[AMD] KIMI fp4 MI355 update docker

benchmarks/single_node/fixed_seq_len/kimik2.5_fp4_mi355x_atom.sh

@@ -21,6 +21,10 @@ echo "TP: $TP, CONC: $CONC, ISL: $ISL, OSL: $OSL, EP_SIZE: $EP_SIZE, DP_ATTENTIO
 
 SERVER_LOG=/workspace/server.log
 
+
+export ATOM_DISABLE_MMAP=true # Model load faster.
+export AITER_QUICK_REDUCE_QUANTIZATION=INT4
+export AITER_MXFP4_INTERMEDIATE=1
 export OMP_NUM_THREADS=1
 
 # Calculate max-model-len based on ISL and OSL
@@ -45,6 +49,7 @@ python3 -m atom.entrypoints.openai_server \
     --model $MODEL \
     --server-port $PORT \
     -tp $TP \
+    --scheduler-delay-factor 1 \
     --kv_cache_dtype fp8 $CALCULATED_MAX_MODEL_LEN $EP \
     --trust-remote-code \
     > $SERVER_LOG 2>&1 &

configs/amd-master.yaml

@@ -809,7 +809,7 @@ kimik2.5-fp4-mi355x-vllm-agentic:
       - { tp: 4, kv-offloading: dram, kv-offload-backend: native, conc-list: [16, 24, 32, 40] }
 
 kimik2.5-fp4-mi355x-atom:
-  image: rocm/atom:rocm7.2.3_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom20260511
+  image: rocm/atom:rocm7.2.4_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.4_202607091539
   model: amd/Kimi-K2.5-MXFP4
   model-prefix: kimik2.5
   runner: mi355x
@@ -821,12 +821,10 @@ kimik2.5-fp4-mi355x-atom:
     - isl: 1024
       osl: 1024
       search-space:
-      - { tp: 8, conc-start: 4, conc-end: 128 }
       - { tp: 4, conc-start: 4, conc-end: 128 }
     - isl: 8192
       osl: 1024
       search-space:
-      - { tp: 8, conc-start: 4, conc-end: 128 }
       - { tp: 4, conc-start: 4, conc-end: 128 }
 
 dsr1-fp8-mi355x-atom:

perf-changelog.yaml

@@ -4587,6 +4587,14 @@
     - "Add high concurrency configs"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1994
 
+- config-keys:
+    - kimik2.5-fp4-mi355x-atom
+  description:
+    - "Update ATOM image from rocm/atom:rocm7.2.3_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom20260511 to rocm7.2.4_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.4_202607091539"
+    - "Enable ATOM_DISABLE_MMAP=true, AITER_QUICK_REDUCE_QUANTIZATION=INT4, AITER_MXFP4_INTERMEDIATE=1; add --scheduler-delay-factor 1 to server launch"
+    - "Conc 4 only for both 1k1k and 8k1k"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2132
+
 - config-keys:
     - dsv4-fp4-mi355x-vllm-agentic
   description: