[AMD] Update MiniMax-M3-MXFP4 MI355X vLLM disagg perf and config / 更新 MiniMax-M3-MXFP4 MI355X vLLM disagg 性能与配置

benchmarks/multi_node/amd_utils/models_vllm.yaml

@@ -41,7 +41,8 @@ MiniMax-M2.5:
 MiniMax-M3-MXFP4:
   prefill_flags: "--tensor-parallel-size 8 --max-num-batched-tokens 32768 --max-num-seqs 512 --block-size 128 --language-model-only --attention-backend TRITON_ATTN --moe-backend aiter --no-enable-prefix-caching --gpu-memory-utilization 0.90 --tool-call-parser minimax_m3 --reasoning-parser minimax_m3 --enable-auto-tool-choice"
   decode_flags: "--tensor-parallel-size 8 --max-num-batched-tokens 32768 --max-num-seqs 512 --block-size 128 --language-model-only --attention-backend TRITON_ATTN --moe-backend aiter --no-enable-prefix-caching --gpu-memory-utilization 0.90 --tool-call-parser minimax_m3 --reasoning-parser minimax_m3 --enable-auto-tool-choice"
-  env: "VLLM_USE_V1=1 VLLM_ROCM_USE_AITER=1 VLLM_ROCM_USE_AITER_MOE=1 VLLM_USE_BREAKABLE_CUDAGRAPH=0 VLLM_ENGINE_READY_TIMEOUT_S=3600"
+  env: "VLLM_USE_V1=1 VLLM_ROCM_USE_AITER=1 VLLM_ROCM_USE_AITER_MOE=1 VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS=1 VLLM_USE_BREAKABLE_CUDAGRAPH=0 VLLM_ENGINE_READY_TIMEOUT_S=3600"
+  prefill_env: "VLLM_ROCM_QUICK_REDUCE_QUANTIZATION=INT4 VLLM_ROCM_QUICK_REDUCE_MAX_SIZE_BYTES_MB=2048"
   hf_dir: "models--amd--MiniMax-M3-MXFP4"
 
 gpt-oss-120b:

benchmarks/multi_node/amd_utils/server_vllm.sh

@@ -145,10 +145,12 @@ pf = bash_escape(m.get('prefill_flags', '--tensor-parallel-size 8'))
 df = bash_escape(m.get('decode_flags', '--tensor-parallel-size 8'))
 ev = bash_escape(m.get('env', ''))
 dev = bash_escape(m.get('decode_env', ''))
+pev = bash_escape(m.get('prefill_env', ''))
 print(f'PREFILL_SERVER_CONFIG=\"{pf}\"')
 print(f'DECODE_SERVER_CONFIG=\"{df}\"')
 print(f'MODEL_ENVS=\"{ev}\"')
 print(f'DECODE_MODEL_ENVS=\"{dev}\"')
+print(f'PREFILL_MODEL_ENVS=\"{pev}\"')
 ")"
 
 echo "Loaded model configuration for: $MODEL_NAME"
@@ -251,6 +253,11 @@ if [ "$NODE_RANK" -eq 0 ]; then
 
     setup_vllm_env
 
+    for env_pair in ${PREFILL_MODEL_ENVS}; do
+        export "$env_pair"
+        echo "[PREFILL_ENV] $env_pair"
+    done
+
     # Router is started as an external container by job.slurm (VLLM_ROUTER_IMAGE)
     echo "Using external vllm-router container (started by job.slurm on this node)"
 
@@ -420,6 +427,11 @@ elif [ "$NODE_RANK" -gt 0 ] && [ "$NODE_RANK" -lt "$xP" ]; then
 
     setup_vllm_env
 
+    for env_pair in ${PREFILL_MODEL_ENVS}; do
+        export "$env_pair"
+        echo "[PREFILL_ENV] $env_pair"
+    done
+
     SERVED_MODEL="${MODEL_NAME}"
     PREFILL_CMD="vllm serve ${MODEL_PATH} \
         --served-model-name ${SERVED_MODEL} \

configs/amd-master.yaml

@@ -2420,7 +2420,7 @@ minimaxm3-fp8-mi355x-vllm-mtp:
 
 # MiniMax-M3 MXFP4 MI355X vLLM disaggregated (prefill/decode) config.
 minimaxm3-fp4-mi355x-vllm-disagg:
-  image: rocm/vllm-dev:vllm-0.23.1-rocm723-mi35x-mori-0625
+  image: vllm/vllm-openai-rocm:nightly-2dfaae752b4db0d43cfc0715c780e33be030d0f1
   model: amd/MiniMax-M3-MXFP4
   model-prefix: minimaxm3
   runner: mi355x-disagg
@@ -2433,9 +2433,9 @@ minimaxm3-fp4-mi355x-vllm-disagg:
     - isl: 8192
       osl: 1024
       search-space:
-      # 1P TP4 + 1D TP4 (2 nodes total), conc sweep 1..512 (single job, looped)
+      # 1P TP4 + 1D TP4 (2 nodes total), conc sweep 1..256 (single job, looped)
       - spec-decoding: "none"
-        conc-list: [ 1, 2, 4, 8, 16, 32, 64, 128, 256, 512 ]
+        conc-list: [ 1, 2, 4, 8, 16, 32, 64, 128, 256 ]
         prefill:
           num-worker: 1
           tp: 4
@@ -2450,23 +2450,6 @@ minimaxm3-fp4-mi355x-vllm-disagg:
           dp-attn: false
           additional-settings:
           - "DECODE_NODES=1"
-      # 2P TP4 + 1D TP4 (3 nodes total), conc 128/256/512 (single job, looped)
-      - spec-decoding: "none"
-        conc-list: [ 128, 256, 512 ]
-        prefill:
-          num-worker: 2
-          tp: 4
-          ep: 1
-          dp-attn: false
-          additional-settings:
-          - "PREFILL_NODES=2"
-        decode:
-          num-worker: 1
-          tp: 4
-          ep: 1
-          dp-attn: false
-          additional-settings:
-          - "DECODE_NODES=1"
 # MiniMax-M3 MXFP4 MI355X vLLM recipe. The pinned nightly includes upstream
 # MiniMax-M3 Quark MXFP4 support (vllm-project/vllm#45794). Use the text-only
 # language-model path and mirror the MXFP8 MI355X search space for a direct

perf-changelog.yaml

@@ -4575,6 +4575,15 @@
     - "Add high concurrency configs"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1994
 
+- config-keys:
+    - minimaxm3-fp4-mi355x-vllm-disagg
+  description:
+    - "Update the MiniMax-M3 MXFP4 MI355X vLLM disagg image to vllm/vllm-openai-rocm:nightly-2dfaae752b4db0d43cfc0715c780e33be030d0f1 (from rocm/vllm-dev:vllm-0.23.1-rocm723-mi35x-mori-0625) for AITER MoE and shared-expert fusion support."
+    - "Export VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS=1 for MiniMax-M3-MXFP4 (both prefill and decode)."
+    - "Enable prefill-only INT4 quick-reduce: set VLLM_ROCM_QUICK_REDUCE_QUANTIZATION=INT4 and VLLM_ROCM_QUICK_REDUCE_MAX_SIZE_BYTES_MB=2048 on the prefill workers via a new prefill_env channel (mirrors the existing decode_env path in server_vllm.sh)."
+    - "Cap the 1P1D TP4 concurrency sweep at 256 (was 512); drop the 2P1D TP4 layout (128/256/512) as it is CI-flaky with negligible curve impact."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1943
+
 - config-keys:
     - minimaxm3-fp8-mi355x-vllm-mtp
   description: