[AMD][ATOM] 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
+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
@@ -44,7 +48,7 @@ set -x
 python3 -m atom.entrypoints.openai_server \
     --model $MODEL \
     --server-port $PORT \
-    -tp $TP \
+    -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-dev:nightly_202607091539 # wait for a release image, cause Kimi is a old model.
   model: amd/Kimi-K2.5-MXFP4
   model-prefix: kimik2.5
   runner: mi355x
@@ -821,13 +821,13 @@ kimik2.5-fp4-mi355x-atom:
     - isl: 1024
       osl: 1024
       search-space:
-      - { tp: 8, conc-start: 4, conc-end: 128 }
+      - { tp: 8, conc-start: 4, conc-end: 4 }
       - { 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 }
+      - { tp: 8, conc-start: 4, conc-end: 4 }
 
 dsr1-fp8-mi355x-atom:
   image: rocm/atom:rocm7.2.3_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom20260511

perf-changelog.yaml

@@ -4650,3 +4650,9 @@
     - "Update SGLang image from lmsysorg/sglang:v0.5.13-cu130 to lmsysorg/sglang:nightly-dev-cu13-20260707-b4155233"
     - "Enable SGLANG_ENABLE_UNIFIED_RADIX_TREE=1 unconditionally (previously hicache-only) and SGLANG_OPT_UNIFIED_CACHE_FREE_OUT_OF_WINDOW_SLOTS=1 to free out-of-window SWA KV slots during chunked prefill, relieving SWA pool pressure and restoring prefix-cache hit rate on multi-turn agentic workloads"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2112
+
+- config-keys:
+    - kimik2.5-fp4-mi355x-atom
+  description:
+    - "TODO & update image"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2132