From a0792b8c23308a839f7a7c209100824e52b2fc63 Mon Sep 17 00:00:00 2001
From: okorzh-amd <okorzh-amd@users.noreply.github.com>
Date: Mon, 13 Jul 2026 16:46:04 -0500
Subject: [PATCH 1/3] Update kimik2.5 fp4 mi355x vLLM config to match
 workloads-inference K2.6

Align the MI355X vLLM launch flags and AITER env vars with the
Kimi-K2.6-MXFP4 recipe from workloads-inference, keeping the kimik2.5
model and existing InferenceX benchmark infrastructure unchanged.
---
 .../fixed_seq_len/kimik2.5_fp4_mi355x.sh      | 32 +++++++++++--------
 1 file changed, 19 insertions(+), 13 deletions(-)

diff --git a/benchmarks/single_node/fixed_seq_len/kimik2.5_fp4_mi355x.sh b/benchmarks/single_node/fixed_seq_len/kimik2.5_fp4_mi355x.sh
index d4616143ac..d99fa64b92 100755
--- a/benchmarks/single_node/fixed_seq_len/kimik2.5_fp4_mi355x.sh
+++ b/benchmarks/single_node/fixed_seq_len/kimik2.5_fp4_mi355x.sh
@@ -46,12 +46,13 @@ if [[ "$version" == "" || $version -lt 177 ]]; then
 fi
 
 export VLLM_ROCM_USE_AITER=1
+export AMDGCN_USE_BUFFER_OPS=1
+export VLLM_ROCM_USE_AITER_MLA_PS=1
 export VLLM_ROCM_QUICK_REDUCE_QUANTIZATION=INT4
-
-# Disable AITER RMSNorm for TP < 8 due to accuracy issues
-if [ "${TP}" -lt 8 ]; then
-  export VLLM_ROCM_USE_AITER_RMSNORM=0
-fi
+export VLLM_ROCM_USE_SKINNY_GEMM=0
+export VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS=1
+export VLLM_ROCM_USE_AITER_TUNED_UNQUANTISED_GEMM=1
+export VLLM_ROCM_DISABLE_ATTENTION_LINEAR_LAYER_DYNAMIC_MXFP4_QUANT=1
 
 if [ "${EP_SIZE:-0}" -gt 1 ]; then
   EP=" --enable-expert-parallel"
@@ -59,9 +60,6 @@ else
   EP=" "
 fi
 
-# following AMD andy luo's recipe
-# https://x.com/linluo77/status/2017024513595301985
-
 # Start GPU monitoring (power, temperature, clocks every second)
 start_gpu_monitor
 
@@ -69,13 +67,21 @@ set -x
 vllm serve $MODEL --port $PORT \
 --tensor-parallel-size=$TP \
 $EP \
---gpu-memory-utilization 0.90 \
---max-model-len $MAX_MODEL_LEN \
---block-size=1 \
---no-enable-prefix-caching \
 --trust-remote-code \
 --no-enable-prefix-caching \
---mm-encoder-tp-mode data > $SERVER_LOG 2>&1 &
+--max-model-len $MAX_MODEL_LEN \
+--max-num-seqs 512 \
+--max-num-batched-tokens 65536 \
+--reasoning-parser kimi_k2 \
+--tool-call-parser kimi_k2 \
+--enable-auto-tool-choice \
+--gpu-memory-utilization 0.9 \
+--mm-encoder-tp-mode data \
+--attention-backend ROCM_AITER_MLA \
+--block-size 1 \
+--kv-cache-dtype fp8 \
+--compilation-config '{"pass_config": {"fuse_allreduce_rms": true, "eliminate_noops": true, "fuse_rope_kvcache_cat_mla": true}, "custom_ops": ["none", "+rms_norm"], "compile_ranges_endpoints": [64], "cudagraph_mode": "full_and_piecewise", "use_inductor_graph_partition": true}' \
+> $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!
 

From deee77c6983c54636e76434a9828426e00482a30 Mon Sep 17 00:00:00 2001
From: ppalanga <ppalanga@amd.com>
Date: Thu, 16 Jul 2026 13:20:01 -0700
Subject: [PATCH 2/3] Update perf-changelog.yaml

---
 perf-changelog.yaml | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 57d510dd15..34d97fb1af 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -4750,3 +4750,10 @@
     - "Image: lmsysorg/sglang:nightly-dev-cu13-20260709-074bb928"
     - "6 topologies across 1k/1k and 8k/1k: 1P1D TP4 STP + wide-EP (DEP4 prefill / DEP16 decode) from 1P1D up to 8P1D, recipes under benchmarks/multi_node/srt-slurm-recipes/sglang/qwen3.5/gb300-fp8/"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2137
+
+
+- config-keys:
+    - kimik2.5-fp4-mi355x-vllm
+  description:
+    - "Update the server configuration for Kimi-K2.5-MXFP4"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2203

From edf9e6efdb04462f2a7bcd85dac2d95cbabda2da Mon Sep 17 00:00:00 2001
From: okorzh-amd <okorzh-amd@users.noreply.github.com>
Date: Thu, 16 Jul 2026 17:30:34 -0500
Subject: [PATCH 3/3] Add trailing newline to perf-changelog.yaml

Fixes CI check requiring the file to end with a newline.
---
 perf-changelog.yaml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 25c0b66215..02e4437f3f 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -4899,4 +4899,4 @@
     - kimik2.5-fp4-mi355x-vllm
   description:
     - "Update the server configuration for Kimi-K2.5-MXFP4"
-  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2203
\ No newline at end of file
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2203