Add Qwen3.5-397B FP8 B200 SGLang benchmark config (#804)

Add Qwen3.5-397B-A17B-FP8 SGLang B200 benchmark (STP)

---------

Co-authored-by: claude[bot] <41898282+claude[bot]@users.noreply.github.com>
Co-authored-by: Kedar Potdar <kedarpotdar-nv@users.noreply.github.com>
Co-authored-by: Ankur Singh <ankursingh-nv@users.noreply.github.com>

Author: 4 people

.github/configs/nvidia-master.yaml

@@ -1804,6 +1804,31 @@ qwen3.5-bf16-b200-sglang:
     search-space:
     - { tp: 8, ep: 1, conc-start: 4, conc-end: 64 }
 
+qwen3.5-fp8-b200-sglang:
+  image: lmsysorg/sglang:v0.5.9-cu129-amd64
+  model: Qwen/Qwen3.5-397B-A17B-FP8
+  model-prefix: qwen3.5
+  runner: b200
+  precision: fp8
+  framework: sglang
+  multinode: false
+  seq-len-configs:
+  - isl: 1024
+    osl: 1024
+    search-space:
+    - { tp: 4, ep: 4, conc-start: 4, conc-end: 16 }
+    - { tp: 4, ep: 4, conc-start: 64, conc-end: 64 }
+  - isl: 1024
+    osl: 8192
+    search-space:
+    - { tp: 8, ep: 1, conc-start: 4, conc-end: 8 }
+    - { tp: 4, ep: 4, conc-start: 8, conc-end: 64}
+  - isl: 8192
+    osl: 1024
+    search-space:
+    - { tp: 8, ep: 1, conc-start: 4, conc-end: 4}
+    - { tp: 4, ep: 4, conc-start: 8, conc-end: 64 }
+
 kimik2.5-int4-b200-vllm:
   image: vllm/vllm-openai:v0.15.1
   model: moonshotai/Kimi-K2.5

benchmarks/single_node/qwen3.5_fp8_b200.sh

@@ -0,0 +1,84 @@
+#!/usr/bin/env bash
+
+source "$(dirname "$0")/../benchmark_lib.sh"
+
+check_env_vars \
+    MODEL \
+    TP \
+    CONC \
+    ISL \
+    OSL \
+    RANDOM_RANGE_RATIO \
+    RESULT_FILENAME \
+    EP_SIZE
+
+if [[ -n "$SLURM_JOB_ID" ]]; then
+  echo "JOB $SLURM_JOB_ID running on $SLURMD_NODENAME"
+fi
+
+nvidia-smi
+
+hf download "$MODEL"
+
+export NCCL_NVLS_ENABLE=1
+export SGL_ENABLE_JIT_DEEPGEMM=false
+export SGLANG_ENABLE_FLASHINFER_GEMM=true
+export PYTHONUNBUFFERED=1
+
+SERVER_LOG=/workspace/server.log
+PORT=${PORT:-8888}
+
+# Default: recv every ~10 requests; if CONC ≥ 16, relax to ~30 requests between scheduler recv polls.
+if [[ $CONC -ge 16 ]]; then
+  SCHEDULER_RECV_INTERVAL=30
+else
+  SCHEDULER_RECV_INTERVAL=10
+fi
+
+MEM_FRAC_STATIC=0.82
+CHUNKED_PREFILL_SIZE=32768
+MAX_PREFILL_TOKENS=32768
+CUDA_GRAPH_MAX_BATCH_SIZE=$CONC
+MAX_RUNNING_REQUESTS=128
+CONTEXT_LENGTH=$((ISL + OSL + 20))
+
+echo "SCHEDULER_RECV_INTERVAL: $SCHEDULER_RECV_INTERVAL, CONC: $CONC, ISL: $ISL, OSL: $OSL"
+
+set -x
+PYTHONNOUSERSITE=1 python3 -m sglang.launch_server --model-path=$MODEL --host=0.0.0.0 --port=$PORT \
+--trust-remote-code \
+--tensor-parallel-size=$TP --data-parallel-size=1 --ep-size $EP_SIZE \
+--quantization fp8 --kv-cache-dtype fp8_e4m3 \
+--mamba-ssm-dtype bfloat16 \
+--cuda-graph-max-bs $CUDA_GRAPH_MAX_BATCH_SIZE --max-running-requests $MAX_RUNNING_REQUESTS \
+--mem-fraction-static $MEM_FRAC_STATIC --chunked-prefill-size $CHUNKED_PREFILL_SIZE --max-prefill-tokens $MAX_PREFILL_TOKENS \
+--context-length $CONTEXT_LENGTH --disable-radix-cache \
+--attention-backend trtllm_mha --moe-runner-backend flashinfer_trtllm \
+--scheduler-recv-interval $SCHEDULER_RECV_INTERVAL \
+--tokenizer-worker-num 6 --stream-interval 30 > $SERVER_LOG 2>&1 &
+
+SERVER_PID=$!
+
+# Wait for server to be ready
+wait_for_server_ready --port "$PORT" --server-log "$SERVER_LOG" --server-pid "$SERVER_PID"
+
+pip install -q datasets pandas
+
+run_benchmark_serving \
+    --model "$MODEL" \
+    --port "$PORT" \
+    --backend vllm \
+    --input-len "$ISL" \
+    --output-len "$OSL" \
+    --random-range-ratio "$RANDOM_RANGE_RATIO" \
+    --num-prompts "$((CONC * 10))" \
+    --max-concurrency "$CONC" \
+    --result-filename "$RESULT_FILENAME" \
+    --result-dir /workspace/
+
+# After throughput, run evaluation only if RUN_EVAL is true
+if [ "${RUN_EVAL}" = "true" ]; then
+    run_eval --framework lm-eval --port "$PORT" --concurrent-requests $CONC
+    append_lm_eval_summary
+fi
+set +x

perf-changelog.yaml

@@ -856,3 +856,12 @@
     - "Switch from TP=8/EP=8 to TP=4/EP=4, concurrency 4-64 for 1k1k, 1k8k, and 8k1k"
     - "Script uses conditional --enable-expert-parallel based on EP_SIZE env var"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/832
+
+- config-keys:
+    - qwen3.5-fp8-b200-sglang
+  description:
+    - "Add Qwen3.5-397B-A17B-FP8 SGLang benchmark configuration for B200"
+    - "Image: lmsysorg/sglang:v0.5.9-cu129-amd64"
+    - "Uses trtllm_mha attention backend and flashinfer_trtllm MOE runner"
+    - "Enable SGLANG_ENABLE_FLASHINFER_GEMM=true, NCCL_NVLS_ENABLE=1"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/804