[AMD] Add MiniMax-M3-FP4 MI355X ATOM EAGLE3 only

.github/configs/amd-master.yaml

@@ -2612,6 +2612,27 @@ minimaxm3-fp4-mi355x-atom:
       - { tp: 4, conc-start: 1, conc-end: 256 }
       - { tp: 8, conc-start: 1, conc-end: 2 }
 
+minimaxm3-fp4-mi355x-atom-mtp:
+  image: rocm/atom-dev:MiniMax-M3-20260622
+  model: amd/MiniMax-M3-MXFP4
+  model-prefix: minimaxm3
+  runner: mi355x
+  precision: fp4
+  framework: atom
+  multinode: false
+  scenarios:
+    fixed-seq-len:
+    - isl: 1024
+      osl: 1024
+      search-space:
+      - { tp: 2, conc-start: 128, conc-end: 256, spec-decoding: mtp }
+      - { tp: 4, conc-start: 1, conc-end: 256, spec-decoding: mtp }
+    - isl: 8192
+      osl: 1024
+      search-space:
+      - { tp: 2, conc-start: 128, conc-end: 128, spec-decoding: mtp }
+      - { tp: 4, conc-start: 1, conc-end: 256, spec-decoding: mtp }
+
 # MiniMax-M3 MXFP8 MI300X day-zero recipe. Reuse the dedicated ROCm image and
 # MI355X serving shape, but retain the default BF16 KV cache because this
 # checkpoint lacks calibrated ROCm FP8 attention scales. Use the TP8-only H100

benchmarks/single_node/fixed_seq_len/minimaxm3_fp4_mi355x_atom.sh → benchmarks/single_node/fixed_seq_len/minimaxm3_fp4_mi355x_atom_mtp.sh

@@ -11,8 +11,7 @@ check_env_vars \
     RANDOM_RANGE_RATIO \
     RESULT_FILENAME \
     EP_SIZE \
-    DP_ATTENTION \
-    MAX_MODEL_LEN
+    DP_ATTENTION 
 
 if [[ -n "$SLURM_JOB_ID" ]]; then
   echo "JOB $SLURM_JOB_ID running on $SLURMD_NODENAME"
@@ -22,41 +21,40 @@ echo "TP: $TP, CONC: $CONC, ISL: $ISL, OSL: $OSL, EP_SIZE: $EP_SIZE, DP_ATTENTIO
 
 SERVER_LOG=/workspace/server.log
 
-export OMP_NUM_THREADS=1
+PARALLEL_ARGS=(-tp "$TP") #TP
+if [ "$DP_ATTENTION" = "true" ]; then
+    if [ "$EP_SIZE" -gt 1 ]; then #DP+EP
+        PARALLEL_ARGS=(-tp "$TP" --enable-expert-parallel --enable-dp-attention )
+    else #DP+TP
+        PARALLEL_ARGS=(-tp "$TP" --enable-dp-attention )
+    fi
+fi 
 
-# Use the matrix-supplied MAX_MODEL_LEN (isl + osl + 256). Eval-only jobs need a
-# larger window for the eval prompts, so override it from the eval context.
-if [ "${EVAL_ONLY}" = "true" ]; then
-    setup_eval_context
-    MAX_MODEL_LEN="$EVAL_MAX_MODEL_LEN"
-fi
-
-if [ "$EP_SIZE" -gt 1 ]; then
-  EP=" --enable-expert-parallel"
-else
-  EP=" "
-fi
+SPEC_ARGS=(--method eagle3 --draft-model Inferact/MiniMax-M3-EAGLE3 --num-speculative-tokens 3 )
 
 # Start GPU monitoring (power, temperature, clocks every second)
 start_gpu_monitor
 MEM_FRAC_STATIC=0.8
 
 set -x
-
-# Flags follow the ATOM MiniMax-M3 MXFP4 recipe (FP4 on 4xMI355 section):
-# https://github.com/ROCm/ATOM/blob/5d42d49f9e4292e5b61475917e92e7ec1b1dacb7/recipes/MiniMax-M3.md
-# --block-size 128 is mandatory for MiniMax MSA. KV cache is left at the default
-# dtype: amd/MiniMax-M3-MXFP4 ships no calibrated FP8 KV scales, so
-# --kv_cache_dtype fp8 trips an assertion (k_scale is None) in the MSA
-# fused_qknorm kernel during init.
+export AITER_QUICK_REDUCE_QUANTIZATION=INT4
+export MAX_MODEL_LEN=32768
+export MAX_NUM_BATCHED_TOKENS=32768
+export MAX_NUM_SEQS=128
+# (srok), not yet
+#    --kv_cache_dtype fp8 \
 python3 -m atom.entrypoints.openai_server \
     --model $MODEL \
     --server-port $PORT \
-    -tp $TP \
-    --max-model-len $MAX_MODEL_LEN $EP \
+    "${PARALLEL_ARGS[@]}" \
+    "${SPEC_ARGS[@]}" \
     --block-size 128 \
     --gpu-memory-utilization $MEM_FRAC_STATIC \
+    --max-model-len $MAX_MODEL_LEN \
+    --max-num-batched-tokens $MAX_NUM_BATCHED_TOKENS \
+    --max-num-seqs $MAX_NUM_SEQS \
     --trust-remote-code \
+    --no-enable_prefix_caching \
     > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
@@ -76,7 +74,7 @@ run_benchmark_serving \
     --max-concurrency "$CONC" \
     --result-filename "$RESULT_FILENAME" \
     --result-dir /workspace/ \
-    --trust-remote-code
+    --trust-remote-code $( [[ ${#SPEC_ARGS[@]} -gt 0 ]] && echo "--use-chat-template" )
 
 # After throughput, run evaluation only if RUN_EVAL is true
 if [ "${RUN_EVAL}" = "true" ]; then

perf-changelog.yaml

@@ -3983,6 +3983,13 @@
     - "Remove the runtime SupportsEagle3 source patch now included in the pinned nightly"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1843
 
+- config-keys:
+    - minimaxm3-fp4-mi355x-atom-mtp
+  description:
+    - "Add minimaxm3-fp4-mi355x-atom-mtp CI recipe: single-node ATOM with EAGLE3 speculative decoding (3 draft tokens) for MiniMax-M3-MXFP4 on MI355X"
+    - "Supports TP/EP/DP-attention combinations; search space mirrors minimaxm3-fp4-mi355x-atom (ISL=1024,8192 OSL=1024 TP2/TP4/TP8)"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1866
+
 - config-keys:
     - minimaxm3-fp8-gb300-dynamo-vllm
   description: