update 8k and 32k

experiments/h20/dsv4_h20_sglang_tp_dp_matrix/matrix.json

@@ -1,7 +1,6 @@
 {
-  "comment": "ISL/OSL matrix for dsv4_h20_sglang_tp_dp_matrix. Y=must test, P=optional (record N on failure), N=skip.",
-  "mode": "Y",
   "comment": "Only mandatory (Y) combinations are tested; 1M ISL is excluded per user request.",
+  "mode": "Y",
   "matrix": {
     "1024": {
       "128": "Y",
@@ -19,6 +18,14 @@
       "2048": "Y",
       "4096": "Y"
     },
+    "8192": {
+      "128": "Y",
+      "256": "Y",
+      "512": "Y",
+      "1024": "Y",
+      "2048": "Y",
+      "4096": "Y"
+    },
     "16384": {
       "128": "Y",
       "256": "Y",
@@ -27,6 +34,14 @@
       "2048": "Y",
       "4096": "P"
     },
+    "32768": {
+      "128": "Y",
+      "256": "Y",
+      "512": "Y",
+      "1024": "Y",
+      "2048": "Y",
+      "4096": "P"
+    },
     "65536": {
       "128": "Y",
       "256": "Y",
@@ -69,13 +84,45 @@
     }
   },
   "concurrency": {
-    "1024": { "low": 1, "high": 128 },
+    "1024": {
-    "4096": { "low": 1, "high": 64 },
+      "low": 1,
-    "16384": { "low": 1, "high": 32 },
+      "high": 128
-    "65536": { "low": 1, "high": 8 },
+    },
-    "131072": { "low": 1, "high": 4 },
+    "4096": {
-    "262144": { "low": 1, "high": 2 },
+      "low": 1,
-    "524288": { "low": 1, "high": 2 },
+      "high": 64
-    "1048576": { "low": 1, "high": 2 }
+    },
+    "8192": {
+      "low": 1,
+      "high": 64
+    },
+    "16384": {
+      "low": 1,
+      "high": 32
+    },
+    "32768": {
+      "low": 1,
+      "high": 16
+    },
+    "65536": {
+      "low": 1,
+      "high": 8
+    },
+    "131072": {
+      "low": 1,
+      "high": 4
+    },
+    "262144": {
+      "low": 1,
+      "high": 2
+    },
+    "524288": {
+      "low": 1,
+      "high": 2
+    },
+    "1048576": {
+      "low": 1,
+      "high": 2
+    }
   }
 }

experiments/h20/dsv4_h20_vllm_tp_dp_matrix/matrix.json

@@ -1,7 +1,6 @@
 {
-  "comment": "ISL/OSL matrix for dsv4_h20_vllm_tp_dp_matrix. Y=must test, P=optional (record N on failure), N=skip.",
-  "mode": "Y",
   "comment": "Only mandatory (Y) combinations are tested; 1M ISL is excluded per user request.",
+  "mode": "Y",
   "matrix": {
     "1024": {
       "128": "Y",
@@ -19,6 +18,14 @@
       "2048": "Y",
       "4096": "Y"
     },
+    "8192": {
+      "128": "Y",
+      "256": "Y",
+      "512": "Y",
+      "1024": "Y",
+      "2048": "Y",
+      "4096": "Y"
+    },
     "16384": {
       "128": "Y",
       "256": "Y",
@@ -27,6 +34,14 @@
       "2048": "Y",
       "4096": "P"
     },
+    "32768": {
+      "128": "Y",
+      "256": "Y",
+      "512": "Y",
+      "1024": "Y",
+      "2048": "Y",
+      "4096": "P"
+    },
     "65536": {
       "128": "Y",
       "256": "Y",
@@ -69,13 +84,45 @@
     }
   },
   "concurrency": {
-    "1024": { "low": 1, "high": 128 },
+    "1024": {
-    "4096": { "low": 1, "high": 64 },
+      "low": 1,
-    "16384": { "low": 1, "high": 32 },
+      "high": 128
-    "65536": { "low": 1, "high": 8 },
+    },
-    "131072": { "low": 1, "high": 4 },
+    "4096": {
-    "262144": { "low": 1, "high": 2 },
+      "low": 1,
-    "524288": { "low": 1, "high": 2 },
+      "high": 64
-    "1048576": { "low": 1, "high": 2 }
+    },
+    "8192": {
+      "low": 1,
+      "high": 64
+    },
+    "16384": {
+      "low": 1,
+      "high": 32
+    },
+    "32768": {
+      "low": 1,
+      "high": 16
+    },
+    "65536": {
+      "low": 1,
+      "high": 8
+    },
+    "131072": {
+      "low": 1,
+      "high": 4
+    },
+    "262144": {
+      "low": 1,
+      "high": 2
+    },
+    "524288": {
+      "low": 1,
+      "high": 2
+    },
+    "1048576": {
+      "low": 1,
+      "high": 2
+    }
   }
 }

experiments/h200/dsv4_h200_sglang_tp_dp_matrix/matrix.json

@@ -1,7 +1,6 @@
 {
-  "comment": "ISL/DSL matrix for dsv4_h200_sglang_tp_dp_matrix. Y=must test, P=optional (record N on failure), N=skip.",
-  "mode": "Y",
   "comment": "Only mandatory (Y) combinations are tested; 1M ISL is excluded per user request.",
+  "mode": "Y",
   "matrix": {
     "1024": {
       "128": "Y",
@@ -19,6 +18,14 @@
       "2048": "Y",
       "4096": "Y"
     },
+    "8192": {
+      "128": "Y",
+      "256": "Y",
+      "512": "Y",
+      "1024": "Y",
+      "2048": "Y",
+      "4096": "Y"
+    },
     "16384": {
       "128": "Y",
       "256": "Y",
@@ -27,6 +34,14 @@
       "2048": "Y",
       "4096": "P"
     },
+    "32768": {
+      "128": "Y",
+      "256": "Y",
+      "512": "Y",
+      "1024": "Y",
+      "2048": "Y",
+      "4096": "P"
+    },
     "65536": {
       "128": "Y",
       "256": "Y",
@@ -69,13 +84,45 @@
     }
   },
   "concurrency": {
-    "1024": { "low": 1, "high": 128 },
+    "1024": {
-    "4096": { "low": 1, "high": 64 },
+      "low": 1,
-    "16384": { "low": 1, "high": 32 },
+      "high": 128
-    "65536": { "low": 1, "high": 8 },
+    },
-    "131072": { "low": 1, "high": 4 },
+    "4096": {
-    "262144": { "low": 1, "high": 2 },
+      "low": 1,
-    "524288": { "low": 1, "high": 2 },
+      "high": 64
-    "1048576": { "low": 1, "high": 2 }
+    },
+    "8192": {
+      "low": 1,
+      "high": 64
+    },
+    "16384": {
+      "low": 1,
+      "high": 32
+    },
+    "32768": {
+      "low": 1,
+      "high": 16
+    },
+    "65536": {
+      "low": 1,
+      "high": 8
+    },
+    "131072": {
+      "low": 1,
+      "high": 4
+    },
+    "262144": {
+      "low": 1,
+      "high": 2
+    },
+    "524288": {
+      "low": 1,
+      "high": 2
+    },
+    "1048576": {
+      "low": 1,
+      "high": 2
+    }
   }
 }

experiments/h200/dsv4_h200_vllm_tp_dp_matrix/matrix.json

@@ -1,7 +1,6 @@
 {
-  "comment": "ISL/DSL matrix for dsv4_h200_vllm_tp_dp_matrix. Y=must test, P=optional (record N on failure), N=skip.",
-  "mode": "Y",
   "comment": "Only mandatory (Y) combinations are tested; 1M ISL is excluded per user request.",
+  "mode": "Y",
   "matrix": {
     "1024": {
       "128": "Y",
@@ -19,6 +18,14 @@
       "2048": "Y",
       "4096": "Y"
     },
+    "8192": {
+      "128": "Y",
+      "256": "Y",
+      "512": "Y",
+      "1024": "Y",
+      "2048": "Y",
+      "4096": "Y"
+    },
     "16384": {
       "128": "Y",
       "256": "Y",
@@ -27,6 +34,14 @@
       "2048": "Y",
       "4096": "P"
     },
+    "32768": {
+      "128": "Y",
+      "256": "Y",
+      "512": "Y",
+      "1024": "Y",
+      "2048": "Y",
+      "4096": "P"
+    },
     "65536": {
       "128": "Y",
       "256": "Y",
@@ -69,13 +84,45 @@
     }
   },
   "concurrency": {
-    "1024": { "low": 1, "high": 128 },
+    "1024": {
-    "4096": { "low": 1, "high": 64 },
+      "low": 1,
-    "16384": { "low": 1, "high": 32 },
+      "high": 128
-    "65536": { "low": 1, "high": 8 },
+    },
-    "131072": { "low": 1, "high": 4 },
+    "4096": {
-    "262144": { "low": 1, "high": 2 },
+      "low": 1,
-    "524288": { "low": 1, "high": 2 },
+      "high": 64
-    "1048576": { "low": 1, "high": 2 }
+    },
+    "8192": {
+      "low": 1,
+      "high": 64
+    },
+    "16384": {
+      "low": 1,
+      "high": 32
+    },
+    "32768": {
+      "low": 1,
+      "high": 16
+    },
+    "65536": {
+      "low": 1,
+      "high": 8
+    },
+    "131072": {
+      "low": 1,
+      "high": 4
+    },
+    "262144": {
+      "low": 1,
+      "high": 2
+    },
+    "524288": {
+      "low": 1,
+      "high": 2
+    },
+    "1048576": {
+      "low": 1,
+      "high": 2
+    }
   }
 }

experiments/p800/dsv4_p800_sglang_tp_dp_matrix/matrix.json

@@ -19,6 +19,14 @@
       "2048": "Y",
       "4096": "Y"
     },
+    "8192": {
+      "128": "Y",
+      "256": "Y",
+      "512": "Y",
+      "1024": "Y",
+      "2048": "Y",
+      "4096": "Y"
+    },
     "16384": {
       "128": "Y",
       "256": "Y",
@@ -27,6 +35,14 @@
       "2048": "Y",
       "4096": "P"
     },
+    "32768": {
+      "128": "Y",
+      "256": "Y",
+      "512": "Y",
+      "1024": "Y",
+      "2048": "Y",
+      "4096": "P"
+    },
     "65536": {
       "128": "Y",
       "256": "Y",
@@ -45,10 +61,33 @@
     }
   },
   "concurrency": {
-    "1024": { "low": 1, "high": 128 },
+    "1024": {
-    "4096": { "low": 1, "high": 64 },
+      "low": 1,
-    "16384": { "low": 1, "high": 32 },
+      "high": 128
-    "65536": { "low": 1, "high": 8 },
+    },
-    "131072": { "low": 1, "high": 4 }
+    "4096": {
+      "low": 1,
+      "high": 64
+    },
+    "8192": {
+      "low": 1,
+      "high": 64
+    },
+    "16384": {
+      "low": 1,
+      "high": 32
+    },
+    "32768": {
+      "low": 1,
+      "high": 16
+    },
+    "65536": {
+      "low": 1,
+      "high": 8
+    },
+    "131072": {
+      "low": 1,
+      "high": 4
+    }
   }
 }

experiments/pro6000/dsv4_pro6000_sglang_tp_dp_matrix/adaptive_config.env

@@ -15,11 +15,16 @@ NUM_PROMPTS_MULTIPLIER="${NUM_PROMPTS_MULTIPLIER:-5}"
 TPS_MIN_GAIN_PCT="${TPS_MIN_GAIN_PCT:-2.0}"
 PLATEAU_PATIENCE="${PLATEAU_PATIENCE:-2}"
 
+# Stop a shape when p95 TTFT exceeds the SLO; keep group skipping disabled.
+TTFT_SLO_MS="${TTFT_SLO_MS:-4000}"
+ENABLE_TTFT_SLO_STOP="${ENABLE_TTFT_SLO_STOP:-1}"
+TTFT_GROUP_SKIP_MS="${TTFT_GROUP_SKIP_MS:-0}"
+
 # Keep the same random workload semantics as the fixed matrix baseline.
 # DATASET_PATH must contain at least SEARCH_MAX_CONCURRENCY times
 # NUM_PROMPTS_MULTIPLIER valid two-turn conversations. Set this explicitly to
 # random-ids to use generated token IDs without a ShareGPT seed dataset.
-BENCH_DATASET_NAME="${BENCH_DATASET_NAME:-random}"
+BENCH_DATASET_NAME="${BENCH_DATASET_NAME:-random-ids}"
 # SGLang interprets 0.0 as Uniform[1, requested_len]. Use 1.0 for fixed
 # ISL/OSL points; lower values intentionally benchmark a length distribution.
 RANDOM_RANGE_RATIO="${RANDOM_RANGE_RATIO:-1.0}"

experiments/pro6000/dsv4_pro6000_sglang_tp_dp_matrix/compare.py

@@ -83,7 +83,7 @@ def main():
         f.write(f"- Model: `{model}`\n")
         f.write(f"- Hardware: {hardware}\n")
         f.write("- Backend: SGLang (Docker)\n")
-        f.write("- Benchmark client: `sglang.bench_serving`\n")
+        f.write("- Benchmark client: `sglang.benchmark.serving`\n")
         f.write(f"- SLO reference: TTFT P95 < {args.ttft_limit}ms, TPOT mean < {args.tpot_limit}ms\n\n")
 
         # Configuration overview.

experiments/pro6000/dsv4_pro6000_sglang_tp_dp_matrix/config.env

@@ -7,7 +7,7 @@
 
 EXPERIMENT="dsv4_pro6000_sglang_tp_dp_matrix"
 MODEL_NAME="DeepSeek-V4-Flash"
-MODEL_PATH="/root/data/models/ds_flash_v4"
+MODEL_PATH="/data/6000D/DeepSeek-V4-Flash"
 SERVED_MODEL_NAME="deepseek-v4-flash"
 
 SGLANG_PORT="${SGLANG_PORT:-30031}"
@@ -20,7 +20,7 @@ VENV_CLIENT="${VENV_CLIENT:-/root/.miniconda3/envs/sglang}"
 # Run the benchmark client natively (0) or inside Docker (1).
 USE_DOCKER_CLIENT="${USE_DOCKER_CLIENT:-1}"
 
-export CUDA_VISIBLE_DEVICES="${CUDA_VISIBLE_DEVICES:-0,1,3,4,5,6,7}"
+export CUDA_VISIBLE_DEVICES="${CUDA_VISIBLE_DEVICES:-0,1,2,3,4,5,6,7}"
 
 # Runtime working directory for logs, pid files, and tmp. Defaults to a local
 # directory under this experiment so the benchmark is self-contained.
@@ -41,12 +41,14 @@ MOE_RUNNER_BACKEND="${MOE_RUNNER_BACKEND:-marlin}"
 
 # Deployment switch. 0 = native sglang venv, 1 = Docker.
 USE_DOCKER="${USE_DOCKER:-1}"
-DOCKER_IMAGE="${DOCKER_IMAGE:-lmsysorg/sglang:latest}"
+DOCKER_IMAGE="${DOCKER_IMAGE:-sglang-sm120-dsv4:0.5.15.post1-fi0.6.14-sm120fix1}"
 
-# Dataset used by sglang.bench_serving --dataset-name random.
+# Offline by default: random-ids generates token IDs without downloading data.
-# The random sampler needs a ShareGPT-style JSON file locally; it falls back to
+# To use ShareGPT, set BENCH_DATASET_NAME=random and DATASET_PATH explicitly.
-# downloading from HuggingFace, which usually fails on offline nodes.
+BENCH_DATASET_NAME="${BENCH_DATASET_NAME:-random-ids}"
-DATASET_PATH="${DATASET_PATH:-/root/data/yy/sskj/datasets/ShareGPT_V3_unfiltered_cleaned_split.json}"
+DATASET_PATH="${DATASET_PATH:-}"
+SGLANG_BENCH_MODULE="${SGLANG_BENCH_MODULE:-sglang.benchmark.serving}"
+CACHE_DIR="${CACHE_DIR:-/data/yy/sskj/sglang_sm120_cache}"
 
 # Matrix and concurrency rules are defined in matrix.json by default.
 MATRIX_FILE="${MATRIX_FILE:-${SCRIPT_DIR:-.}/matrix.json}"

experiments/pro6000/dsv4_pro6000_sglang_tp_dp_matrix/matrix.json

@@ -1,7 +1,6 @@
 {
-  "comment": "ISL/DSL matrix for dsv4_h200_sglang_tp_dp_matrix. Y=must test, P=optional (record N on failure), N=skip.",
-  "mode": "Y",
   "comment": "Only mandatory (Y) combinations are tested; 1M ISL is excluded per user request.",
+  "mode": "Y",
   "matrix": {
     "1024": {
       "128": "Y",
@@ -19,6 +18,14 @@
       "2048": "Y",
       "4096": "Y"
     },
+    "8192": {
+      "128": "Y",
+      "256": "Y",
+      "512": "Y",
+      "1024": "Y",
+      "2048": "Y",
+      "4096": "Y"
+    },
     "16384": {
       "128": "Y",
       "256": "Y",
@@ -27,6 +34,14 @@
       "2048": "Y",
       "4096": "P"
     },
+    "32768": {
+      "128": "Y",
+      "256": "Y",
+      "512": "Y",
+      "1024": "Y",
+      "2048": "Y",
+      "4096": "P"
+    },
     "65536": {
       "128": "Y",
       "256": "Y",
@@ -69,13 +84,45 @@
     }
   },
   "concurrency": {
-    "1024": { "low": 1, "high": 128 },
+    "1024": {
-    "4096": { "low": 1, "high": 64 },
+      "low": 1,
-    "16384": { "low": 1, "high": 32 },
+      "high": 128
-    "65536": { "low": 1, "high": 8 },
+    },
-    "131072": { "low": 1, "high": 4 },
+    "4096": {
-    "262144": { "low": 1, "high": 2 },
+      "low": 1,
-    "524288": { "low": 1, "high": 2 },
+      "high": 64
-    "1048576": { "low": 1, "high": 2 }
+    },
+    "8192": {
+      "low": 1,
+      "high": 64
+    },
+    "16384": {
+      "low": 1,
+      "high": 32
+    },
+    "32768": {
+      "low": 1,
+      "high": 16
+    },
+    "65536": {
+      "low": 1,
+      "high": 8
+    },
+    "131072": {
+      "low": 1,
+      "high": 4
+    },
+    "262144": {
+      "low": 1,
+      "high": 2
+    },
+    "524288": {
+      "low": 1,
+      "high": 2
+    },
+    "1048576": {
+      "low": 1,
+      "high": 2
+    }
   }
 }

experiments/pro6000/dsv4_pro6000_sglang_tp_dp_matrix/run_adaptive_concurrency.sh

@@ -27,7 +27,7 @@ else
   PYTHON="$(command -v python3)"
 fi
 
-DOCKER_IMAGE="${DOCKER_IMAGE:-lmsysorg/sglang:latest}"
+DOCKER_IMAGE="${DOCKER_IMAGE:-sglang-sm120-dsv4:0.5.15.post1-fi0.6.14-sm120fix1}"
 
 engine_is_healthy() {
   curl --fail --silent --show-error --max-time 5 \
@@ -68,7 +68,7 @@ engine_build_server_args() {
   local tp="$1"
   local dp="$2"
   local -a args=(
-    sglang serve --model-path "$MODEL_PATH"
+    python3 -m sglang.launch_server --model-path "$MODEL_PATH"
     --trust-remote-code
     --tp-size "$tp"
     --moe-runner-backend "$MOE_RUNNER_BACKEND"
@@ -167,15 +167,16 @@ engine_run_bench() {
       --network host \
       "${volume_args[@]}" \
       -e PYTHONUNBUFFERED=1 \
+      --entrypoint python3 \
       "$DOCKER_IMAGE" \
-      python -m sglang.bench_serving "${bench_args[@]}"
+      -m "$SGLANG_BENCH_MODULE" "${bench_args[@]}"
   else
-    "$PYTHON" -m sglang.bench_serving "${bench_args[@]}"
+    "$PYTHON" -m "$SGLANG_BENCH_MODULE" "${bench_args[@]}"
   fi
 }
 
 export -f engine_run_bench
 export ENGINE_PORT MODEL_PATH RESULT_BASE DOCKER_IMAGE USE_DOCKER_CLIENT
-export BENCH_DATASET_NAME DATASET_PATH RANDOM_RANGE_RATIO BENCH_WARMUP_MAX_REQUESTS PYTHON
+export BENCH_DATASET_NAME DATASET_PATH RANDOM_RANGE_RATIO BENCH_WARMUP_MAX_REQUESTS PYTHON SGLANG_BENCH_MODULE
 
 adaptive_main "$@"

experiments/pro6000/dsv4_pro6000_sglang_tp_dp_matrix/run_adaptive_concurrency_add16.sh

@@ -27,7 +27,7 @@ else
   PYTHON="$(command -v python3)"
 fi
 
-DOCKER_IMAGE="${DOCKER_IMAGE:-lmsysorg/sglang:latest}"
+DOCKER_IMAGE="${DOCKER_IMAGE:-sglang-sm120-dsv4:0.5.15.post1-fi0.6.14-sm120fix1}"
 
 engine_is_healthy() {
   curl --fail --silent --show-error --max-time 5 \
@@ -68,7 +68,7 @@ engine_build_server_args() {
   local tp="$1"
   local dp="$2"
   local -a args=(
-    sglang serve --model-path "$MODEL_PATH"
+    python3 -m sglang.launch_server --model-path "$MODEL_PATH"
     --trust-remote-code
     --tp-size "$tp"
     --moe-runner-backend "$MOE_RUNNER_BACKEND"
@@ -167,16 +167,17 @@ engine_run_bench() {
       --network host \
       "${volume_args[@]}" \
       -e PYTHONUNBUFFERED=1 \
+      --entrypoint python3 \
       "$DOCKER_IMAGE" \
-      python -m sglang.bench_serving "${bench_args[@]}"
+      -m "$SGLANG_BENCH_MODULE" "${bench_args[@]}"
   else
-    "$PYTHON" -m sglang.bench_serving "${bench_args[@]}"
+    "$PYTHON" -m "$SGLANG_BENCH_MODULE" "${bench_args[@]}"
   fi
 }
 
 export -f engine_run_bench
 export ENGINE_PORT MODEL_PATH RESULT_BASE DOCKER_IMAGE USE_DOCKER_CLIENT
-export BENCH_DATASET_NAME DATASET_PATH RANDOM_RANGE_RATIO BENCH_WARMUP_MAX_REQUESTS PYTHON
+export BENCH_DATASET_NAME DATASET_PATH RANDOM_RANGE_RATIO BENCH_WARMUP_MAX_REQUESTS PYTHON SGLANG_BENCH_MODULE
 
 export SEARCH_START_CONCURRENCY=16
 export SEARCH_ADDEND=16

experiments/pro6000/dsv4_pro6000_sglang_tp_dp_matrix/run_bench.sh

@@ -22,7 +22,7 @@ DRY_RUN="${DRY_RUN:-0}"
 GRID_LIMIT="${GRID_LIMIT:-0}"
 
 # Export variables used inside functions that are called via bash -c subshells.
-export DATASET_PATH MODEL_PATH RESULT_BASE DOCKER_IMAGE USE_DOCKER_CLIENT
+export BENCH_DATASET_NAME DATASET_PATH MODEL_PATH RESULT_BASE DOCKER_IMAGE USE_DOCKER_CLIENT SGLANG_BENCH_MODULE
 
 if [[ -x "${VENV_CLIENT}/bin/python" ]]; then
   PYTHON="${VENV_CLIENT}/bin/python"
@@ -30,7 +30,7 @@ else
   PYTHON="$(command -v python3)"
 fi
 
-DOCKER_IMAGE="${DOCKER_IMAGE:-lmsysorg/sglang:latest}"
+DOCKER_IMAGE="${DOCKER_IMAGE:-sglang-sm120-dsv4:0.5.15.post1-fi0.6.14-sm120fix1}"
 
 log_dir_global="${RESULT_BASE}/${RUN_ID}/logs"
 mkdir -p "$log_dir_global"
@@ -79,9 +79,9 @@ stop_server() {
     docker rm -f "${EXPERIMENT}_sglang_tp${tp}_dp${dp}" >/dev/null 2>&1 || true
   fi
 
-  # Fallback: kill any sglang serve processes for this model.
+  # Fallback: kill any SGLang launch processes for this model.
-  pkill -9 -f "sglang serve.*${MODEL_NAME}" 2>/dev/null || true
+  pkill -9 -f "sglang.launch_server.*${MODEL_NAME}" 2>/dev/null || true
-  pkill -9 -f "sglang serve.*${MODEL_PATH}" 2>/dev/null || true
+  pkill -9 -f "sglang.launch_server.*${MODEL_PATH}" 2>/dev/null || true
   sleep 2
 }
 
@@ -90,7 +90,7 @@ build_server_args() {
   local dp="$2"
 
   local args=(
-    "sglang serve" "$MODEL_PATH"
+    "python3 -m sglang.launch_server" --model-path "$MODEL_PATH"
     --trust-remote-code
     --tp-size "$tp"
     --moe-runner-backend "$MOE_RUNNER_BACKEND"
@@ -138,11 +138,17 @@ restart_server() {
 }
 
 run_bench_serving() {
-  # Run sglang.bench_serving either natively or inside the SGLang Docker image.
+  # Inject the offline workload choice consistently for warmup and main runs.
+  local -a dataset_args=(--dataset-name "$BENCH_DATASET_NAME")
+  if [[ "$BENCH_DATASET_NAME" == "random" ]]; then
+    dataset_args+=(--dataset-path "$DATASET_PATH")
+  else
+    dataset_args+=(--tokenize-prompt)
+  fi
   if [[ "${USE_DOCKER_CLIENT:-1}" == "1" ]]; then
     local vol_args=()
     vol_args+=("-v" "${MODEL_PATH}:${MODEL_PATH}:ro")
-    if [[ -n "${DATASET_PATH:-}" ]]; then
+    if [[ "$BENCH_DATASET_NAME" == "random" ]]; then
       vol_args+=("-v" "${DATASET_PATH}:${DATASET_PATH}:ro")
     fi
     vol_args+=("-v" "${RESULT_BASE}:${RESULT_BASE}")
@@ -150,10 +156,11 @@ run_bench_serving() {
       --network host \
       "${vol_args[@]}" \
       -e PYTHONUNBUFFERED=1 \
+      --entrypoint python3 \
       "${DOCKER_IMAGE}" \
-      python -m sglang.bench_serving "$@"
+      -m "$SGLANG_BENCH_MODULE" "${dataset_args[@]}" "$@"
   else
-    "$PYTHON" -m sglang.bench_serving "$@"
+    "$PYTHON" -m "$SGLANG_BENCH_MODULE" "${dataset_args[@]}" "$@"
   fi
 }
 export -f run_bench_serving
@@ -168,8 +175,6 @@ run_warmup() {
       --backend sglang \
       --host 127.0.0.1 \
       --port "'"$SGLANG_PORT"'" \
-      --dataset-name random \
-      --dataset-path "'"$DATASET_PATH"'" \
       --random-input-len "'"$input_len"'" \
       --random-output-len "'"$output_len"'" \
       --num-prompts 1 \
@@ -319,7 +324,7 @@ skip_remaining_scenarios() {
     fi
     append_scenario_record "$result_root" \
       "name=${sname}" \
-      "config=$(jq -n --arg phase main --argjson c "$conc" --argjson i "$isl" --argjson o "$dsl" --arg dataset random --argjson n "$num" '{phase: $phase, concurrency: $c, input_len: $i, output_len: $o, dataset: $dataset, num_prompts: $n}')" \
+      "config=$(jq -n --arg phase main --argjson c "$conc" --argjson i "$isl" --argjson o "$dsl" --arg dataset "$BENCH_DATASET_NAME" --argjson n "$num" '{phase: $phase, concurrency: $c, input_len: $i, output_len: $o, dataset: $dataset, num_prompts: $n}')" \
       "status=\"${status}\"" \
       "note=\"${reason}\""
     record_skipped_csv "$skipped_csv" \
@@ -447,8 +452,6 @@ run_parallel_config() {
         --backend sglang \
         --host 127.0.0.1 \
         --port "'"$SGLANG_PORT"'" \
-        --dataset-name random \
-        --dataset-path "'"$DATASET_PATH"'" \
         --random-input-len "'"$isl"'" \
         --random-output-len "'"$dsl"'" \
         --num-prompts "'"$num"'" \
@@ -465,7 +468,7 @@ run_parallel_config() {
       log "finished ${config_label} scenario: output=${output_file}"
       append_scenario_record "$result_root" \
         "name=${sname}" \
-        "config=$(jq -n --arg phase main --argjson c "$conc" --argjson i "$isl" --argjson o "$dsl" --arg dataset random --argjson n "$num" '{phase: $phase, concurrency: $c, input_len: $i, output_len: $o, dataset: $dataset, num_prompts: $n}')" \
+        "config=$(jq -n --arg phase main --argjson c "$conc" --argjson i "$isl" --argjson o "$dsl" --arg dataset "$BENCH_DATASET_NAME" --argjson n "$num" '{phase: $phase, concurrency: $c, input_len: $i, output_len: $o, dataset: $dataset, num_prompts: $n}')" \
         "status=\"completed\"" \
         "note=\"benchmark finished successfully\""
       continue
@@ -476,7 +479,7 @@ run_parallel_config() {
       log "ERROR: ${config_label} scenario ${sname} triggered OOM; stopping config"
       append_scenario_record "$result_root" \
         "name=${sname}" \
-        "config=$(jq -n --arg phase main --argjson c "$conc" --argjson i "$isl" --argjson o "$dsl" --arg dataset random --argjson n "$num" '{phase: $phase, concurrency: $c, input_len: $i, output_len: $o, dataset: $dataset, num_prompts: $n}')" \
+        "config=$(jq -n --arg phase main --argjson c "$conc" --argjson i "$isl" --argjson o "$dsl" --arg dataset "$BENCH_DATASET_NAME" --argjson n "$num" '{phase: $phase, concurrency: $c, input_len: $i, output_len: $o, dataset: $dataset, num_prompts: $n}')" \
         "status=\"OOM\"" \
         "note=\"detected CUDA out-of-memory\""
       record_skipped_csv "$skipped_csv" \
@@ -491,7 +494,7 @@ run_parallel_config() {
       log "optional (P) scenario failed; recording as skipped and continuing"
       append_scenario_record "$result_root" \
         "name=${sname}" \
-        "config=$(jq -n --arg phase main --argjson c "$conc" --argjson i "$isl" --argjson o "$dsl" --arg dataset random --argjson n "$num" '{phase: $phase, concurrency: $c, input_len: $i, output_len: $o, dataset: $dataset, num_prompts: $n}')" \
+        "config=$(jq -n --arg phase main --argjson c "$conc" --argjson i "$isl" --argjson o "$dsl" --arg dataset "$BENCH_DATASET_NAME" --argjson n "$num" '{phase: $phase, concurrency: $c, input_len: $i, output_len: $o, dataset: $dataset, num_prompts: $n}')" \
         "status=\"skipped_optional\"" \
         "note=\"optional scenario failed (rc=${bench_rc})\""
       record_skipped_csv "$skipped_csv" \
@@ -509,7 +512,7 @@ run_parallel_config() {
     log "ERROR: ${config_label} server restart failed; skipping remaining scenarios"
     append_scenario_record "$result_root" \
       "name=${sname}" \
-      "config=$(jq -n --arg phase main --argjson c "$conc" --argjson i "$isl" --argjson o "$dsl" --arg dataset random --argjson n "$num" '{phase: $phase, concurrency: $c, input_len: $i, output_len: $o, dataset: $dataset, num_prompts: $n}')" \
+      "config=$(jq -n --arg phase main --argjson c "$conc" --argjson i "$isl" --argjson o "$dsl" --arg dataset "$BENCH_DATASET_NAME" --argjson n "$num" '{phase: $phase, concurrency: $c, input_len: $i, output_len: $o, dataset: $dataset, num_prompts: $n}')" \
       "status=\"FAILED\"" \
       "note=\"scenario failed and server restart failed (rc=${bench_rc})\""
     record_skipped_csv "$skipped_csv" \

experiments/pro6000/dsv4_pro6000_sglang_tp_dp_matrix/run_sglang_in_container.sh

@@ -22,7 +22,7 @@ rm -f "$PID_FILE"
 
 # Build server args.
 SERVER_ARGS=(
-  serve
+  -m sglang.launch_server
   --model-path "$MODEL_PATH"
   --trust-remote-code
   --tp-size "$TP"
@@ -44,24 +44,24 @@ SERVER_ARGS_STR="${SERVER_ARGS[*]}"
 
 echo "=== Starting SGLang server inside container (TP=${TP}, DP=${DP}) ==="
 echo "Container: $NAME"
-echo "Command: sglang ${SERVER_ARGS_STR}"
+echo "Command: python3 ${SERVER_ARGS_STR}"
 echo "Log: $LOG"
 
 # Kill any existing sglang process inside container first.
-docker exec "$NAME" pkill -9 -f "sglang serve" 2>/dev/null || true
+docker exec "$NAME" pkill -9 -f "sglang.launch_server" 2>/dev/null || true
 sleep 3
 
-# Start sglang serve inside the container in background.
+# Start the SGLang launch module inside the container in background.
 # We use nohup so it survives after docker exec returns.
 docker exec -d \
   -e CUDA_VISIBLE_DEVICES="${CUDA_VISIBLE_DEVICES}" \
   -e PYTHONUNBUFFERED=1 \
   "$NAME" \
-  bash -c "nohup sglang ${SERVER_ARGS_STR} > /tmp/sglang_server.log 2>&1 &"
+  bash -c "nohup python3 ${SERVER_ARGS_STR} > /tmp/sglang_server.log 2>&1 &"
 
 # Wait for the server process to appear inside container.
 sleep 2
-SERVER_PID=$(docker exec "$NAME" pgrep -f "sglang serve" | head -n 1 || true)
+SERVER_PID=$(docker exec "$NAME" pgrep -f "sglang.launch_server" | head -n 1 || true)
 if [[ -z "$SERVER_PID" ]]; then
   echo "ERROR: sglang server process not found inside container"
   docker exec "$NAME" cat /tmp/sglang_server.log 2>/dev/null | tail -50 || true

experiments/pro6000/dsv4_pro6000_sglang_tp_dp_matrix/start_sglang_container.sh

@@ -13,14 +13,13 @@ source "${SCRIPT_DIR}/config.env"
 RUNTIME_BASE="${RUNTIME_BASE:-${SCRIPT_DIR}/runtime}"
 mkdir -p "${RUNTIME_BASE}/logs" "${RUNTIME_BASE}/tmp"
 
-IMAGE="${DOCKER_IMAGE:-lmsysorg/sglang:latest}"
+IMAGE="${DOCKER_IMAGE:-sglang-sm120-dsv4:0.5.15.post1-fi0.6.14-sm120fix1}"
 PORT="${SGLANG_PORT:-30031}"
 NAME="${EXPERIMENT}_sglang_container"
 PID_FILE="${RUNTIME_BASE}/${EXPERIMENT}_container.pid"
 
-# Persistent cache directory on host for DeepGEMM JIT kernels.
+# Persistent cache directory for SM120 JIT kernels and tokenizer metadata.
-CACHE_DIR="${RUNTIME_BASE}/cache"
+mkdir -p "$CACHE_DIR"
-mkdir -p "${CACHE_DIR}/deep_gemm" "${CACHE_DIR}/tvm-ffi"
 
 # Clean up any stale container.
 docker rm -f "$NAME" >/dev/null 2>&1 || true
@@ -40,16 +39,19 @@ echo "Log: $LOG"
 docker run -d \
   --name "$NAME" \
   --gpus all \
-  --ipc host \
+  --privileged \
-  --shm-size 16g \
+  --ipc=host \
+  --network host \
+  --ulimit memlock=-1 \
+  --ulimit stack=67108864 \
   --entrypoint /bin/bash \
-  -p "${PORT}:${PORT}" \
   -v "${MODEL_PATH}:${MODEL_PATH}:ro" \
-  -v "${CACHE_DIR}/deep_gemm:/root/.cache/deep_gemm" \
+  -v "${CACHE_DIR}:/root/.cache" \
-  -v "${CACHE_DIR}/tvm-ffi:/root/.cache/tvm-ffi" \
   -v "${RUNTIME_BASE}/tmp:/tmp" \
   -e CUDA_VISIBLE_DEVICES="${CUDA_VISIBLE_DEVICES}" \
   -e PYTHONUNBUFFERED=1 \
+  -e HF_HUB_OFFLINE=1 \
+  -e TRANSFORMERS_OFFLINE=1 \
   "$IMAGE" \
   -c "sleep infinity" \
   > "$LOG" 2>&1

experiments/pro6000/dsv4_pro6000_sglang_tp_dp_matrix/start_sglang_docker.sh

@@ -2,8 +2,8 @@
 # Start SGLang server in Docker for a given TPxDP configuration.
 # Usage: start_sglang_docker.sh <TP> <DP>
 #
-# Uses the lmsysorg/sglang image and the same argument set as the
+# Uses the verified SGLang SM120 image and keeps its JIT cache on
-# bare-metal start script. The container is removed automatically on stop.
+# persistent storage. The container is removed automatically on stop.
 set -e
 
 TP="${1}"
@@ -14,9 +14,9 @@ SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 source "${SCRIPT_DIR}/config.env"
 
 RUNTIME_BASE="${RUNTIME_BASE:-${SCRIPT_DIR}/runtime}"
-mkdir -p "${RUNTIME_BASE}/logs" "${RUNTIME_BASE}/tmp"
+mkdir -p "${RUNTIME_BASE}/logs" "${RUNTIME_BASE}/tmp" "$CACHE_DIR"
 
-IMAGE="${DOCKER_IMAGE:-lmsysorg/sglang:latest}"
+IMAGE="${DOCKER_IMAGE:-sglang-sm120-dsv4:0.5.15.post1-fi0.6.14-sm120fix1}"
 PORT="${SGLANG_PORT:-30031}"
 NAME="${EXPERIMENT}_sglang_tp${TP}_dp${DP}"
 PID_FILE="${RUNTIME_BASE}/${EXPERIMENT}_sglang_tp${TP}_dp${DP}.pid"
@@ -28,7 +28,7 @@ rm -f "$PID_FILE"
 docker rm -f "$NAME" >/dev/null 2>&1 || true
 
 SERVER_ARGS=(
-  serve
+  -m sglang.launch_server
   --model-path "$MODEL_PATH"
   --trust-remote-code
   --tp-size "$TP"
@@ -53,7 +53,7 @@ echo "Image: $IMAGE"
 echo "Model: $MODEL_PATH"
 echo "Container name: $NAME"
 echo "Host port: $PORT"
-echo "Command: sglang ${SERVER_ARGS_STR}"
+echo "Command: python3 ${SERVER_ARGS_STR}"
 echo "Log: $LOG"
 
 # Run docker in the foreground so that killing the host process stops the
@@ -62,14 +62,19 @@ echo "Log: $LOG"
 nohup docker run --rm \
   --name "$NAME" \
   --gpus all \
-  --ipc host \
+  --privileged \
-  --shm-size 16g \
+  --ipc=host \
-  --entrypoint sglang \
+  --network host \
-  -p "${PORT}:${PORT}" \
+  --ulimit memlock=-1 \
+  --ulimit stack=67108864 \
+  --entrypoint python3 \
   -v "${MODEL_PATH}:${MODEL_PATH}:ro" \
+  -v "${CACHE_DIR}:/root/.cache" \
   -v "${RUNTIME_BASE}/tmp:/tmp" \
   -e CUDA_VISIBLE_DEVICES="${CUDA_VISIBLE_DEVICES}" \
   -e PYTHONUNBUFFERED=1 \
+  -e HF_HUB_OFFLINE=1 \
+  -e TRANSFORMERS_OFFLINE=1 \
   "$IMAGE" \
   "${SERVER_ARGS[@]}" \
   > "$LOG" 2>&1 &

experiments/pro6000/dsv4_pro6000_sglang_tp_dp_matrix/start_sglang_dp.sh

@@ -33,7 +33,7 @@ PID_FILE="${RUNTIME_BASE}/${EXPERIMENT}_sglang_tp${TP}_dp${DP}.pid"
 rm -f "$PID_FILE"
 
 SERVER_ARGS=(
-  sglang serve
+  python3 -m sglang.launch_server
   --model-path "$MODEL_PATH"
   --trust-remote-code
   --tp-size "$TP"

experiments/pro6000/dsv4_pro6000_vllm_tp_dp_matrix/adaptive_config.env

@@ -15,11 +15,16 @@ NUM_PROMPTS_MULTIPLIER="${NUM_PROMPTS_MULTIPLIER:-5}"
 TPS_MIN_GAIN_PCT="${TPS_MIN_GAIN_PCT:-2.0}"
 PLATEAU_PATIENCE="${PLATEAU_PATIENCE:-2}"
 
+# Stop a shape when p95 TTFT exceeds the SLO; keep group skipping disabled.
+TTFT_SLO_MS="${TTFT_SLO_MS:-4000}"
+ENABLE_TTFT_SLO_STOP="${ENABLE_TTFT_SLO_STOP:-1}"
+TTFT_GROUP_SKIP_MS="${TTFT_GROUP_SKIP_MS:-0}"
+
 # Keep the same random workload semantics as the fixed matrix baseline.
 # DATASET_PATH must contain at least SEARCH_MAX_CONCURRENCY times
 # NUM_PROMPTS_MULTIPLIER valid two-turn conversations. Set this explicitly to
 # random-ids to use generated token IDs without a ShareGPT seed dataset.
-BENCH_DATASET_NAME="${BENCH_DATASET_NAME:-random}"
+BENCH_DATASET_NAME="${BENCH_DATASET_NAME:-random-ids}"
 # SGLang interprets 0.0 as Uniform[1, requested_len]. Use 1.0 for fixed
 # ISL/OSL points; lower values intentionally benchmark a length distribution.
 RANDOM_RANGE_RATIO="${RANDOM_RANGE_RATIO:-1.0}"

experiments/pro6000/dsv4_pro6000_vllm_tp_dp_matrix/config.env

@@ -6,7 +6,7 @@
 
 EXPERIMENT="dsv4_pro6000_vllm_tp_dp_matrix"
 MODEL_NAME="DeepSeek-V4-Flash"
-MODEL_PATH="/data/hf_models/DeepSeek-V4-Flash"
+MODEL_PATH="/data/6000D/DeepSeek-V4-Flash"
 SERVED_MODEL_NAME="deepseek-v4-flash"
 
 VLLM_PORT="${VLLM_PORT:-30030}"
@@ -41,17 +41,19 @@ BLOCK_SIZE="${BLOCK_SIZE:-256}"
 
 # Deployment switch. 0 = native vllm venv, 1 = Docker.
 USE_DOCKER="${USE_DOCKER:-1}"
-DOCKER_IMAGE="${DOCKER_IMAGE:-vllm/vllm-openai:latest}"
+DOCKER_IMAGE="${DOCKER_IMAGE:-vllm-sm120-dsv4:0.25.1-fi0.6.14}"
 
-# Benchmark client Docker image. vLLM's image does not include sglang.bench_serving,
+# Benchmark client Docker image. vLLM's image does not include the SGLang benchmark client,
 # so the client runs inside the SGLang image and targets the vLLM backend via the
 # OpenAI-compatible API.
-DOCKER_CLIENT_IMAGE="${DOCKER_CLIENT_IMAGE:-lmsysorg/sglang:latest}"
+DOCKER_CLIENT_IMAGE="${DOCKER_CLIENT_IMAGE:-sglang-sm120-dsv4:0.5.15.post1-fi0.6.14-sm120fix1}"
 
-# Dataset used by sglang.bench_serving --dataset-name random.
+# Offline by default: random-ids generates token IDs without downloading data.
-# The random sampler needs a ShareGPT-style JSON file locally; it falls back to
+# To use ShareGPT, set BENCH_DATASET_NAME=random and DATASET_PATH explicitly.
-# downloading from HuggingFace, which usually fails on offline nodes.
+BENCH_DATASET_NAME="${BENCH_DATASET_NAME:-random-ids}"
-DATASET_PATH="${DATASET_PATH:-/data/yy/sskj/datasets/ShareGPT_V3_unfiltered_cleaned_split.json}"
+DATASET_PATH="${DATASET_PATH:-}"
+SGLANG_BENCH_MODULE="${SGLANG_BENCH_MODULE:-sglang.benchmark.serving}"
+CACHE_DIR="${CACHE_DIR:-/data/yy/sskj/vllm_sm120_cache}"
 
 # Matrix and concurrency rules are defined in matrix.json by default.
 MATRIX_FILE="${MATRIX_FILE:-${SCRIPT_DIR:-.}/matrix.json}"

experiments/pro6000/dsv4_pro6000_vllm_tp_dp_matrix/matrix.json

@@ -0,0 +1,128 @@
+{
+  "comment": "ISL/OSL matrix for dsv4_pro6000_vllm_tp_dp_matrix. Y=must test, P=optional (record N on failure), N=skip.",
+  "mode": "Y",
+  "matrix": {
+    "1024": {
+      "128": "Y",
+      "256": "Y",
+      "512": "Y",
+      "1024": "Y",
+      "2048": "Y",
+      "4096": "Y"
+    },
+    "4096": {
+      "128": "Y",
+      "256": "Y",
+      "512": "Y",
+      "1024": "Y",
+      "2048": "Y",
+      "4096": "Y"
+    },
+    "8192": {
+      "128": "Y",
+      "256": "Y",
+      "512": "Y",
+      "1024": "Y",
+      "2048": "Y",
+      "4096": "Y"
+    },
+    "16384": {
+      "128": "Y",
+      "256": "Y",
+      "512": "Y",
+      "1024": "Y",
+      "2048": "Y",
+      "4096": "P"
+    },
+    "32768": {
+      "128": "Y",
+      "256": "Y",
+      "512": "Y",
+      "1024": "Y",
+      "2048": "Y",
+      "4096": "P"
+    },
+    "65536": {
+      "128": "Y",
+      "256": "Y",
+      "512": "Y",
+      "1024": "Y",
+      "2048": "P",
+      "4096": "N"
+    },
+    "131072": {
+      "128": "Y",
+      "256": "Y",
+      "512": "Y",
+      "1024": "P",
+      "2048": "N",
+      "4096": "N"
+    },
+    "262144": {
+      "128": "Y",
+      "256": "Y",
+      "512": "P",
+      "1024": "N",
+      "2048": "N",
+      "4096": "N"
+    },
+    "524288": {
+      "128": "Y",
+      "256": "P",
+      "512": "N",
+      "1024": "N",
+      "2048": "N",
+      "4096": "N"
+    },
+    "1048576": {
+      "128": "Y",
+      "256": "P",
+      "512": "N",
+      "1024": "N",
+      "2048": "N",
+      "4096": "N"
+    }
+  },
+  "concurrency": {
+    "1024": {
+      "low": 1,
+      "high": 128
+    },
+    "4096": {
+      "low": 1,
+      "high": 64
+    },
+    "8192": {
+      "low": 1,
+      "high": 64
+    },
+    "16384": {
+      "low": 1,
+      "high": 32
+    },
+    "32768": {
+      "low": 1,
+      "high": 16
+    },
+    "65536": {
+      "low": 1,
+      "high": 8
+    },
+    "131072": {
+      "low": 1,
+      "high": 4
+    },
+    "262144": {
+      "low": 1,
+      "high": 2
+    },
+    "524288": {
+      "low": 1,
+      "high": 2
+    },
+    "1048576": {
+      "low": 1,
+      "high": 2
+    }
+  }
+}

experiments/pro6000/dsv4_pro6000_vllm_tp_dp_matrix/run_adaptive_concurrency.sh

@@ -27,8 +27,8 @@ else
   PYTHON="$(command -v python3)"
 fi
 
-DOCKER_IMAGE="${DOCKER_IMAGE:-vllm/vllm-openai:latest}"
+DOCKER_IMAGE="${DOCKER_IMAGE:-vllm-sm120-dsv4:0.25.1-fi0.6.14}"
-DOCKER_CLIENT_IMAGE="${DOCKER_CLIENT_IMAGE:-lmsysorg/sglang:latest}"
+DOCKER_CLIENT_IMAGE="${DOCKER_CLIENT_IMAGE:-sglang-sm120-dsv4:0.5.15.post1-fi0.6.14-sm120fix1}"
 
 engine_is_healthy() {
   curl --fail --silent --show-error --max-time 5 \
@@ -68,7 +68,6 @@ engine_build_server_args() {
     --gpu-memory-utilization "$GPU_MEMORY_UTILIZATION"
     --max-model-len "$MAX_MODEL_LEN"
     --max-num-seqs "$MAX_NUM_SEQS"
-    --no-enable-flashinfer-autotune
     --host 0.0.0.0
     --port "$ENGINE_PORT"
   )
@@ -150,15 +149,16 @@ engine_run_bench() {
       -e PYTHONUNBUFFERED=1 \
       -e HF_HUB_OFFLINE=1 \
       -e TRANSFORMERS_OFFLINE=1 \
+      --entrypoint python3 \
       "$DOCKER_CLIENT_IMAGE" \
-      python -m sglang.bench_serving "${bench_args[@]}"
+      -m "$SGLANG_BENCH_MODULE" "${bench_args[@]}"
   else
-    "$PYTHON" -m sglang.bench_serving "${bench_args[@]}"
+    "$PYTHON" -m "$SGLANG_BENCH_MODULE" "${bench_args[@]}"
   fi
 }
 
 export -f engine_run_bench
 export ENGINE_PORT MODEL_PATH RESULT_BASE DOCKER_CLIENT_IMAGE USE_DOCKER_CLIENT
-export BENCH_DATASET_NAME DATASET_PATH RANDOM_RANGE_RATIO BENCH_WARMUP_MAX_REQUESTS PYTHON
+export BENCH_DATASET_NAME DATASET_PATH RANDOM_RANGE_RATIO BENCH_WARMUP_MAX_REQUESTS PYTHON SGLANG_BENCH_MODULE
 
 adaptive_main "$@"

experiments/pro6000/dsv4_pro6000_vllm_tp_dp_matrix/run_adaptive_concurrency_add16.sh

@@ -27,8 +27,8 @@ else
   PYTHON="$(command -v python3)"
 fi
 
-DOCKER_IMAGE="${DOCKER_IMAGE:-vllm/vllm-openai:latest}"
+DOCKER_IMAGE="${DOCKER_IMAGE:-vllm-sm120-dsv4:0.25.1-fi0.6.14}"
-DOCKER_CLIENT_IMAGE="${DOCKER_CLIENT_IMAGE:-lmsysorg/sglang:latest}"
+DOCKER_CLIENT_IMAGE="${DOCKER_CLIENT_IMAGE:-sglang-sm120-dsv4:0.5.15.post1-fi0.6.14-sm120fix1}"
 
 engine_is_healthy() {
   curl --fail --silent --show-error --max-time 5 \
@@ -68,7 +68,6 @@ engine_build_server_args() {
     --gpu-memory-utilization "$GPU_MEMORY_UTILIZATION"
     --max-model-len "$MAX_MODEL_LEN"
     --max-num-seqs "$MAX_NUM_SEQS"
-    --no-enable-flashinfer-autotune
     --host 0.0.0.0
     --port "$ENGINE_PORT"
   )
@@ -150,16 +149,17 @@ engine_run_bench() {
       -e PYTHONUNBUFFERED=1 \
       -e HF_HUB_OFFLINE=1 \
       -e TRANSFORMERS_OFFLINE=1 \
+      --entrypoint python3 \
       "$DOCKER_CLIENT_IMAGE" \
-      python -m sglang.bench_serving "${bench_args[@]}"
+      -m "$SGLANG_BENCH_MODULE" "${bench_args[@]}"
   else
-    "$PYTHON" -m sglang.bench_serving "${bench_args[@]}"
+    "$PYTHON" -m "$SGLANG_BENCH_MODULE" "${bench_args[@]}"
   fi
 }
 
 export -f engine_run_bench
 export ENGINE_PORT MODEL_PATH RESULT_BASE DOCKER_CLIENT_IMAGE USE_DOCKER_CLIENT
-export BENCH_DATASET_NAME DATASET_PATH RANDOM_RANGE_RATIO BENCH_WARMUP_MAX_REQUESTS PYTHON
+export BENCH_DATASET_NAME DATASET_PATH RANDOM_RANGE_RATIO BENCH_WARMUP_MAX_REQUESTS PYTHON SGLANG_BENCH_MODULE
 
 export SEARCH_START_CONCURRENCY=16
 export SEARCH_ADDEND=16

experiments/pro6000/dsv4_pro6000_vllm_tp_dp_matrix/start_vllm_docker.sh

@@ -2,8 +2,8 @@
 # Start vLLM server in Docker for a given TPxDP configuration.
 # Usage: start_vllm_docker.sh <TP> <DP>
 #
-# Uses the vllm/vllm-openai image and the same argument set as the
+# Uses the verified vLLM SM120 image and keeps its cache on persistent
-# bare-metal start script. The container is removed automatically on stop.
+# storage. The container is removed automatically on stop.
 set -e
 
 TP="${1}"
@@ -14,9 +14,9 @@ SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 source "${SCRIPT_DIR}/config.env"
 
 RUNTIME_BASE="${RUNTIME_BASE:-${SCRIPT_DIR}/runtime}"
-mkdir -p "${RUNTIME_BASE}/logs" "${RUNTIME_BASE}/tmp"
+mkdir -p "${RUNTIME_BASE}/logs" "${RUNTIME_BASE}/tmp" "$CACHE_DIR"
 
-IMAGE="${DOCKER_IMAGE:-vllm/vllm-openai:latest}"
+IMAGE="${DOCKER_IMAGE:-vllm-sm120-dsv4:0.25.1-fi0.6.14}"
 PORT="${VLLM_PORT:-30030}"
 NAME="${EXPERIMENT}_vllm_tp${TP}_dp${DP}"
 PID_FILE="${RUNTIME_BASE}/${EXPERIMENT}_vllm_tp${TP}_dp${DP}.pid"
@@ -36,7 +36,6 @@ SERVER_ARGS=(
   --gpu-memory-utilization "$GPU_MEMORY_UTILIZATION"
   --max-model-len "$MAX_MODEL_LEN"
   --max-num-seqs "$MAX_NUM_SEQS"
-  --no-enable-flashinfer-autotune
   --host 0.0.0.0
   --port "$PORT"
 )
@@ -63,12 +62,14 @@ echo "Log: $LOG"
 nohup docker run --rm \
   --name "$NAME" \
   --gpus all \
-  --ipc host \
+  --privileged \
-  --shm-size 16g \
+  --ipc=host \
+  --network host \
   --ulimit memlock=-1 \
+  --ulimit stack=67108864 \
   --entrypoint vllm \
-  -p "${PORT}:${PORT}" \
   -v "${MODEL_PATH}:${MODEL_PATH}:ro" \
+  -v "${CACHE_DIR}:/root/.cache" \
   -v "${RUNTIME_BASE}/tmp:/tmp" \
   -e CUDA_VISIBLE_DEVICES="${CUDA_VISIBLE_DEVICES}" \
   -e PYTHONUNBUFFERED=1 \

experiments/pro6000/dsv4_pro6000_vllm_tp_dp_matrix/start_vllm_dp.sh

@@ -41,7 +41,6 @@ SERVER_ARGS=(
   --gpu-memory-utilization "$GPU_MEMORY_UTILIZATION"
   --max-model-len "$MAX_MODEL_LEN"
   --max-num-seqs "$MAX_NUM_SEQS"
-  --no-enable-flashinfer-autotune
   --host 0.0.0.0
   --port "$VLLM_PORT"
 )

scripts/common/platform.sh

@@ -21,7 +21,9 @@ if [[ -z "${PLATFORM:-}" ]]; then
     PLATFORM="kunlun_p800"
   elif command -v nvidia-smi >/dev/null 2>&1; then
     _GPU_NAME=$(nvidia-smi --query-gpu=name --format=csv,noheader 2>/dev/null | head -n 1 || true)
-    if [[ -n "$_GPU_NAME" ]] && grep -qi "H20" <<< "$_GPU_NAME"; then
+    if [[ -n "$_GPU_NAME" ]] && grep -qiE "RTX 6000D|PRO 6000D" <<< "$_GPU_NAME"; then
+      PLATFORM="nvidia_rtx6000d"
+    elif [[ -n "$_GPU_NAME" ]] && grep -qi "H20" <<< "$_GPU_NAME"; then
       PLATFORM="nvidia_h20"
     else
       PLATFORM="nvidia_h200"