From 00277ceaa7ced729180d5f697878f12082af7a12 Mon Sep 17 00:00:00 2001
From: Honglie Yi <hyi@crsuse2-m2m-185.us-east2-a.compute.internal>
Date: Wed, 19 Aug 2026 04:22:25 +0000
Subject: [PATCH 01/13] feat: add DeepSeek V4 ATOM AgentX MTP recipe
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Add the MI355X TP8 AgentX sweep for DeepSeek-V4-Pro on ATOM with the validated FP8 cache, prefix-caching, checkpointing, and three-token MTP settings. Use golden synthetic acceptance for throughput and real MTP acceptance for evaluation.

新增 DeepSeek-V4-Pro 在 MI355X TP8 ATOM 上的 AgentX sweep，采用已验证的 FP8 缓存、前缀缓存、状态检查点和三 token MTP 配置。吞吐测试使用 golden synthetic acceptance，评测使用真实 MTP acceptance。
---
 .../agentic/dsv4_fp4_mi355x_atom_mtp.sh       | 131 ++++++++++++++++++
 configs/amd-master.yaml                       |  16 +++
 perf-changelog.yaml                           |  10 ++
 3 files changed, 157 insertions(+)
 create mode 100755 benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh

diff --git a/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh b/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh
new file mode 100755
index 0000000000..cbf4bc253f
--- /dev/null
+++ b/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh
@@ -0,0 +1,131 @@
+#!/usr/bin/env bash
+set -euo pipefail
+set -x
+
+# Agentic trace replay benchmark for DeepSeek-V4-Pro FP4 on MI355X using
+# ATOM MTP. Throughput runs use the committed golden synthetic acceptance;
+# eval-only runs use the model's real MTP acceptance.
+
+source "$(dirname "$0")/../../benchmark_lib.sh"
+
+check_env_vars MODEL TP CONC KV_OFFLOADING TOTAL_CPU_DRAM_GB RESULT_DIR DURATION EP_SIZE DP_ATTENTION
+
+if [[ -n "${SLURM_JOB_ID:-}" ]]; then
+    echo "JOB $SLURM_JOB_ID running on ${SLURMD_NODENAME:-unknown}"
+fi
+
+if [ "$TP" -ne 8 ] || [ "$EP_SIZE" -ne 1 ] || [ "$DP_ATTENTION" != "false" ]; then
+    echo "This recipe requires TP=8, EP_SIZE=1, and DP_ATTENTION=false" >&2
+    exit 1
+fi
+require_agentic_kv_offload_none
+
+if [[ -n "${ROCR_VISIBLE_DEVICES:-}" ]]; then
+    export HIP_VISIBLE_DEVICES="$ROCR_VISIBLE_DEVICES"
+fi
+
+if [[ -n "${MODEL_PATH:-}" ]]; then
+    if [[ ! -d "$MODEL_PATH" || -z "$(ls -A "$MODEL_PATH" 2>/dev/null)" ]]; then
+        hf download "$MODEL" --local-dir "$MODEL_PATH"
+    fi
+else
+    hf download "$MODEL"
+    export MODEL_PATH="$MODEL"
+fi
+
+rocm-smi || true
+amd-smi || true
+
+resolve_trace_source
+install_agentic_deps
+
+# ATOM runtime settings validated with the DeepSeek-V4-Pro AgentX baseline.
+export AITER_BF16_FP8_MOE_BOUND=0
+export AITER_LOG_LEVEL=WARNING
+export ATOM_MOE_GU_ITLV=1
+export ATOM_DISABLE_MMAP=true
+export ATOM_DEBUG_PREFIX_HITS=1
+export ATOM_PROFILER_MORE=0
+export ATOM_PROFILER_TIMEOUT=1200
+
+# AgentX/AIPerf network, failure, warmup, and trace-gap settings from the
+# validated one-hour baseline.
+export AIPERF_HTTP_TCP_USER_TIMEOUT=900000
+export AIPERF_FAILED_REQUEST_THRESHOLD=0.10
+export AIPERF_LIVE_FAILED_REQUEST_THRESHOLD=0.10
+export AIPERF_TRACE_IDLE_GAP_CAP_SECONDS=300
+export AIPERF_WARMUP_REQUESTS_PER_LANE=10
+export AIPERF_BENCHMARK_GRACE_PERIOD=30
+
+# Require ATOM Prometheus metrics in every official result.
+export AIPERF_SERVER_METRICS_URLS="http://localhost:${PORT}/metrics"
+export AIPERF_REQUIRED_SERVER_METRIC_PREFIX="atom:"
+
+wait_for_amd_gpu_clean
+
+SERVER_LOG="$RESULT_DIR/server.log"
+mkdir -p "$RESULT_DIR"
+
+SERVER_PID=""
+cleanup_atom_server() {
+    local exit_code=$?
+    trap - EXIT INT TERM
+    set +e
+    stop_background_process_tree "$SERVER_PID" "ATOM server" 60
+    exit "$exit_code"
+}
+trap cleanup_atom_server EXIT
+trap 'exit 130' INT
+trap 'exit 143' TERM
+
+# AgentX concurrency counts session trees. Keep 2x scheduler headroom for the
+# request bursts produced by subagent fan-out.
+MAX_NUM_SEQS=$((2 * CONC))
+
+# golden_al_distribution/dsv4_mtp.yaml: thinking_on, 3 draft tokens -> AL 2.49
+# acceptance rate = (2.49 - 1) / 3 = 0.4966666667.
+NUM_SPEC_TOKENS=3
+SPEC_ACCEPTANCE_RATE=0.4966666667
+SPEC_ARGS=(
+    --method mtp
+    --num-speculative-tokens "$NUM_SPEC_TOKENS"
+)
+if [ "${EVAL_ONLY:-false}" != "true" ]; then
+    SPEC_ARGS+=(--spec-decode-acceptance-rate "$SPEC_ACCEPTANCE_RATE")
+fi
+
+echo "Starting ATOM server with MAX_NUM_SEQS=$MAX_NUM_SEQS NUM_SPEC_TOKENS=$NUM_SPEC_TOKENS SPEC_ACCEPTANCE_RATE=$SPEC_ACCEPTANCE_RATE EVAL_ONLY=${EVAL_ONLY:-false}"
+ATOM_CMD=(
+    python3 -u -m atom.entrypoints.openai_server
+    --model "$MODEL_PATH"
+    --served-model-name "$MODEL"
+    --host 0.0.0.0
+    --server-port "$PORT"
+    --tensor-parallel-size "$TP"
+    --kv-cache-dtype fp8
+    --index-cache-dtype fp8
+    --enable-prefix-caching
+    --gpu-memory-utilization 0.9
+    --max-num-batched-tokens 16384
+    --attn-prefill-chunk-size 16384
+    --state-checkpoint-interval-tokens 32768
+    --level 3
+    --cudagraph-mode FULL
+    "${SPEC_ARGS[@]}"
+    --max-num-seqs "$MAX_NUM_SEQS"
+)
+write_command "$RESULT_DIR/server_command.txt" "${ATOM_CMD[@]}"
+"${ATOM_CMD[@]}" > "$SERVER_LOG" 2>&1 &
+SERVER_PID=$!
+echo "Server PID: $SERVER_PID"
+
+wait_for_server_ready --port "$PORT" --server-log "$SERVER_LOG" --server-pid "$SERVER_PID"
+
+if [ "${EVAL_ONLY:-false}" = "true" ]; then
+    run_eval --port "$PORT"
+else
+    # AgentX DSv4 traces already carry fully formed chat payloads; do not apply
+    # AIPerf's generic chat template on top of them.
+    build_replay_cmd "$RESULT_DIR"
+    run_agentic_replay_and_write_outputs "$RESULT_DIR"
+fi
diff --git a/configs/amd-master.yaml b/configs/amd-master.yaml
index cf5cdfeb6d..0c48bd0197 100644
--- a/configs/amd-master.yaml
+++ b/configs/amd-master.yaml
@@ -1284,6 +1284,22 @@ dsv4-fp4-mi355x-vllm-agentic-mtp:
       # while MTP creates two. Restore these points after the upstream hybrid
       # KV recovery fix lands: https://github.com/vllm-project/vllm/pull/45497
 
+# DeepSeek-V4-Pro FP4 AgentX on one MI355X node using ATOM MTP. Throughput
+# uses the thinking_on golden AL 2.49 for three draft tokens; eval uses real
+# MTP acceptance. max-num-seqs is set to 2x concurrency by the recipe.
+dsv4-fp4-mi355x-atom-agentic-mtp:
+  image: rocm/atom-dev:nightly_202608181633
+  model: deepseek-ai/DeepSeek-V4-Pro
+  model-prefix: dsv4
+  runner: cluster:mi355x-amds
+  precision: fp4
+  framework: atom
+  multinode: false
+  scenarios:
+    agentic-coding:
+    - search-space:
+      - { tp: 8, ep: 1, dp-attn: false, kv-offloading: none, spec-decoding: mtp, conc-list: [1, 2, 4, 8, 16, 32, 48] }
+
 dsr1-fp4-mi355x-sglang-disagg-mtp:
   image: lmsysorg/sglang-rocm:v0.5.12-rocm720-mi35x-20260519
   model: amd/DeepSeek-R1-0528-MXFP4-v2
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 6b0d955301..291151844a 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -6171,3 +6171,13 @@
     - "Use AITER INT4 quick-reduce, ptpc_fp8 online quantization excluding embeddings, lm_head, gates, and experts, an FP8 KV cache, and three-token MTP with golden synthetic acceptance length 2.99 for benchmark runs."
     - "Set max-num-seqs to twice the concurrency, select CUDA graph capture sizes by concurrency, and cap max-num-batched-tokens at 16384."
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2576
+
+- config-keys:
+    - dsv4-fp4-mi355x-atom-agentic-mtp
+  scenario-type:
+    - agentic-coding
+  description:
+    - "Add DeepSeek-V4-Pro FP4 ATOM MTP AgentX on one 8x MI355X node at concurrency 1, 2, 4, 8, 16, 32, and 48, with no KV offload and max-num-seqs set to twice concurrency."
+    - "Use rocm/atom-dev:nightly_202608181633 (digest sha256:fdc5650f2b6d13c22f1f1b3873ee6dc61278e6b0b90dc24d0f6c55810895032f), FP8 KV/index caches, prefix caching, 32K state checkpoints, 16K batching/prefill chunks, and FULL cudagraph mode."
+    - "Use three-token MTP with the committed DeepSeek-V4 thinking-mode golden AL 2.49 (synthetic acceptance rate 0.4966666667) for throughput, while eval-only runs measure real MTP acceptance."
+  pr-link: TBD

From 89cac1dffa02c10b5f2536da2008109a269bd575 Mon Sep 17 00:00:00 2001
From: Honglie Yi <hyi@crsuse2-m2m-185.us-east2-a.compute.internal>
Date: Wed, 19 Aug 2026 04:24:23 +0000
Subject: [PATCH 02/13] chore: link DeepSeek V4 ATOM AgentX changelog
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Backfill the performance changelog with the submitted InferenceX pull request URL.

在性能变更日志中回填已提交的 InferenceX PR 链接。
---
 perf-changelog.yaml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 291151844a..48d7dc64f8 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -6180,4 +6180,4 @@
     - "Add DeepSeek-V4-Pro FP4 ATOM MTP AgentX on one 8x MI355X node at concurrency 1, 2, 4, 8, 16, 32, and 48, with no KV offload and max-num-seqs set to twice concurrency."
     - "Use rocm/atom-dev:nightly_202608181633 (digest sha256:fdc5650f2b6d13c22f1f1b3873ee6dc61278e6b0b90dc24d0f6c55810895032f), FP8 KV/index caches, prefix caching, 32K state checkpoints, 16K batching/prefill chunks, and FULL cudagraph mode."
     - "Use three-token MTP with the committed DeepSeek-V4 thinking-mode golden AL 2.49 (synthetic acceptance rate 0.4966666667) for throughput, while eval-only runs measure real MTP acceptance."
-  pr-link: TBD
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2668

From 9c755e6d1ace0a199be70b77e132930c610ebc2f Mon Sep 17 00:00:00 2001
From: Honglie Yi <hyi@crsuse2-m2m-271.us-east2-a.compute.internal>
Date: Wed, 19 Aug 2026 10:07:25 +0000
Subject: [PATCH 03/13] fix(agentx): extend ATOM saturation warmup
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Use a 3600-second warmup grace period for the ATOM C32 and C48 saturation arms while keeping C1 through C16 at the shared 1800-second default.\n\n中文：将 ATOM C32 和 C48 饱和并发点的 warmup grace period 调整为 3600 秒，C1 至 C16 继续使用共享的 1800 秒默认值。
---
 .../single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh      | 6 ++++++
 perf-changelog.yaml                                      | 9 +++++++++
 2 files changed, 15 insertions(+)

diff --git a/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh b/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh
index cbf4bc253f..6d1197362c 100755
--- a/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh
+++ b/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh
@@ -82,6 +82,12 @@ trap 'exit 143' TERM
 # request bursts produced by subagent fan-out.
 MAX_NUM_SEQS=$((2 * CONC))
 
+# Saturation arms carry a larger in-flight working set than the 30-minute
+# default warmup drain allows.
+if [ "$CONC" -ge 32 ]; then
+    export AGENTIC_WARMUP_GRACE_PERIOD=3600
+fi
+
 # golden_al_distribution/dsv4_mtp.yaml: thinking_on, 3 draft tokens -> AL 2.49
 # acceptance rate = (2.49 - 1) / 3 = 0.4966666667.
 NUM_SPEC_TOKENS=3
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index d0aadc8983..6331366bd0 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -6205,3 +6205,12 @@
     - "Use rocm/atom-dev:nightly_202608181633 (digest sha256:fdc5650f2b6d13c22f1f1b3873ee6dc61278e6b0b90dc24d0f6c55810895032f), FP8 KV/index caches, prefix caching, 32K state checkpoints, 16K batching/prefill chunks, and FULL cudagraph mode."
     - "Use three-token MTP with the committed DeepSeek-V4 thinking-mode golden AL 2.49 (synthetic acceptance rate 0.4966666667) for throughput, while eval-only runs measure real MTP acceptance."
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2668
+
+- config-keys:
+    - dsv4-fp4-mi355x-atom-agentic-mtp
+  scenario-type:
+    - agentic-coding
+  description:
+    - "Increase the AgentX warmup grace period from the 1800-second default to 3600 seconds at concurrency 32 and 48, matching the DeepSeek-V4-Pro SGLang saturation recipe so in-flight warmup requests can drain and reused long prefixes are fully primed before profiling."
+    - "Keep concurrency 1 through 16 at the default 1800 seconds; the profiling duration remains 3600 seconds for every concurrency arm."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2668

From ee37e1804f1c1ff711cac78141f932b1f98f372a Mon Sep 17 00:00:00 2001
From: seungrokj <144636725+seungrokj@users.noreply.github.com>
Date: Wed, 19 Aug 2026 20:02:52 +0900
Subject: [PATCH 04/13] Update dsv4_fp4_mi355x_atom_mtp.sh

---
 benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh b/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh
index 6d1197362c..f8169be4e4 100755
--- a/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh
+++ b/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh
@@ -112,9 +112,9 @@ ATOM_CMD=(
     --index-cache-dtype fp8
     --enable-prefix-caching
     --gpu-memory-utilization 0.9
-    --max-num-batched-tokens 16384
-    --attn-prefill-chunk-size 16384
-    --state-checkpoint-interval-tokens 32768
+    --max-num-batched-tokens 8192
+    --attn-prefill-chunk-size 8192
+    --state-checkpoint-interval-tokens 8192
     --level 3
     --cudagraph-mode FULL
     "${SPEC_ARGS[@]}"

From 097b9ae5424efcffd7dff1f10dad998aa67a6f08 Mon Sep 17 00:00:00 2001
From: seungrokj <144636725+seungrokj@users.noreply.github.com>
Date: Wed, 19 Aug 2026 20:03:38 +0900
Subject: [PATCH 05/13] Update amd-master.yaml

---
 configs/amd-master.yaml | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/configs/amd-master.yaml b/configs/amd-master.yaml
index 0c48bd0197..8e5db0957a 100644
--- a/configs/amd-master.yaml
+++ b/configs/amd-master.yaml
@@ -1298,7 +1298,8 @@ dsv4-fp4-mi355x-atom-agentic-mtp:
   scenarios:
     agentic-coding:
     - search-space:
-      - { tp: 8, ep: 1, dp-attn: false, kv-offloading: none, spec-decoding: mtp, conc-list: [1, 2, 4, 8, 16, 32, 48] }
+      #- { tp: 8, ep: 1, dp-attn: false, kv-offloading: none, spec-decoding: mtp, conc-list: [1, 2, 4, 8, 16, 32, 48] }
+      - { tp: 8, ep: 1, dp-attn: false, kv-offloading: none, spec-decoding: mtp, conc-list: [48] }
 
 dsr1-fp4-mi355x-sglang-disagg-mtp:
   image: lmsysorg/sglang-rocm:v0.5.12-rocm720-mi35x-20260519

From 9ff0a4304797a7ef5761f831245133e238b5442a Mon Sep 17 00:00:00 2001
From: seungrokj <144636725+seungrokj@users.noreply.github.com>
Date: Wed, 19 Aug 2026 20:10:37 +0900
Subject: [PATCH 06/13] Update perf-changelog.yaml

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

diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 6331366bd0..1281c43911 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -6204,13 +6204,6 @@
     - "Add DeepSeek-V4-Pro FP4 ATOM MTP AgentX on one 8x MI355X node at concurrency 1, 2, 4, 8, 16, 32, and 48, with no KV offload and max-num-seqs set to twice concurrency."
     - "Use rocm/atom-dev:nightly_202608181633 (digest sha256:fdc5650f2b6d13c22f1f1b3873ee6dc61278e6b0b90dc24d0f6c55810895032f), FP8 KV/index caches, prefix caching, 32K state checkpoints, 16K batching/prefill chunks, and FULL cudagraph mode."
     - "Use three-token MTP with the committed DeepSeek-V4 thinking-mode golden AL 2.49 (synthetic acceptance rate 0.4966666667) for throughput, while eval-only runs measure real MTP acceptance."
-  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2668
-
-- config-keys:
-    - dsv4-fp4-mi355x-atom-agentic-mtp
-  scenario-type:
-    - agentic-coding
-  description:
     - "Increase the AgentX warmup grace period from the 1800-second default to 3600 seconds at concurrency 32 and 48, matching the DeepSeek-V4-Pro SGLang saturation recipe so in-flight warmup requests can drain and reused long prefixes are fully primed before profiling."
     - "Keep concurrency 1 through 16 at the default 1800 seconds; the profiling duration remains 3600 seconds for every concurrency arm."
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2668

From b09f6caaf46bd74a3e7f63ac2821a6c7d786a27d Mon Sep 17 00:00:00 2001
From: seungrokj <144636725+seungrokj@users.noreply.github.com>
Date: Wed, 19 Aug 2026 21:43:11 +0900
Subject: [PATCH 07/13] Update dsv4_fp4_mi355x_atom_mtp.sh

---
 benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh b/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh
index f8169be4e4..6d1197362c 100755
--- a/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh
+++ b/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh
@@ -112,9 +112,9 @@ ATOM_CMD=(
     --index-cache-dtype fp8
     --enable-prefix-caching
     --gpu-memory-utilization 0.9
-    --max-num-batched-tokens 8192
-    --attn-prefill-chunk-size 8192
-    --state-checkpoint-interval-tokens 8192
+    --max-num-batched-tokens 16384
+    --attn-prefill-chunk-size 16384
+    --state-checkpoint-interval-tokens 32768
     --level 3
     --cudagraph-mode FULL
     "${SPEC_ARGS[@]}"

From 0f467e11df32642eab60a495736374edd628aa63 Mon Sep 17 00:00:00 2001
From: seungrokj <144636725+seungrokj@users.noreply.github.com>
Date: Wed, 19 Aug 2026 21:43:46 +0900
Subject: [PATCH 08/13] Update amd-master.yaml

---
 configs/amd-master.yaml | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/configs/amd-master.yaml b/configs/amd-master.yaml
index 8e5db0957a..0c48bd0197 100644
--- a/configs/amd-master.yaml
+++ b/configs/amd-master.yaml
@@ -1298,8 +1298,7 @@ dsv4-fp4-mi355x-atom-agentic-mtp:
   scenarios:
     agentic-coding:
     - search-space:
-      #- { tp: 8, ep: 1, dp-attn: false, kv-offloading: none, spec-decoding: mtp, conc-list: [1, 2, 4, 8, 16, 32, 48] }
-      - { tp: 8, ep: 1, dp-attn: false, kv-offloading: none, spec-decoding: mtp, conc-list: [48] }
+      - { tp: 8, ep: 1, dp-attn: false, kv-offloading: none, spec-decoding: mtp, conc-list: [1, 2, 4, 8, 16, 32, 48] }
 
 dsr1-fp4-mi355x-sglang-disagg-mtp:
   image: lmsysorg/sglang-rocm:v0.5.12-rocm720-mi35x-20260519

From 013df3dbed341f1385a51294bcd41a1ca9e226da Mon Sep 17 00:00:00 2001
From: seungrokj <144636725+seungrokj@users.noreply.github.com>
Date: Thu, 20 Aug 2026 09:48:37 +0900
Subject: [PATCH 09/13] Update amd-master.yaml

---
 configs/amd-master.yaml | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/configs/amd-master.yaml b/configs/amd-master.yaml
index 0c48bd0197..8e5db0957a 100644
--- a/configs/amd-master.yaml
+++ b/configs/amd-master.yaml
@@ -1298,7 +1298,8 @@ dsv4-fp4-mi355x-atom-agentic-mtp:
   scenarios:
     agentic-coding:
     - search-space:
-      - { tp: 8, ep: 1, dp-attn: false, kv-offloading: none, spec-decoding: mtp, conc-list: [1, 2, 4, 8, 16, 32, 48] }
+      #- { tp: 8, ep: 1, dp-attn: false, kv-offloading: none, spec-decoding: mtp, conc-list: [1, 2, 4, 8, 16, 32, 48] }
+      - { tp: 8, ep: 1, dp-attn: false, kv-offloading: none, spec-decoding: mtp, conc-list: [48] }
 
 dsr1-fp4-mi355x-sglang-disagg-mtp:
   image: lmsysorg/sglang-rocm:v0.5.12-rocm720-mi35x-20260519

From 486becce06c8391c0edab8e1280e682c07e50610 Mon Sep 17 00:00:00 2001
From: seungrokj <144636725+seungrokj@users.noreply.github.com>
Date: Thu, 20 Aug 2026 09:49:14 +0900
Subject: [PATCH 10/13] Update dsv4_fp4_mi355x_atom_mtp.sh

---
 .../single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh       | 8 +-------
 1 file changed, 1 insertion(+), 7 deletions(-)

diff --git a/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh b/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh
index 6d1197362c..29adf7950c 100755
--- a/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh
+++ b/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh
@@ -82,12 +82,6 @@ trap 'exit 143' TERM
 # request bursts produced by subagent fan-out.
 MAX_NUM_SEQS=$((2 * CONC))
 
-# Saturation arms carry a larger in-flight working set than the 30-minute
-# default warmup drain allows.
-if [ "$CONC" -ge 32 ]; then
-    export AGENTIC_WARMUP_GRACE_PERIOD=3600
-fi
-
 # golden_al_distribution/dsv4_mtp.yaml: thinking_on, 3 draft tokens -> AL 2.49
 # acceptance rate = (2.49 - 1) / 3 = 0.4966666667.
 NUM_SPEC_TOKENS=3
@@ -114,7 +108,7 @@ ATOM_CMD=(
     --gpu-memory-utilization 0.9
     --max-num-batched-tokens 16384
     --attn-prefill-chunk-size 16384
-    --state-checkpoint-interval-tokens 32768
+    --state-checkpoint-interval-tokens 8192
     --level 3
     --cudagraph-mode FULL
     "${SPEC_ARGS[@]}"

From 9b32a5443a1ec06567a5fab21a40db310ff03262 Mon Sep 17 00:00:00 2001
From: seungrokj <144636725+seungrokj@users.noreply.github.com>
Date: Thu, 20 Aug 2026 12:19:32 +0900
Subject: [PATCH 11/13] Update amd-master.yaml

---
 configs/amd-master.yaml | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/configs/amd-master.yaml b/configs/amd-master.yaml
index 8e5db0957a..0c48bd0197 100644
--- a/configs/amd-master.yaml
+++ b/configs/amd-master.yaml
@@ -1298,8 +1298,7 @@ dsv4-fp4-mi355x-atom-agentic-mtp:
   scenarios:
     agentic-coding:
     - search-space:
-      #- { tp: 8, ep: 1, dp-attn: false, kv-offloading: none, spec-decoding: mtp, conc-list: [1, 2, 4, 8, 16, 32, 48] }
-      - { tp: 8, ep: 1, dp-attn: false, kv-offloading: none, spec-decoding: mtp, conc-list: [48] }
+      - { tp: 8, ep: 1, dp-attn: false, kv-offloading: none, spec-decoding: mtp, conc-list: [1, 2, 4, 8, 16, 32, 48] }
 
 dsr1-fp4-mi355x-sglang-disagg-mtp:
   image: lmsysorg/sglang-rocm:v0.5.12-rocm720-mi35x-20260519

From a93da29ea2f21958064a5dcfd83ae5ccc1508f62 Mon Sep 17 00:00:00 2001
From: seungrokj <144636725+seungrokj@users.noreply.github.com>
Date: Thu, 20 Aug 2026 15:14:45 +0900
Subject: [PATCH 12/13] Update dsv4_fp4_mi355x_atom_mtp.sh

---
 benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh b/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh
index 29adf7950c..825778f836 100755
--- a/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh
+++ b/benchmarks/single_node/agentic/dsv4_fp4_mi355x_atom_mtp.sh
@@ -103,7 +103,7 @@ ATOM_CMD=(
     --server-port "$PORT"
     --tensor-parallel-size "$TP"
     --kv-cache-dtype fp8
-    --index-cache-dtype fp8
+    --index-cache-dtype fp4
     --enable-prefix-caching
     --gpu-memory-utilization 0.9
     --max-num-batched-tokens 16384

From 109dafc5f3a05b86e3b1bd6486cdc26070f6d644 Mon Sep 17 00:00:00 2001
From: seungrokj <144636725+seungrokj@users.noreply.github.com>
Date: Thu, 20 Aug 2026 19:59:42 +0900
Subject: [PATCH 13/13] Update amd-master.yaml

---
 configs/amd-master.yaml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/configs/amd-master.yaml b/configs/amd-master.yaml
index 0bc948d5d6..10604c32bd 100644
--- a/configs/amd-master.yaml
+++ b/configs/amd-master.yaml
@@ -1288,7 +1288,7 @@ dsv4-fp4-mi355x-vllm-agentic-mtp:
 # uses the thinking_on golden AL 2.49 for three draft tokens; eval uses real
 # MTP acceptance. max-num-seqs is set to 2x concurrency by the recipe.
 dsv4-fp4-mi355x-atom-agentic-mtp:
-  image: rocm/atom-dev:nightly_202608181633
+  image: rocm/atom-dev:nightly_202608201032
   model: deepseek-ai/DeepSeek-V4-Pro
   model-prefix: dsv4
   runner: cluster:mi355x-amds