Add B200 config: dsv4-fp4-vllm (DeepSeek-V4-Pro) 🐋 🐋

.github/configs/nvidia-master.yaml

@@ -2409,6 +2409,27 @@ dsr1-fp8-b200-trt-mtp:
     # TP8 for all points
     - { tp: 8, ep: 1, conc-start: 4, conc-end: 256, spec-decoding: mtp }
 
+# DeepSeek-V4-Pro recipe from https://vllm.ai/blog/deepseek-v4
+# Runs on 8xB200 with DP=8 + expert parallelism (TP=1 per replica),
+# FP8 KV cache, block size 256, and FP4 indexer cache.
+dsv4-fp4-b200-vllm:
+  image: vllm/vllm-openai:deepseekv4-cu130
+  model: deepseek-ai/DeepSeek-V4-Pro
+  model-prefix: dsv4
+  runner: b200
+  precision: fp4
+  framework: vllm
+  multinode: false
+  seq-len-configs:
+  - isl: 1024
+    osl: 1024
+    search-space:
+    - { tp: 8, ep: 8, dp-attn: true, conc-start: 4, conc-end: 1024 }
+  - isl: 8192
+    osl: 1024
+    search-space:
+    - { tp: 8, ep: 8, dp-attn: true, conc-start: 4, conc-end: 512 }
+
 dsr1-fp8-h200-sglang:
   image: lmsysorg/sglang:v0.5.9-cu130
   model: deepseek-ai/DeepSeek-R1-0528

benchmarks/single_node/dsv4_fp4_b200_vllm.sh

@@ -0,0 +1,82 @@
+#!/usr/bin/env bash
+
+source "$(dirname "$0")/../benchmark_lib.sh"
+
+check_env_vars \
+    MODEL \
+    TP \
+    CONC \
+    ISL \
+    OSL \
+    MAX_MODEL_LEN \
+    RANDOM_RANGE_RATIO \
+    RESULT_FILENAME
+
+if [[ -n "$SLURM_JOB_ID" ]]; then
+  echo "JOB $SLURM_JOB_ID running on $SLURMD_NODENAME"
+fi
+
+nvidia-smi
+
+hf download "$MODEL"
+
+SERVER_LOG=/workspace/server.log
+PORT=${PORT:-8888}
+
+if [ "${EVAL_ONLY}" = "true" ]; then
+    setup_eval_context
+    MAX_MODEL_LEN="$EVAL_MAX_MODEL_LEN"
+fi
+
+# Start GPU monitoring (power, temperature, clocks every second)
+start_gpu_monitor
+
+# Per https://vllm.ai/blog/deepseek-v4 the DeepSeek-V4-Pro recipe on 8xB200
+# runs with EP + DP=8 (no --tensor-parallel-size flag). TP from the search
+# space is used only for GPU allocation by the runner and as the DP size.
+set -x
+VLLM_ENGINE_READY_TIMEOUT_S=3600 \
+vllm serve $MODEL --host 0.0.0.0 --port $PORT \
+--trust-remote-code \
+--kv-cache-dtype fp8 \
+--block-size 256 \
+--no-enable-prefix-caching \
+--enable-expert-parallel \
+--data-parallel-size $TP \
+--max-model-len $MAX_MODEL_LEN \
+--compilation-config '{"cudagraph_mode":"FULL_AND_PIECEWISE","custom_ops":["all"]}' \
+--attention_config.use_fp4_indexer_cache=True \
+--tokenizer-mode deepseek_v4 \
+--tool-call-parser deepseek_v4 \
+--enable-auto-tool-choice \
+--reasoning-parser deepseek_v4 > $SERVER_LOG 2>&1 &
+
+SERVER_PID=$!
+
+# 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 "$RESULT_FILENAME" \
+    --result-dir /workspace/ \
+    --trust-remote-code
+
+# After throughput, run evaluation only if RUN_EVAL is true
+if [ "${RUN_EVAL}" = "true" ]; then
+    run_eval --framework lm-eval --port "$PORT"
+    append_lm_eval_summary
+fi
+
+# Stop GPU monitoring
+stop_gpu_monitor
+set +x

perf-changelog.yaml

@@ -1755,7 +1755,7 @@
     - "VLLM_ENGINE_READY_TIMEOUT_S=3600 to accommodate large weight loading"
     - "Configs: 1k1k conc 4-64, 8k1k conc 4-64"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1130
-
+  
 - config-keys:
     - dsv4-fp4-b300-sglang
   description:
@@ -1812,3 +1812,17 @@
     - "Topologies: low-conc 1p1d-dep8-tep8 (4 nodes, mirrored from NVIDIA srt-slurm PR #71 with offload kept and numa-bind dropped); mid 1p1d-dep8-dep16 (6 nodes) and high 3p1d-dep8-dep16 (10 nodes) hand-rolled, structurally derived from the kimi-k2.5 1k/1k pattern"
     - "Recipes stored under benchmarks/multi_node/srt-slurm-recipes/ and overlaid onto the upstream srt-slurm checkout at runtime"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1129
+
+- config-keys:
+    - dsv4-fp4-b200-vllm
+  description:
+    - "Add DeepSeek-V4-Pro vLLM B200 benchmark per https://vllm.ai/blog/deepseek-v4"
+    - "Image: vllm/vllm-openai:deepseekv4-cu130"
+    - "Model: deepseek-ai/DeepSeek-V4-Pro"
+    - "8xB200 recipe: DP=8 + expert parallelism (TP=1/replica), FP8 KV cache, block size 256, FP4 indexer cache"
+    - "Flags: --trust-remote-code, --kv-cache-dtype fp8, --block-size 256, --enable-expert-parallel,"
+    - "  --data-parallel-size 8, --compilation-config cudagraph_mode=FULL_AND_PIECEWISE custom_ops=all,"
+    - "  --attention_config.use_fp4_indexer_cache=True, --tokenizer-mode deepseek_v4,"
+    - "  --tool-call-parser deepseek_v4, --enable-auto-tool-choice, --reasoning-parser deepseek_v4"
+    - "Configs: 1k1k conc 4-1024, 8k1k conc 4-512"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/XXXX

runners/launch_b200-cw.sh

@@ -4,9 +4,19 @@ export HF_HUB_CACHE_MOUNT="/tmp/gharunner/hf-hub-cache"
 export PORT=8888
 
 MODEL_CODE="${EXP_NAME%%_*}"
-FRAMEWORK_SUFFIX=$([[ "$FRAMEWORK" == "trt" ]] && printf '_trt' || printf '')
 SPEC_SUFFIX=$([[ "$SPEC_DECODING" == "mtp" ]] && printf '_mtp' || printf '')
 
+# Prefer a framework-tagged script (e.g. dsv4_fp4_b200_vllm.sh) so models
+# with multiple inference engines can coexist; fall back to the historical
+# name without an engine suffix (`_trt` for trt, bare for everyone else)
+# for scripts that haven't been retagged yet.
+BENCH_BASE="benchmarks/single_node/${MODEL_CODE}_${PRECISION}_b200"
+BENCH_SCRIPT="${BENCH_BASE}_${FRAMEWORK}${SPEC_SUFFIX}.sh"
+if [[ ! -f "$BENCH_SCRIPT" ]]; then
+    LEGACY_FW_SUFFIX=$([[ "$FRAMEWORK" == "trt" ]] && printf '_trt' || printf '')
+    BENCH_SCRIPT="${BENCH_BASE}${LEGACY_FW_SUFFIX}${SPEC_SUFFIX}.sh"
+fi
+
 PARTITION="b200"
 SQUASH_FILE="/tmp/gharunner/squash/$(echo "$IMAGE" | sed 's/[\/:@#]/_/g').sqsh"
 LOCK_FILE="${SQUASH_FILE}.lock"
@@ -58,6 +68,6 @@ srun --jobid=$JOB_ID \
 --container-mount-home \
 --container-workdir=$CONTAINER_MOUNT_DIR \
 --no-container-entrypoint --export=ALL \
-bash benchmarks/single_node/${MODEL_CODE}_${PRECISION}_b200${FRAMEWORK_SUFFIX}${SPEC_SUFFIX}.sh
+bash "$BENCH_SCRIPT"
 
 scancel $JOB_ID

runners/launch_b200-dgxc-slurm.sh

@@ -251,8 +251,17 @@ else
 
     HF_HUB_CACHE_MOUNT="/scratch/fsw/gharunners/hf-hub-cache"
     SQUASH_FILE="/home/sa-shared/containers/$(echo "$IMAGE" | sed 's/[\/:@#]/_/g').sqsh"
-    FRAMEWORK_SUFFIX=$([[ "$FRAMEWORK" == "trt" ]] && printf '_trt' || printf '')
     SPEC_SUFFIX=$([[ "$SPEC_DECODING" == "mtp" ]] && printf '_mtp' || printf '')
+    # Prefer a framework-tagged script (e.g. dsv4_fp4_b200_vllm.sh) so models
+    # with multiple inference engines can coexist; fall back to the historical
+    # name without an engine suffix (`_trt` for trt, bare for everyone else)
+    # for scripts that haven't been retagged yet.
+    BENCH_BASE="benchmarks/single_node/${EXP_NAME%%_*}_${PRECISION}_b200"
+    BENCH_SCRIPT="${BENCH_BASE}_${FRAMEWORK}${SPEC_SUFFIX}.sh"
+    if [[ ! -f "$BENCH_SCRIPT" ]]; then
+        LEGACY_FW_SUFFIX=$([[ "$FRAMEWORK" == "trt" ]] && printf '_trt' || printf '')
+        BENCH_SCRIPT="${BENCH_BASE}${LEGACY_FW_SUFFIX}${SPEC_SUFFIX}.sh"
+    fi
     LOCK_FILE="${SQUASH_FILE}.lock"
 
     # TODO(Cam): lmsysorg/sglang:deepseek-v4-blackwell installs sglang editable at
@@ -290,5 +299,5 @@ else
         --no-container-mount-home \
         --container-workdir=$CONTAINER_MOUNT_DIR \
         --no-container-entrypoint --export=ALL,PORT=8888 \
-        bash benchmarks/single_node/${EXP_NAME%%_*}_${PRECISION}_b200${FRAMEWORK_SUFFIX}${SPEC_SUFFIX}.sh
+        bash "$BENCH_SCRIPT"
 fi

runners/launch_b200-nb.sh

@@ -2,9 +2,19 @@
 
 HF_HUB_CACHE_MOUNT="/mnt/data/gharunners/hf-hub-cache/"
 PARTITION="main"
-FRAMEWORK_SUFFIX=$([[ "$FRAMEWORK" == "trt" ]] && printf '_trt' || printf '')
 SPEC_SUFFIX=$([[ "$SPEC_DECODING" == "mtp" ]] && printf '_mtp' || printf '')
 
+# Prefer a framework-tagged script (e.g. dsv4_fp4_b200_vllm.sh) so models
+# with multiple inference engines can coexist; fall back to the historical
+# name without an engine suffix (`_trt` for trt, bare for everyone else)
+# for scripts that haven't been retagged yet.
+BENCH_BASE="benchmarks/single_node/${EXP_NAME%%_*}_${PRECISION}_b200"
+BENCH_SCRIPT="${BENCH_BASE}_${FRAMEWORK}${SPEC_SUFFIX}.sh"
+if [[ ! -f "$BENCH_SCRIPT" ]]; then
+    LEGACY_FW_SUFFIX=$([[ "$FRAMEWORK" == "trt" ]] && printf '_trt' || printf '')
+    BENCH_SCRIPT="${BENCH_BASE}${LEGACY_FW_SUFFIX}${SPEC_SUFFIX}.sh"
+fi
+
 UCX_NET_DEVICES=eth0
 
 # TODO(Cam): lmsysorg/sglang:deepseek-v4-blackwell installs sglang editable at
@@ -27,4 +37,4 @@ srun --partition=$PARTITION --gres=gpu:$TP --exclusive --job-name="$RUNNER_NAME"
 --container-writable \
 --container-workdir=$CONTAINER_MOUNT_DIR \
 --no-container-entrypoint --export=ALL,PORT=8888,UCX_NET_DEVICES=$UCX_NET_DEVICES \
-bash benchmarks/single_node/${EXP_NAME%%_*}_${PRECISION}_b200${FRAMEWORK_SUFFIX}${SPEC_SUFFIX}.sh
+bash "$BENCH_SCRIPT"