Update vLLM MiniMax-M3 image tags / 更新 vLLM MiniMax-M3 镜像标签

benchmarks/single_node/fixed_seq_len/minimaxm3_fp4_b200.sh

@@ -33,7 +33,6 @@ SERVER_LOG=/workspace/server.log
 
 export VLLM_ENGINE_READY_TIMEOUT_S=3600
 export VLLM_FLOAT32_MATMUL_PRECISION=high
-export VLLM_FLASHINFER_ALLREDUCE_BACKEND=trtllm
 
 if [ "${DP_ATTENTION}" = "true" ]; then
   PARALLEL_ARGS="--tensor-parallel-size=1 --data-parallel-size=$TP --enable-expert-parallel"
@@ -52,6 +51,7 @@ start_gpu_monitor
 set -x
 vllm serve $MODEL --port $PORT \
 $PARALLEL_ARGS \
+--attention_config.indexer_kv_dtype fp8 \
 --gpu-memory-utilization 0.95 \
 --max-model-len $MAX_MODEL_LEN \
 --kv-cache-dtype fp8 \

benchmarks/single_node/fixed_seq_len/minimaxm3_fp4_b200_mtp.sh

@@ -47,7 +47,6 @@ SERVER_LOG=/workspace/server.log
 
 export VLLM_ENGINE_READY_TIMEOUT_S=3600
 export VLLM_FLOAT32_MATMUL_PRECISION=high
-export VLLM_FLASHINFER_ALLREDUCE_BACKEND=trtllm
 
 if [ "${DP_ATTENTION}" = "true" ]; then
   PARALLEL_ARGS="--tensor-parallel-size=1 --data-parallel-size=$TP --enable-expert-parallel"

benchmarks/single_node/fixed_seq_len/minimaxm3_fp4_b300.sh

@@ -38,7 +38,6 @@ SERVER_LOG=/workspace/server.log
 
 export VLLM_ENGINE_READY_TIMEOUT_S=3600
 export VLLM_FLOAT32_MATMUL_PRECISION=high
-export VLLM_FLASHINFER_ALLREDUCE_BACKEND=trtllm
 
 if [ "${DP_ATTENTION}" = "true" ]; then
   PARALLEL_ARGS="--tensor-parallel-size=1 --data-parallel-size=$TP --enable-expert-parallel"
@@ -57,6 +56,7 @@ start_gpu_monitor
 set -x
 vllm serve "$MODEL_PATH" --served-model-name "$MODEL" --host 0.0.0.0 --port $PORT \
 $PARALLEL_ARGS \
+--attention_config.indexer_kv_dtype fp8 \
 --gpu-memory-utilization 0.95 \
 --max-model-len $MAX_MODEL_LEN \
 --kv-cache-dtype fp8 \

benchmarks/single_node/fixed_seq_len/minimaxm3_fp4_b300_mtp.sh

@@ -52,7 +52,6 @@ SERVER_LOG=/workspace/server.log
 
 export VLLM_ENGINE_READY_TIMEOUT_S=3600
 export VLLM_FLOAT32_MATMUL_PRECISION=high
-export VLLM_FLASHINFER_ALLREDUCE_BACKEND=trtllm
 
 if [ "${DP_ATTENTION}" = "true" ]; then
   PARALLEL_ARGS="--tensor-parallel-size=1 --data-parallel-size=$TP --enable-expert-parallel"

configs/nvidia-master.yaml

@@ -7527,7 +7527,7 @@ minimaxm3-fp8-b300-vllm:
 # weights are pre-staged read-only at /scratch/models/MiniMax-M3-NVFP4 (added to
 # the STAGED_MODELS allow-list in launch_b300-nv.sh).
 minimaxm3-fp4-b300-vllm:
-  image: vllm/vllm-openai:nightly-93d8f834dd8acf33eb0e2a75b2711b628cb6e226
+  image: vllm/vllm-openai:nightly-4080263bb2c5d10deac17aaeb88e0823bc35bca9
   model: nvidia/MiniMax-M3-NVFP4
   model-prefix: minimaxm3
   runner: b300
@@ -7553,7 +7553,7 @@ minimaxm3-fp4-b300-vllm:
 # /scratch/models/MiniMax-M3-NVFP4 (added to the STAGED_MODELS allow-list in
 # launch_b300-nv.sh); the EAGLE3 draft is downloaded to the writable models dir.
 minimaxm3-fp4-b300-vllm-mtp:
-  image: vllm/vllm-openai:nightly-93d8f834dd8acf33eb0e2a75b2711b628cb6e226
+  image: vllm/vllm-openai:nightly-4080263bb2c5d10deac17aaeb88e0823bc35bca9
   model: nvidia/MiniMax-M3-NVFP4
   model-prefix: minimaxm3
   runner: b300
@@ -7608,7 +7608,7 @@ minimaxm3-fp8-b200-vllm:
 # (MSA sparse/index cache); weights are pre-staged at /scratch/fsw/models/MiniMax-M3-NVFP4
 # (launch_b200-dgxc.sh resolves MODEL_PATH for minimaxm3-fp4).
 minimaxm3-fp4-b200-vllm:
-  image: vllm/vllm-openai:nightly-93d8f834dd8acf33eb0e2a75b2711b628cb6e226
+  image: vllm/vllm-openai:nightly-4080263bb2c5d10deac17aaeb88e0823bc35bca9
   model: nvidia/MiniMax-M3-NVFP4
   model-prefix: minimaxm3
   runner: b200-dgxc
@@ -7663,7 +7663,7 @@ minimaxm3-fp8-b200-vllm-mtp:
 # resolves MODEL_PATH for minimaxm3-fp4); the EAGLE3 draft is fetched next to
 # the target weights.
 minimaxm3-fp4-b200-vllm-mtp:
-  image: vllm/vllm-openai:vllm-minimax-m3-perf-x86_64-13.0.1-8b00f41
+  image: vllm/vllm-openai:nightly-4080263bb2c5d10deac17aaeb88e0823bc35bca9
   model: nvidia/MiniMax-M3-NVFP4
   model-prefix: minimaxm3
   runner: b200-dgxc

perf-changelog.yaml

@@ -5191,3 +5191,13 @@
     - "Agentic experiment Variant K (probe, of the #2374 Variant G): REMOVES the kimi-k3-container-deps.sh in-container patch (setup_script, script file, and launcher copy) to verify whether the mamba_hybrid index_fill_ dtype patch is still required by the current vllm/vllm-openai:kimi-k3 image — expected to fail at the first decode step if it is"
     - "Recipe: benchmarks/multi_node/srt-slurm-recipes/vllm/kimi-k3/agentic/agg-b200-tp8pp2-agentic.yaml on the cluster:b200-dgxc pool"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2391
+
+- config-keys:
+    - minimaxm3-fp4-b200-vllm
+    - minimaxm3-fp4-b200-vllm-mtp
+    - minimaxm3-fp4-b300-vllm
+    - minimaxm3-fp4-b300-vllm-mtp
+  description:
+    - "Update vllm image, remove setting trtllm allreduce backend"
+    - "Use fp8 indexer cache dtype for the B200/B300 non-eagle configs"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2120