Add MI300X GLM-5.2 FP8 AgentX MTP

benchmarks/single_node/agentic/glm5.2_fp8_mi300x_mtp.sh

@@ -0,0 +1,118 @@
+#!/usr/bin/env bash
+set -eo pipefail
+set -x
+
+# GLM-5.2 FP8 on one 8xMI300X node with native EAGLE MTP.
+
+source "$(dirname "$0")/../../benchmark_lib.sh"
+
+export EVAL_FRAMEWORK="lm-eval"
+
+check_env_vars \
+    MODEL TP CONC EP_SIZE KV_OFFLOADING PORT EVAL_ONLY \
+    RESULT_DIR DURATION
+
+require_agentic_kv_offload_none
+
+if [[ -n "${SLURM_JOB_ID:-}" ]]; then
+    echo "JOB $SLURM_JOB_ID running on ${SLURMD_NODENAME:-unknown}"
+fi
+if [[ -n "${ROCR_VISIBLE_DEVICES:-}" ]]; then
+    export HIP_VISIBLE_DEVICES="$ROCR_VISIBLE_DEVICES"
+fi
+
+if [[ -n "${MODEL_PATH:-}" ]]; then
+    if [[ ! -d "$MODEL_PATH" || -z "$(ls -A "$MODEL_PATH" 2>/dev/null)" ]]; then
+        hf download "$MODEL" --local-dir "$MODEL_PATH"
+    fi
+else
+    hf download "$MODEL"
+    export MODEL_PATH="$MODEL"
+fi
+rocm-smi || true
+amd-smi || true
+
+export WEKA_LOADER_OVERRIDE=semianalysis_cc_traces_weka_062126
+resolve_trace_source
+install_agentic_deps
+
+SERVER_LOG="$RESULT_DIR/server.log"
+mkdir -p "$RESULT_DIR"
+SERVER_PID=""
+
+cleanup_agentic_services() {
+    local exit_code=$?
+    trap - EXIT INT TERM
+    set +e
+    stop_background_process_tree "$SERVER_PID" "SGLang server" 60
+    exit "$exit_code"
+}
+trap cleanup_agentic_services EXIT
+trap 'exit 130' INT
+trap 'exit 143' TERM
+
+export PYTHONNOUSERSITE=1
+export AIPERF_HTTP_TCP_USER_TIMEOUT=900000
+export SGLANG_TIMEOUT_KEEP_ALIVE=900
+export SGLANG_DSA_FUSE_TOPK=false
+export SGLANG_OPT_USE_TOPK_V2=false
+
+# Throughput replays use the committed GLM-5.2 thinking-on K=3 golden AL;
+# evals retain real target-model verification.
+if [[ "${EVAL_ONLY:-false}" != "true" ]]; then
+    export SGLANG_SIMULATE_ACC_LEN=2.99
+    export SGLANG_SIMULATE_ACC_METHOD=match-expected
+    export SGLANG_SIMULATE_ACC_TOKEN_MODE=real-draft-token
+fi
+
+MAX_RUNNING_REQUESTS=$((2 * CONC))
+
+SGLANG_CMD=(
+    python3 -m sglang.launch_server
+    --model-path "$MODEL_PATH"
+    --served-model-name "$MODEL"
+    --host 0.0.0.0
+    --port "$PORT"
+    --trust-remote-code
+    --tp "$TP"
+    --ep-size "$EP_SIZE"
+    --dsa-prefill-backend tilelang
+    --dsa-decode-backend tilelang
+    --dsa-topk-backend torch
+    --kv-cache-dtype bfloat16
+    --tool-call-parser glm47
+    --reasoning-parser glm45
+    --context-length 1048576
+    --max-total-tokens 1048576
+    --chunked-prefill-size 131072
+    # Full 131072-token DSA prefills need transient workspace beyond the KV
+    # pool; 0.70 leaves enough headroom on 192 GB MI300X ranks.
+    --mem-fraction-static 0.70
+    --max-running-requests "$MAX_RUNNING_REQUESTS"
+    --cuda-graph-max-bs "$MAX_RUNNING_REQUESTS"
+    --speculative-algorithm EAGLE
+    --speculative-num-steps 3
+    --speculative-eagle-topk 1
+    --speculative-num-draft-tokens 4
+    --watchdog-timeout 1800
+    --enable-metrics
+    --enable-cache-report
+)
+
+write_command "$RESULT_DIR/sglang_command.txt" "${SGLANG_CMD[@]}"
+"${SGLANG_CMD[@]}" > "$SERVER_LOG" 2>&1 &
+SERVER_PID=$!
+
+wait_for_server_ready --port "$PORT" --server-log "$SERVER_LOG" --server-pid "$SERVER_PID"
+
+if [[ "${EVAL_ONLY:-false}" == "true" ]]; then
+    export SWEBENCH_AGENT_STEP_LIMIT=150
+    run_eval --port "$PORT"
+else
+    # Aggregate serving exposes one logical SGLang Prometheus target.
+    export AIPERF_SERVER_METRICS_URLS="http://localhost:$PORT/metrics"
+    export AIPERF_REQUIRED_SERVER_METRIC_PREFIX="sglang:"
+    build_replay_cmd "$RESULT_DIR"
+    REPLAY_CMD+=" --benchmark-grace-period 1800"
+    run_agentic_replay_and_write_outputs "$RESULT_DIR"
+fi

configs/amd-master.yaml

@@ -1539,6 +1539,23 @@ glm5.2-fp8-mi325x-sglang-agentic-mtp:
       search-space:
       - { tp: 8, ep: 1, kv-offloading: none, spec-decoding: mtp, conc-list: [1, 2, 3, 4, 5, 6, 8] }
 
+# Fast discovery found a sharp quality cliff after c3. Keep the verified TP and
+# TEP c3 points; lower concurrency repeatedly failed quality or memory gates.
+glm5.2-fp8-mi300x-sglang-agentic-mtp:
+  image: lmsysorg/sglang:v0.5.16-rocm720-mi30x
+  model: zai-org/GLM-5.2-FP8
+  model-prefix: glm5.2
+  runner: cluster:mi300x-amds
+  precision: fp8
+  framework: sglang
+  multinode: false
+  scenarios:
+    agentic-coding:
+    - dram-utilization: 0.80
+      search-space:
+      - { tp: 8, ep: 1, kv-offloading: none, spec-decoding: mtp, conc-list: [3] }
+      - { tp: 8, ep: 8, kv-offloading: none, spec-decoding: mtp, conc-list: [3] }
+
 minimaxm3-fp8-mi325x-vllm-agentic:
   image: vllm/vllm-openai-rocm:nightly-04c2a8deac44fdb1ca3e2b5ec3e6bf16f3f6a914
   model: MiniMaxAI/MiniMax-M3-MXFP8

perf-changelog.yaml

@@ -5918,3 +5918,11 @@
     - "Add a TEP2 arm (tp 2, ep 2) to the qwen3.5-fp4-b200-sglang-mtp 8k/1k sweep at concurrency 16, 32, and 64"
     - "Rides on the NVFP4-V2 checkpoint switch from #2205"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2550
+
+- config-keys:
+    - glm5.2-fp8-mi300x-sglang-agentic-mtp
+  scenario-type:
+    - agentic-coding
+  description:
+    - "Add MI300X GLM-5.2 FP8 AgentX with native SGLang EAGLE MTP and the verified resident Pareto matrix: TP8 c3 plus TEP8 c3."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2581