From 9bf2a5732a3c8bc0b0134eb258d1a4c1d1179ac8 Mon Sep 17 00:00:00 2001
From: seungrokj <seungrok.jung@amd.com>
Date: Fri, 24 Apr 2026 15:51:10 +0900
Subject: [PATCH 1/2] [AMD/ROCM] atom qwen3.5 fp4 on mi355x

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---
 .github/configs/amd-master.yaml               | 20 +++++
 .../single_node/qwen3.5_fp4_mi355x_atom.sh    | 82 +++++++++++++++++++
 perf-changelog.yaml                           |  9 ++
 3 files changed, 111 insertions(+)
 create mode 100644 benchmarks/single_node/qwen3.5_fp4_mi355x_atom.sh

diff --git a/.github/configs/amd-master.yaml b/.github/configs/amd-master.yaml
index 0cd82774e..0e7539c69 100644
--- a/.github/configs/amd-master.yaml
+++ b/.github/configs/amd-master.yaml
@@ -307,6 +307,26 @@ qwen3.5-fp4-mi355x-sglang:
     - { tp: 2, conc-start: 4, conc-end: 256 }
     - { tp: 4, conc-start: 4, conc-end: 16 }
 
+qwen3.5-fp4-mi355x-atom:
+  image: rocm/atom:rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.2.post
+  model: amd/Qwen3.5-397B-A17B-MXFP4
+  model-prefix: qwen3.5
+  runner: mi355x
+  precision: fp4
+  framework: atom
+  multinode: false
+  seq-len-configs:
+  - isl: 1024
+    osl: 1024
+    search-space:
+    - { tp: 2, conc-start: 4, conc-end: 256 }
+    - { tp: 4, conc-start: 4, conc-end: 16 }
+  - isl: 8192
+    osl: 1024
+    search-space:
+    - { tp: 2, conc-start: 4, conc-end: 256 }
+    - { tp: 4, conc-start: 4, conc-end: 16 }
+
 qwen3.5-fp8-mi300x-sglang:
   image: lmsysorg/sglang:v0.5.10-rocm720-mi30x
   model: Qwen/Qwen3.5-397B-A17B-FP8
diff --git a/benchmarks/single_node/qwen3.5_fp4_mi355x_atom.sh b/benchmarks/single_node/qwen3.5_fp4_mi355x_atom.sh
new file mode 100644
index 000000000..2a8c67da0
--- /dev/null
+++ b/benchmarks/single_node/qwen3.5_fp4_mi355x_atom.sh
@@ -0,0 +1,82 @@
+#!/usr/bin/env bash
+
+source "$(dirname "$0")/../benchmark_lib.sh"
+
+check_env_vars \
+    MODEL \
+    TP \
+    CONC \
+    ISL \
+    OSL \
+    RANDOM_RANGE_RATIO \
+    RESULT_FILENAME \
+    EP_SIZE \
+    DP_ATTENTION
+
+if [[ -n "$SLURM_JOB_ID" ]]; then
+  echo "JOB $SLURM_JOB_ID running on $SLURMD_NODENAME"
+fi
+
+echo "TP: $TP, CONC: $CONC, ISL: $ISL, OSL: $OSL, EP_SIZE: $EP_SIZE, DP_ATTENTION: $DP_ATTENTION"
+
+SERVER_LOG=/workspace/server.log
+PORT=${PORT:-8888}
+
+export OMP_NUM_THREADS=1
+
+# Calculate max-model-len based on ISL and OSL
+if [ "$ISL" = "1024" ] && [ "$OSL" = "1024" ]; then
+    CALCULATED_MAX_MODEL_LEN=""
+else
+    CALCULATED_MAX_MODEL_LEN=" --max-model-len 10240 "
+fi
+
+if [ "$EP_SIZE" -gt 1 ]; then
+  EP=" --enable-expert-parallel"
+else
+  EP=" "
+fi
+
+# Start GPU monitoring (power, temperature, clocks every second)
+start_gpu_monitor
+MEM_FRAC_STATIC=0.9
+
+set -x
+
+python3 -m atom.entrypoints.openai_server \
+    --model $MODEL \
+    --server-port $PORT \
+    -tp $TP \
+    --kv_cache_dtype fp8 $CALCULATED_MAX_MODEL_LEN $EP \
+    --gpu-memory-utilization $MEM_FRAC_STATIC \
+    --trust-remote-code \
+    > $SERVER_LOG 2>&1 &
+
+SERVER_PID=$!
+
+# Wait for server to be ready
+wait_for_server_ready --port "$PORT" --server-log "$SERVER_LOG" --server-pid "$SERVER_PID"
+
+export PYTHONDONTWRITEBYTECODE=1
+run_benchmark_serving \
+    --model "$MODEL" \
+    --port "$PORT" \
+    --backend vllm \
+    --input-len "$ISL" \
+    --output-len "$OSL" \
+    --random-range-ratio "$RANDOM_RANGE_RATIO" \
+    --num-prompts "$((CONC * 10))" \
+    --max-concurrency "$CONC" \
+    --result-filename "$RESULT_FILENAME" \
+    --result-dir /workspace/ \
+    --trust-remote-code
+
+# After throughput, run evaluation only if RUN_EVAL is true
+if [ "${RUN_EVAL}" = "true" ]; then
+    run_eval --framework lm-eval --port "$PORT" 
+    append_lm_eval_summary
+fi
+
+# Stop GPU monitoring
+stop_gpu_monitor
+set +x
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index ddc6409c2..17b3db7f8 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -1733,3 +1733,12 @@
     - "TP=2 and TP=4, concurrency 4-256 for 1k1k and 8k1k sequence lengths"
     - "Add --max-num-seqs and --gpu-memory-utilization 0.9 to server launch"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1043
+
+- config-keys:
+    - qwen3.5-fp4-mi355x-atom
+  description:
+    - "Add Qwen3.5-397B-A17B MXFP4 single-node MI355X ATOM benchmark"
+    - "Image: rocm/atom:rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.2.post"
+    - "TP=2 and TP=4, concurrency 4-256 for 1k1k and 8k1k sequence lengths"
+    - "Add --gpu-memory-utilization 0.9 to server launch"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/

From bb9d95f5fa88f72bdebdd0466b1fb6f5e11d89ac Mon Sep 17 00:00:00 2001
From: seungrokj <seungrok.jung@amd.com>
Date: Fri, 24 Apr 2026 15:51:43 +0900
Subject: [PATCH 2/2] Add PR link to perf-changelog for qwen3.5 fp4 atom config

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---
 perf-changelog.yaml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 17b3db7f8..b5510055c 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -1741,4 +1741,4 @@
     - "Image: rocm/atom:rocm7.2.2_ubuntu24.04_py3.12_pytorch_release_2.10.0_atom0.1.2.post"
     - "TP=2 and TP=4, concurrency 4-256 for 1k1k and 8k1k sequence lengths"
     - "Add --gpu-memory-utilization 0.9 to server launch"
-  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1133