diff --git a/.github/configs/amd-master.yaml b/.github/configs/amd-master.yaml
index c396e678..55651efe 100644
--- a/.github/configs/amd-master.yaml
+++ b/.github/configs/amd-master.yaml
@@ -176,3 +176,73 @@ gptoss-fp4-mi355x-vllm:
     - { tp: 1, conc-start: 4, conc-end: 128 }
     - { tp: 4, conc-start: 4, conc-end: 4 }
     - { tp: 8, conc-start: 4, conc-end: 8 }
+
+dsr1-fp4-mi355x-atom:
+  image: rocm/atom:rocm7.1.1-ubuntu24.04-pytorch2.9-atom0.1.1-MI350x
+  model: amd/DeepSeek-R1-0528-MXFP4-Preview
+  model-prefix: dsr1
+  runner: mi355x-m15-17
+  precision: fp4
+  framework: atom
+  multinode: false
+  seq-len-configs:
+  - isl: 1024
+    osl: 1024
+    search-space:
+    - { tp: 4, conc-start: 4, conc-end: 128 }
+    - { tp: 8, conc-start: 4, conc-end: 128 }
+  - isl: 1024
+    osl: 8192
+    search-space:
+    - { tp: 8, conc-start: 4, conc-end: 128 }
+  - isl: 8192
+    osl: 1024
+    search-space:
+    - { tp: 8, conc-start: 4, conc-end: 128 }
+
+dsr1-fp8-mi355x-atom:
+  image: rocm/atom:rocm7.1.1-ubuntu24.04-pytorch2.9-atom0.1.1-MI350x
+  model: deepseek-ai/DeepSeek-R1-0528
+  model-prefix: dsr1
+  runner: mi355x-m15-17
+  precision: fp8
+  framework: atom
+  multinode: false
+  seq-len-configs:
+  - isl: 1024
+    osl: 1024
+    search-space:
+    - { tp: 8, conc-start: 4, conc-end: 128 }
+  - isl: 1024
+    osl: 8192
+    search-space:
+    - { tp: 8, conc-start: 4, conc-end: 128 }
+  - isl: 8192
+    osl: 1024
+    search-space:
+    - { tp: 8, conc-start: 4, conc-end: 128 }
+
+gptoss-fp4-mi355x-atom:
+  image: rocm/atom:rocm7.1.1-ubuntu24.04-pytorch2.9-atom0.1.1-MI350x
+  model: openai/gpt-oss-120b
+  model-prefix: gptoss
+  runner: mi355x-m15-17
+  precision: fp4
+  framework: atom
+  multinode: false
+  seq-len-configs:
+  - isl: 1024
+    osl: 1024
+    search-space:
+    - { tp: 1, conc-start: 4, conc-end: 128 }
+    - { tp: 8, conc-start: 4, conc-end: 128 }
+  - isl: 1024
+    osl: 8192
+    search-space:
+    - { tp: 1, conc-start: 4, conc-end: 128 }
+    - { tp: 8, conc-start: 4, conc-end: 128 }
+  - isl: 8192
+    osl: 1024
+    search-space:
+    - { tp: 1, conc-start: 4, conc-end: 128 }
+    - { tp: 8, conc-start: 4, conc-end: 128 }
diff --git a/.github/configs/runners.yaml b/.github/configs/runners.yaml
index e414af03..88c47bd3 100644
--- a/.github/configs/runners.yaml
+++ b/.github/configs/runners.yaml
@@ -53,5 +53,8 @@ mi355x:
 - 'mi355x-amd_1'
 - 'mi355x-amd_2'
 - 'mi355x-amd_3'
+mi355x-m15-17:
+- 'mi355x-amdatom'
+- 'mi355x-amdatomtw'
 gb200:
 - gb200-nv_0
diff --git a/.github/workflows/benchmark-tmpl.yml b/.github/workflows/benchmark-tmpl.yml
index 27ebd215..963366ab 100644
--- a/.github/workflows/benchmark-tmpl.yml
+++ b/.github/workflows/benchmark-tmpl.yml
@@ -120,13 +120,10 @@ jobs:
                 # nvidia-smi --gpu-reset -i 0,1,2,3,4,5,6,7 2>/dev/null || true
               fi
             else
-              echo "[Docker] Cleaning up resources ..."
-              docker ps -aq | xargs -r docker rm -f
-              docker network prune -f
-              while [ -n "$(docker ps -aq)" ]; do
-                docker ps -a
-                sleep 5
-              done
+              echo "[Docker] Cleaning up bmk-server container ..."
+              docker stop bmk-server 2>/dev/null || true
+              docker rm -f bmk-server 2>/dev/null || true
+              docker network rm bmk-net 2>/dev/null || true
             fi
           fi
           if command -v squeue >/dev/null 2>&1; then
diff --git a/benchmarks/dsr1_fp4_mi355x_atom_docker.sh b/benchmarks/dsr1_fp4_mi355x_atom_docker.sh
new file mode 100755
index 00000000..105e3539
--- /dev/null
+++ b/benchmarks/dsr1_fp4_mi355x_atom_docker.sh
@@ -0,0 +1,24 @@
+#!/usr/bin/env bash
+
+# ========= Required Env Vars =========
+# HF_TOKEN
+# HF_HUB_CACHE
+# MODEL
+# PORT
+# TP
+# CONC
+# MAX_MODEL_LEN
+
+# Calculate max-model-len based on ISL and OSL
+if [ "$ISL" = "1024" ] && [ "$OSL" = "1024" ]; then
+    CALCULATED_MAX_MODEL_LEN=""
+else
+    CALCULATED_MAX_MODEL_LEN=" --max-model-len 10240 "
+fi
+
+set -x
+python3 -m atom.entrypoints.openai_server \
+    --model $MODEL \
+    --server-port $PORT \
+    -tp $TP \
+    --kv_cache_dtype fp8 $CALCULATED_MAX_MODEL_LEN
diff --git a/benchmarks/dsr1_fp8_mi355x_atom_docker.sh b/benchmarks/dsr1_fp8_mi355x_atom_docker.sh
new file mode 100755
index 00000000..032291b9
--- /dev/null
+++ b/benchmarks/dsr1_fp8_mi355x_atom_docker.sh
@@ -0,0 +1,27 @@
+#!/usr/bin/env bash
+
+# ========= Required Env Vars =========
+# HF_TOKEN
+# HF_HUB_CACHE
+# MODEL
+# PORT
+# TP
+# CONC
+# MAX_MODEL_LEN
+
+# Calculate max-model-len based on ISL and OSL
+if [ "$ISL" = "1024" ] && [ "$OSL" = "1024" ]; then
+    CALCULATED_MAX_MODEL_LEN=""
+else
+    CALCULATED_MAX_MODEL_LEN=" --max-model-len 10240 "
+fi
+
+set -x
+
+BLOCK_SIZE=${BLOCK_SIZE:-16}
+python3 -m atom.entrypoints.openai_server \
+    --model $MODEL \
+    --server-port $PORT \
+    -tp $TP \
+    --kv_cache_dtype fp8 $CALCULATED_MAX_MODEL_LEN \
+    --block-size $BLOCK_SIZE
diff --git a/benchmarks/gptoss_fp4_mi355x_atom_docker.sh b/benchmarks/gptoss_fp4_mi355x_atom_docker.sh
new file mode 100755
index 00000000..4fc5d21c
--- /dev/null
+++ b/benchmarks/gptoss_fp4_mi355x_atom_docker.sh
@@ -0,0 +1,25 @@
+#!/usr/bin/env bash
+
+# ========= Required Env Vars =========
+# HF_TOKEN
+# HF_HUB_CACHE
+# MODEL
+# PORT
+# TP
+# CONC
+# MAX_MODEL_LEN
+
+# Calculate max-model-len based on ISL and OSL
+if [ "$ISL" = "1024" ] && [ "$OSL" = "1024" ]; then
+    CALCULATED_MAX_MODEL_LEN=""
+else
+    CALCULATED_MAX_MODEL_LEN=" --max-model-len 10240 "
+fi
+
+set -x
+export HSA_NO_SCRATCH_RECLAIM=1
+python3 -m atom.entrypoints.openai_server \
+    --model $MODEL \
+    --server-port $PORT \
+    -tp $TP \
+    --kv_cache_dtype fp8 $CALCULATED_MAX_MODEL_LEN
diff --git a/runners/launch_mi355x-amdatom.sh b/runners/launch_mi355x-amdatom.sh
new file mode 100755
index 00000000..cae55690
--- /dev/null
+++ b/runners/launch_mi355x-amdatom.sh
@@ -0,0 +1,110 @@
+#!/usr/bin/env bash
+
+# === Workflow-defined Env Vars ===
+# IMAGE
+# MODEL
+# TP
+# HF_HUB_CACHE
+# ISL
+# OSL
+# MAX_MODEL_LEN
+# RANDOM_RANGE_RATIO
+# CONC
+# GITHUB_WORKSPACE
+# RESULT_FILENAME
+# HF_TOKEN
+# FRAMEWORK
+
+HF_HUB_CACHE_MOUNT="/mnt/hf_hub_cache/"  # Temp solution
+PORT=8888
+
+# Determine framework suffix for benchmark script
+FRAMEWORK_SUFFIX=$([[ "$FRAMEWORK" == "atom" ]] && printf '_atom' || printf '')
+
+network_name="bmk-net"
+server_name="bmk-server"
+client_name="bmk-client"
+
+# Cleanup: stop server container and remove network
+docker stop $server_name 2>/dev/null || true
+docker rm $server_name 2>/dev/null || true
+docker network rm $network_name 2>/dev/null || true
+
+docker network create $network_name
+
+set -x
+docker pull $IMAGE
+DIGEST=$(docker inspect --format='{{index .RepoDigests 0}}' "$IMAGE" | cut -d'@' -f2)
+echo "The image digest is: $DIGEST"
+
+set -x
+docker run --rm -d --ipc=host --shm-size=16g --network=$network_name --name=$server_name \
+--privileged --cap-add=CAP_SYS_ADMIN --device=/dev/kfd --device=/dev/dri --device=/dev/mem \
+--cap-add=SYS_PTRACE --security-opt seccomp=unconfined \
+-v $HF_HUB_CACHE_MOUNT:$HF_HUB_CACHE \
+-v $GITHUB_WORKSPACE:/workspace/ -w /workspace/ \
+-e HF_TOKEN -e HF_HUB_CACHE -e MODEL -e TP -e CONC -e MAX_MODEL_LEN -e PORT=$PORT \
+-e ISL -e OSL \
+--entrypoint=/bin/bash \
+$IMAGE \
+benchmarks/"${EXP_NAME%%_*}_${PRECISION}_mi355x${FRAMEWORK_SUFFIX}_docker.sh"
+
+set +x
+while IFS= read -r line; do
+    printf '%s\n' "$line"
+    if [[ "$line" =~ Application\ startup\ complete ]]; then
+        break
+    fi
+done < <(docker logs -f --tail=0 $server_name 2>&1)
+
+if [[ "$MODEL" == "amd/DeepSeek-R1-0528-MXFP4-Preview" || "$MODEL" == "deepseek-ai/DeepSeek-R1-0528" ]]; then
+  if [[ "$OSL" == "8192" ]]; then
+    #NUM_PROMPTS=$(( CONC * 20 ))
+    NUM_PROMPTS=$(( CONC * 2 )) # atom has no much compilation overhead for dsr1
+  else
+    #NUM_PROMPTS=$(( CONC * 50 ))
+    NUM_PROMPTS=$(( CONC * 10 )) # atom has no much compilation overhead for dsr1
+  fi
+else
+  if [[ "$OSL" == "8192" ]]; then
+    NUM_PROMPTS=$(( CONC * 2 ))
+  else
+    NUM_PROMPTS=$(( CONC * 10 ))
+  fi
+fi
+
+set -x
+echo $GITHUB_WORKSPACE
+git clone https://github.com/kimbochen/bench_serving.git
+git clone https://github.com/kimbochen/bench_serving.git $GITHUB_WORKSPACE/bench_serving
+
+sleep 5
+
+set -x
+docker run --rm --network=$network_name --name=$client_name \
+-v $GITHUB_WORKSPACE:/workspace/ -w /workspace/ \
+-e HF_TOKEN -e PYTHONPYCACHEPREFIX=/tmp/pycache/ \
+--entrypoint=/bin/bash \
+$(echo "$IMAGE" | sed 's/#/\//') \
+-lc "pip install -q datasets pandas && \
+python3 bench_serving/benchmark_serving.py \
+--model=$MODEL --backend=vllm --base-url="http://$server_name:$PORT" \
+--dataset-name=random \
+--random-input-len=$ISL --random-output-len=$OSL --random-range-ratio=$RANDOM_RANGE_RATIO \
+--num-prompts=$NUM_PROMPTS \
+--max-concurrency=$CONC \
+--trust-remote-code \
+--request-rate=inf --ignore-eos \
+--save-result --percentile-metrics="ttft,tpot,itl,e2el" \
+--result-dir=/workspace/ --result-filename=$RESULT_FILENAME.json"
+
+if ls gpucore.* 1> /dev/null 2>&1; then
+  echo "gpucore files exist. not good"
+  rm -f gpucore.*
+fi
+
+
+# Cleanup: stop server container and remove network
+docker stop $server_name 2>/dev/null || true
+docker rm $server_name 2>/dev/null || true
+docker network rm $network_name 2>/dev/null || true