorg/sglang-rocm:v0.5.16-rocm720-mi35x-20260730
+  image: lmsysorg/sglang-rocm:v0.5.18-rocm720-mi35x-20260825
   model: amd/Qwen3.5-397B-A17B-MXFP4
   model-prefix: qwen3.5
   runner: mi355x
@@ -345,7 +345,7 @@ qwen3.5-fp4-mi355x-atom:
       - { tp: 4, conc-start: 4, conc-end: 16 }
 
 qwen3.5-fp4-mi355x-sglang-mtp:
-  image: lmsysorg/sglang-rocm:v0.5.16-rocm720-mi35x-20260730
+  image: lmsysorg/sglang-rocm:v0.5.18-rocm720-mi35x-20260825
   model: amd/Qwen3.5-397B-A17B-MXFP4
   model-prefix: qwen3.5
   runner: mi355x

perf-changelog.yaml

@@ -6427,7 +6427,6 @@
     - "Raise dram-utilization from 0.80 to 0.95, matching the vLLM agentic lane on the same runner, and the TP>=8 hicache ratio from 2 to 3 so the CPU tier reflects the allocated budget. DSv4 HiCache takes a host/device token ratio, not bytes, so the recipe cannot apply TOTAL_CPU_DRAM_GB directly, and host capacity scales with both the ratio and device KV -- meaning it also grows with mem-fraction-static. Measured at ratio 2 / mem-fraction 0.835 the tier allocates 999 GB (124.88 GB per rank across 8 ranks), a third of the 2,849 GB the vLLM agentic lane consumes. ratio 4 at mem-fraction 0.93 overshot the other way, leaving 5.84 GB free on a 2,964 GB node and failing to allocate the V4 paged pool; ratio 3 keeps the tier near 2 TB with room for the paged pool, page cache, AIPerf and the router. The TP<8 ratio is left at its existing value of 8."
     - "Costs TTFT (p50 2.2x, p90 3.3x); AgentX is closed-loop so this is already priced into the throughput figures"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2701
-
 
 - config-keys:
     - dsv4-fp4-gb200-dynamo-sglang
@@ -6466,4 +6465,13 @@
     - "Warm the aggregate low-latency profiles with the standard ten requests per trajectory lane instead of one, so they enter the profiling window from the same live state as every other AgentX recipe."
     - "Reduce the concurrency-1536 disaggregated topology from six to five DEP8 prefill workers while retaining one DEP16 decode worker."
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2644
-
+
+- config-keys:
+    - qwen3.5-fp4-mi355x-sglang
+    - qwen3.5-fp4-mi355x-sglang-mtp
+  scenario-type:
+    - fixed-seq-len
+  description:
+    - "Bump image from lmsysorg/sglang-rocm:v0.5.16-rocm720-mi35x-20260730 to lmsysorg/sglang-rocm:v0.5.18-rocm720-mi35x-20260825"
+    - "Serve both arms with an fp8_e4m3 KV cache."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2640