From d0fb0170969fb3c5851f770737fa7787de23d2ab Mon Sep 17 00:00:00 2001
From: ChangLiu0709 <cliu1004@amd.com>
Date: Tue, 11 Aug 2026 11:24:30 +0000
Subject: [PATCH 01/10] [AMD] Qwen3.5-FP8 MI355X SGLang disagg: image bump,
 TP4P+TP8D baseline, add MTP config

- Bump qwen3.5-fp8-mi355x-sglang-disagg image from v0.5.14 to v0.5.16
  (3-5% throughput improvement across all concurrency levels)
- Switch baseline from TP8P+TP8D (16 GPU) to TP4P+TP8D (12 GPU): TP4
  prefill delivers identical decode interactivity with 24-31% better
  throughput/GPU
- Add qwen3.5-fp8-mi355x-sglang-disagg-mtp: EAGLE speculative decoding
  (MTP depth 3) on TP4P+TP8D, conc-list [8,16,32,64,128]
- models.yaml: add mtp_flags for Qwen3.5-397B-A17B-FP8
  (--speculative-algorithm EAGLE --speculative-eagle-topk 1)
- Infra: add docker_sg.sh wrapper and update job.slurm DOCKER_CMD_DETECT
  for nodes where Slurm doesn't activate the docker supplementary group

Co-authored-by: Cursor <cursoragent@cursor.com>
---
 benchmarks/multi_node/amd_utils/docker_sg.sh | 20 +++++++
 benchmarks/multi_node/amd_utils/job.slurm    |  6 +-
 benchmarks/multi_node/amd_utils/models.yaml  |  5 +-
 configs/amd-master.yaml                      | 62 ++++++++++++++++----
 perf-changelog.yaml                          | 10 ++++
 5 files changed, 91 insertions(+), 12 deletions(-)
 create mode 100755 benchmarks/multi_node/amd_utils/docker_sg.sh

diff --git a/benchmarks/multi_node/amd_utils/docker_sg.sh b/benchmarks/multi_node/amd_utils/docker_sg.sh
new file mode 100755
index 0000000000..575c6a74a8
--- /dev/null
+++ b/benchmarks/multi_node/amd_utils/docker_sg.sh
@@ -0,0 +1,20 @@
+#!/usr/bin/env bash
+# Run `docker` under the 'docker' group.
+#
+# Why: Slurm launches job steps without the user's 'docker' supplementary group
+# in the active credential set. The user IS a docker-group member (getent group
+# docker lists them) but the group is missing from `id -G` inside the step, so
+# the group-owned socket (/var/run/docker.sock, 0660 root:docker) is unreachable
+# via a plain `docker` call. `sg docker -c` re-activates the group for this one
+# command — no sudo, no password, no persistent host change.
+#
+# Used by job.slurm's DOCKER_CMD detection as the fallback when plain `docker`
+# fails but `sg docker -c 'docker ps'` succeeds.
+#
+# argv is passed across the `sg` shell hop by NUL-delimited base64 (NOT string
+# re-quoting): naive `printf %q` mangles the big multiline `docker run ... bash
+# -lc '<script>'` argument (trailing newline became a literal 'n', spawning a
+# stray `$n`). base64 round-trips arbitrary bytes (newlines, quotes) exactly,
+# and only the base64 blob (safe chars) is interpolated into the sg command.
+b64=$(printf '%s\0' "$@" | base64 | tr -d '\n')
+exec sg docker -c "bash -c 'mapfile -d \"\" -t __A < <(printf %s \"$b64\" | base64 -d); exec docker \"\${__A[@]}\"'"
diff --git a/benchmarks/multi_node/amd_utils/job.slurm b/benchmarks/multi_node/amd_utils/job.slurm
index 3c8a6f16a8..d6ac8df9f3 100755
--- a/benchmarks/multi_node/amd_utils/job.slurm
+++ b/benchmarks/multi_node/amd_utils/job.slurm
@@ -246,7 +246,11 @@ SELECTED_NODELIST_STR=$(echo "$SELECTED_NODES" | tr '\n' ',' | sed 's/,$//')
 
 # Docker privilege detection — evaluated per-node since group membership varies.
 # Exported as a snippet so every srun participant resolves it locally.
-export DOCKER_CMD_DETECT='if docker ps &>/dev/null 2>&1; then DOCKER_CMD=docker; else DOCKER_CMD="sudo docker"; fi'
+# Order: (1) plain docker if the socket is reachable; (2) sg-docker wrapper if the
+# user is a docker-group member but Slurm didn't activate the group in the step's
+# credentials (common here); (3) sudo docker as a last resort.
+SG_DOCKER_WRAPPER="${DI_REPO_DIR}/benchmarks/multi_node/amd_utils/docker_sg.sh"
+export DOCKER_CMD_DETECT="if docker ps &>/dev/null 2>&1; then DOCKER_CMD=docker; elif sg docker -c 'docker ps' &>/dev/null 2>&1; then DOCKER_CMD='${SG_DOCKER_WRAPPER}'; else DOCKER_CMD='sudo docker'; fi"
 
 # Update SLURM environment variables
 export SLURM_NNODES=$NUM_NODES
diff --git a/benchmarks/multi_node/amd_utils/models.yaml b/benchmarks/multi_node/amd_utils/models.yaml
index 2231e6fe60..c3223cbb19 100644
--- a/benchmarks/multi_node/amd_utils/models.yaml
+++ b/benchmarks/multi_node/amd_utils/models.yaml
@@ -202,7 +202,10 @@ Qwen3.5-397B-A17B-MXFP4:
 
 Qwen3.5-397B-A17B-FP8:
   base_flags: "--decode-log-interval 1000 --log-level warning --watchdog-timeout 3600 --load-balance-method round_robin --kv-cache-dtype fp8_e4m3 --attention-backend aiter --disaggregation-transfer-backend mori --moe-dense-tp-size 1"
-  mtp_flags: ""
+  # Qwen3.5 built-in MTP head uses EAGLE (in-checkpoint draft, no draft-model-path);
+  # build_server_config appends --speculative-num-steps=DECODE_MTP_SIZE and
+  # --speculative-num-draft-tokens=DECODE_MTP_SIZE+1.
+  mtp_flags: "--speculative-algorithm EAGLE --speculative-eagle-topk 1"
   dp_flags: "--enable-dp-attention --enable-dp-lm-head"
   ep_flags: "--moe-a2a-backend mori"
   prefill:
diff --git a/configs/amd-master.yaml b/configs/amd-master.yaml
index d26796a52f..8c337fca5a 100644
--- a/configs/amd-master.yaml
+++ b/configs/amd-master.yaml
@@ -262,7 +262,7 @@ qwen3.5-fp8-mi355x-atom-mtp:
       - { tp: 8, ep: 1, conc-start: 4, conc-end: 256, spec-decoding: mtp }
 
 qwen3.5-fp8-mi355x-sglang-disagg:
-  image: lmsysorg/sglang:v0.5.14-rocm720-mi35x
+  image: lmsysorg/sglang:v0.5.16-rocm720-mi35x
   model: Qwen/Qwen3.5-397B-A17B-FP8
   model-prefix: qwen3.5
   runner: mi355x-disagg
@@ -277,21 +277,20 @@ qwen3.5-fp8-mi355x-sglang-disagg:
     - isl: 8192
       osl: 1024
       search-space:
-      # 1P+1D TP8/EP1 low-concurrency sweep.
-      # dp-attn intentionally false (matches the 1k1k row): with
-      # --enable-dp-attention + --moe-a2a-backend mori, sglang auto-promotes
-      # moe_ep_size=tp_size=8, but is_deepep_class_backend() excludes MoRI,
-      # so num_shared_slots stays at the global value (1) and the
+      # 1P+1D TP4P+TP8D/EP1 baseline (no speculative decoding).
+      # TP4 prefill saves 4 GPUs vs TP8P while delivering identical decode
+      # interactivity and 24-31% better throughput/GPU (12 vs 16 GPUs).
+      # dp-attn intentionally false: with --enable-dp-attention +
+      # --moe-a2a-backend mori, sglang auto-promotes moe_ep_size=tp_size,
+      # but is_deepep_class_backend() excludes MoRI, so
+      # num_shared_slots stays at the global value (1) and the
       # (num_experts - num_shared_slots) % moe_ep_size assertion in
       # fused_moe_triton/layer.py fires for Qwen3.5 (512 routed + 1 shared).
-      # Track upstream sglang for a fix; flip back to dp-attn=true once
-      # MoRI is added to is_deepep_class_backend() or shared-slot
-      # accounting is reconciled.
       - spec-decoding: "none"
         conc-list: [ 8, 16, 32, 64, 128 ]
         prefill:
           num-worker: 1
-          tp: 8
+          tp: 4
           ep: 1
           dp-attn: false
           additional-settings:
@@ -305,6 +304,49 @@ qwen3.5-fp8-mi355x-sglang-disagg:
           - "DECODE_NODES=1"
           - "DECODE_MTP_SIZE=0"
 
+# MTP-enabled variant of qwen3.5-fp8-mi355x-sglang-disagg. Same 1P+1D
+# TP4P+TP8D/EP1 topology with multi-token prediction on the decode worker
+# via spec-decoding=mtp. Qwen3.5 uses the EAGLE speculative algorithm
+# (models.yaml mtp_flags); DECODE_MTP_SIZE=3 sets --speculative-num-steps 3,
+# --speculative-num-draft-tokens 4, reproducing the validated single-node
+# recipe (EAGLE, num-steps 3, draft-tokens 4, topk 1).
+# bench.sh auto-adds --use-chat-template for the MTP benchmark (required for
+# meaningful acceptance).
+qwen3.5-fp8-mi355x-sglang-disagg-mtp:
+  image: lmsysorg/sglang:v0.5.16-rocm720-mi35x
+  model: Qwen/Qwen3.5-397B-A17B-FP8
+  model-prefix: qwen3.5
+  runner: mi355x-disagg
+  precision: fp8
+  framework: sglang-disagg
+  router: { name: sglang-router, version: "0.3.2" }
+  kv-p2p-transfer: mori
+  multinode: true
+  disagg: true
+  scenarios:
+    fixed-seq-len:
+    - isl: 8192
+      osl: 1024
+      search-space:
+      # 1P+1D TP4P+TP8D/EP1, EAGLE MTP depth 3 (num-steps 3).
+      - spec-decoding: "mtp"
+        conc-list: [ 8, 16, 32, 64, 128 ]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 1
+          dp-attn: false
+          additional-settings:
+          - "PREFILL_NODES=1"
+        decode:
+          num-worker: 1
+          tp: 8
+          ep: 1
+          dp-attn: false
+          additional-settings:
+          - "DECODE_NODES=1"
+          - "DECODE_MTP_SIZE=3"
+
 qwen3.5-fp4-mi355x-sglang:
   image: lmsysorg/sglang-rocm:v0.5.16-rocm720-mi35x-20260730
   model: amd/Qwen3.5-397B-A17B-MXFP4
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 64f87b4479..c74df5cd45 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -5770,3 +5770,13 @@
     - "models.yaml: add mtp_flags (--speculative-algorithm EAGLE --speculative-eagle-topk 1) to DeepSeek-V4-Pro-AgentX; num-steps/num-draft-tokens are derived from DECODE_MTP_SIZE in server_sglang.sh. DeepSeek-V4's MTP head is a native NextN/EAGLE draft shipped with the model, so no --speculative-draft-model-path is needed."
     - "Two search-space arms: TP8/EP1 no-DP at conc-list [2,4,8,16,32], and TP8/EP8/DPA (ep=8, dp-attn=true) at conc-list [64,96,128], both on image lmsysorg/sglang-rocm:v0.5.15.post1-rocm720-mi35x-20260719."
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2309
+
+- config-keys:
+    - qwen3.5-fp8-mi355x-sglang-disagg
+    - qwen3.5-fp8-mi355x-sglang-disagg-mtp
+  description:
+    - "Bump qwen3.5-fp8-mi355x-sglang-disagg image from v0.5.14 to v0.5.16 (3-5% throughput improvement) and switch from TP8P+TP8D (16 GPU) to TP4P+TP8D (12 GPU) — TP4 prefill delivers identical decode interactivity with 24-31% better throughput/GPU."
+    - "Add qwen3.5-fp8-mi355x-sglang-disagg-mtp: EAGLE speculative decoding (MTP depth 3) on the TP4P+TP8D topology. MTP improves TPOT by 3-14% at conc≤64; at conc=128 the decode TP8 bottleneck causes MTP collapse (-38% tput/GPU), which is resolved by switching to TP4D (see Section 13 of the Qwen3.5 DI perf tuning analysis)."
+    - "models.yaml: add mtp_flags (--speculative-algorithm EAGLE --speculative-eagle-topk 1) to Qwen3.5-397B-A17B-FP8."
+    - "Infra: add docker_sg.sh wrapper and update job.slurm DOCKER_CMD_DETECT to use sg-docker fallback for nodes where Slurm doesn't activate the docker supplementary group."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2557

From 3a57f317990555bc32a789e80be40530ebc00d37 Mon Sep 17 00:00:00 2001
From: ChangLiu0709 <cliu1004@amd.com>
Date: Tue, 11 Aug 2026 11:32:33 +0000
Subject: [PATCH 02/10] fix(qwen3.5-disagg-mtp): split MTP config for conc=128
 TP4D decode

At conc=128, TP8 decode bottlenecks MTP causing throughput collapse.
Use TP4P+TP4D (8 GPU) for conc=128 while keeping TP4P+TP8D (12 GPU)
for conc 8-64, matching the validated local benchmark results.

Co-authored-by: Cursor <cursoragent@cursor.com>
---
 configs/amd-master.yaml | 26 ++++++++++++++++++++++++--
 1 file changed, 24 insertions(+), 2 deletions(-)

diff --git a/configs/amd-master.yaml b/configs/amd-master.yaml
index 8c337fca5a..ea013195aa 100644
--- a/configs/amd-master.yaml
+++ b/configs/amd-master.yaml
@@ -328,9 +328,9 @@ qwen3.5-fp8-mi355x-sglang-disagg-mtp:
     - isl: 8192
       osl: 1024
       search-space:
-      # 1P+1D TP4P+TP8D/EP1, EAGLE MTP depth 3 (num-steps 3).
+      # 1P+1D TP4P+TP8D/EP1, EAGLE MTP depth 3 — conc 8-64.
       - spec-decoding: "mtp"
-        conc-list: [ 8, 16, 32, 64, 128 ]
+        conc-list: [ 8, 16, 32, 64 ]
         prefill:
           num-worker: 1
           tp: 4
@@ -346,6 +346,28 @@ qwen3.5-fp8-mi355x-sglang-disagg-mtp:
           additional-settings:
           - "DECODE_NODES=1"
           - "DECODE_MTP_SIZE=3"
+      # 1P+1D TP4P+TP4D/EP1, EAGLE MTP depth 3 — conc 128.
+      # At conc=128, TP8 decode bottlenecks MTP (draft verification
+      # saturates 8 GPUs, causing -38% tput/GPU and TTFT explosion).
+      # Switching to TP4D resolves the collapse: +83% tput/GPU, 7.5x
+      # better TTFT vs TP8D, while interactivity converges (39.5 vs 49.5).
+      - spec-decoding: "mtp"
+        conc-list: [ 128 ]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 1
+          dp-attn: false
+          additional-settings:
+          - "PREFILL_NODES=1"
+        decode:
+          num-worker: 1
+          tp: 4
+          ep: 1
+          dp-attn: false
+          additional-settings:
+          - "DECODE_NODES=1"
+          - "DECODE_MTP_SIZE=3"
 
 qwen3.5-fp4-mi355x-sglang:
   image: lmsysorg/sglang-rocm:v0.5.16-rocm720-mi35x-20260730

From 2f9c1249e2b83ae9e6920e94eea238d7b494ff73 Mon Sep 17 00:00:00 2001
From: ChangLiu0709 <cliu1004@amd.com>
Date: Tue, 11 Aug 2026 14:38:13 +0000
Subject: [PATCH 03/10] fix(job.slurm): add kill -9 fallback for stuck Docker
 containers

When a GPU hang leaves a container in an unkillable state ("did not
receive an exit event"), docker rm -f is insufficient. Add a fallback
that retrieves the container's init PID and kills it directly so the
daemon can release GPU VRAM and the cgroup.

Applied to pre-run cleanup, trap handler, and post-run cleanup.

Co-authored-by: Cursor <cursoragent@cursor.com>
---
 benchmarks/multi_node/amd_utils/job.slurm | 34 +++++++++++++++++++++--
 1 file changed, 32 insertions(+), 2 deletions(-)

diff --git a/benchmarks/multi_node/amd_utils/job.slurm b/benchmarks/multi_node/amd_utils/job.slurm
index d6ac8df9f3..de132ae7a8 100755
--- a/benchmarks/multi_node/amd_utils/job.slurm
+++ b/benchmarks/multi_node/amd_utils/job.slurm
@@ -386,7 +386,7 @@ cleanup() {
   # on every allocated node. Scoped to $DOCKER_CONT_NAME so it never touches
   # other users' containers. (Ported from InferenceY 51ebfa88.)
   srun --nodelist="$SELECTED_NODELIST_SRUN" \
-    bash -c 'eval "$DOCKER_CMD_DETECT"; $DOCKER_CMD rm -f '"$DOCKER_CONT_NAME"' 2>/dev/null || true' 2>/dev/null || true
+    bash -c 'eval "$DOCKER_CMD_DETECT"; $DOCKER_CMD rm -f '"$DOCKER_CONT_NAME"' 2>/dev/null || true; _pid=$($DOCKER_CMD inspect --format "{{.State.Pid}}" '"$DOCKER_CONT_NAME"' 2>/dev/null || true); if [ -n "$_pid" ] && [ "$_pid" != "0" ]; then sudo kill -9 "$_pid" 2>/dev/null || true; sleep 2; $DOCKER_CMD rm -f '"$DOCKER_CONT_NAME"' 2>/dev/null || true; fi' 2>/dev/null || true
   rm -rf ${SLURM_SUBMIT_DIR}/logs 2>/dev/null || true
   echo "[${SLURM_JOB_ID}] cleanup done."
 }
@@ -685,6 +685,23 @@ fi  # end: if ENGINE == atom-disagg
 \$DOCKER_CMD ps -aq --filter \"$CONT_FILTER\" | xargs -r \$DOCKER_CMD rm -f || true
 \$DOCKER_CMD ps -aq | xargs -r \$DOCKER_CMD stop -t 15 || true
 \$DOCKER_CMD ps -aq | xargs -r \$DOCKER_CMD rm -f || true
+
+# Fallback for stuck containers: if docker rm -f fails (\"did not receive an
+# exit event\"), the container's shim/init PID is stuck in uninterruptible
+# sleep (common after a GPU hang). Kill the container PIDs directly so the
+# daemon can clean up the cgroup and release GPU VRAM.
+_stuck=\$(\$DOCKER_CMD ps -aq 2>/dev/null || true)
+if [[ -n \"\$_stuck\" ]]; then
+    echo \"[pre-clean] containers still present after docker rm -f; attempting direct PID kill\"
+    for _cid in \$_stuck; do
+        _pid=\$(\$DOCKER_CMD inspect --format '{{.State.Pid}}' \"\$_cid\" 2>/dev/null || true)
+        if [[ -n \"\$_pid\" && \"\$_pid\" != \"0\" ]]; then
+            sudo kill -9 \"\$_pid\" 2>/dev/null || true
+        fi
+    done
+    sleep 3
+    \$DOCKER_CMD ps -aq | xargs -r \$DOCKER_CMD rm -f 2>/dev/null || true
+fi
 sleep 2
 
 # GPU sanity gate: containers are stopped, so any remaining VRAM use is a bare
@@ -782,7 +799,20 @@ exit \$DOCKER_EXIT_CODE
 "
 
 if [[ "${KEEP_CONTAINERS}" != "1" ]]; then
-    srun --nodelist="$SELECTED_NODELIST_SRUN" bash -c 'eval "$DOCKER_CMD_DETECT"; $DOCKER_CMD rm -f '"$DOCKER_CONT_NAME"' '"$CLIENT_CONT_NAME"' 2>/dev/null || true'
+    srun --nodelist="$SELECTED_NODELIST_SRUN" bash -c '
+        eval "$DOCKER_CMD_DETECT"
+        for _cont in '"$DOCKER_CONT_NAME"' '"$CLIENT_CONT_NAME"'; do
+            $DOCKER_CMD rm -f "$_cont" 2>/dev/null || true
+            if $DOCKER_CMD inspect "$_cont" &>/dev/null; then
+                _pid=$($DOCKER_CMD inspect --format "{{.State.Pid}}" "$_cont" 2>/dev/null || true)
+                if [ -n "$_pid" ] && [ "$_pid" != "0" ]; then
+                    sudo kill -9 "$_pid" 2>/dev/null || true
+                    sleep 2
+                    $DOCKER_CMD rm -f "$_cont" 2>/dev/null || true
+                fi
+            fi
+        done
+    ' || true
 
     # Clean up vLLM external router container on node 0
     if [[ "$ENGINE" == "vllm-disagg" && "$ROUTER_TYPE" == "vllm-router" ]]; then

From faabfa24ecb4074cb39c290784603d4e5fc32a60 Mon Sep 17 00:00:00 2001
From: ChangLiu0709 <cliu1004@amd.com>
Date: Wed, 12 Aug 2026 10:04:35 +0000
Subject: [PATCH 04/10] fix(job.slurm): pre-pull Docker image on all nodes
 before barrier

The container creation barrier (300s timeout) races against Docker
image downloads when one node has the image cached and another doesn't.
A multi-GB SGLang image pull can easily exceed 300s, causing the
barrier to timeout and the job to fail with "No logs directory found".

Add a best-effort srun pre-pull step before the main benchmark srun
so all nodes have the image cached before the barrier starts.

Co-authored-by: Cursor <cursoragent@cursor.com>
---
 benchmarks/multi_node/amd_utils/job.slurm | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/benchmarks/multi_node/amd_utils/job.slurm b/benchmarks/multi_node/amd_utils/job.slurm
index de132ae7a8..7b67e5ed7b 100755
--- a/benchmarks/multi_node/amd_utils/job.slurm
+++ b/benchmarks/multi_node/amd_utils/job.slurm
@@ -588,6 +588,14 @@ if [[ -n "${CLIENT_IMAGE:-}" ]]; then
     srun --nodelist="$SELECTED_NODELIST_SRUN" bash -c 'eval "$DOCKER_CMD_DETECT"; $DOCKER_CMD pull '"$CLIENT_IMAGE"' >/dev/null 2>&1 || true' 2>/dev/null || true
 fi
 
+# Pre-pull the main Docker image on every node so the container creation
+# barrier (300s) doesn't race against a multi-GB image download on nodes
+# where the image isn't cached. Best-effort: failure here is non-fatal
+# since docker run will pull as a fallback.
+echo "[pre-pull] Pulling $DOCKER_IMAGE_NAME on all nodes..."
+srun --nodelist="$SELECTED_NODELIST_SRUN" \
+  bash -c 'eval "$DOCKER_CMD_DETECT"; echo "[pre-pull] $(hostname): pulling..."; $DOCKER_CMD pull '"$DOCKER_IMAGE_NAME"' && echo "[pre-pull] $(hostname): done" || echo "[pre-pull] $(hostname): pull failed (will retry on docker run)"' || true
+
 srun \
   --nodelist="$SELECTED_NODELIST_SRUN" \
   --kill-on-bad-exit=1 \

From 6f1de4651d751a578a17525569caf11c90c16ea5 Mon Sep 17 00:00:00 2001
From: ChangLiu0709 <cliu1004@amd.com>
Date: Wed, 12 Aug 2026 14:24:41 +0000
Subject: [PATCH 05/10] ci: re-trigger sweep with fresh priority classification

The previous run (31587251501) was assigned priority 0.000, likely due
to a misclassification. Re-triggering to get a correct priority score.

Co-authored-by: Cursor <cursoragent@cursor.com>

From b5361fbf51d2eb3e7c9c5b1fdf7a36a618815e05 Mon Sep 17 00:00:00 2001
From: ChangLiu0709 <cliu1004@amd.com>
Date: Wed, 12 Aug 2026 14:29:11 +0000
Subject: [PATCH 06/10] fix: restore deleted minimaxm3 entry in
 perf-changelog.yaml

The merge conflict resolution accidentally dropped the minimaxm3-fp4
entry (PR #2487) from the append-only changelog, causing
process_changelog.py to reject the diff.

Co-authored-by: Cursor <cursoragent@cursor.com>
---
 perf-changelog.yaml | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 3324c01700..dee32cde03 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -5828,3 +5828,10 @@
   description:
     - "Extend the SimpleCPUOffloadConnector grid to c8/c12/c16/c20/c24/c28/c32/c48/c64 to locate its crossover against the resident curve"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2475
+
+- config-keys:
+    - minimaxm3-fp4-mi355x-vllm-agentic-mtp
+  description:
+    - "Add the MI355X MiniMax-M3 FP4 vLLM AgentX MTP submission on vLLM v0.27.1, using the committed EAGLE3 synthetic acceptance length of 3.35 for throughput and real target verification for eval."
+    - "Sweep TP2/TP4 GPU-resident configurations and TP4 LMCache MP DRAM-offload points around the capacity knee, with vLLM server metrics enabled."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2487

From f2f806c4c574dc47bca2b38c7b51d32eedbf7d54 Mon Sep 17 00:00:00 2001
From: ChangLiu0709 <cliu1004@amd.com>
Date: Wed, 12 Aug 2026 14:50:35 +0000
Subject: [PATCH 07/10] ci: re-trigger sweep after removing conflicting label

Removed full-sweep-fail-fast, keeping full-sweep-enabled only.

Co-authored-by: Cursor <cursoragent@cursor.com>

From a7ac0ee833688099e0ea93f068e50e4bc8c429b0 Mon Sep 17 00:00:00 2001
From: ChangLiu0709 <cliu1004@amd.com>
Date: Wed, 12 Aug 2026 16:33:24 +0000
Subject: [PATCH 08/10] fix(submit.sh): exclude all known bad-docker nodes

Add g14, g16, g17, g19, g31, g37 to the default SLURM_EXCLUDE_NODES
list. These nodes have stale docker-group membership (falls back to
sudo without password), causing container launch failures in CI.

Co-authored-by: Cursor <cursoragent@cursor.com>
---
 benchmarks/multi_node/amd_utils/submit.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/benchmarks/multi_node/amd_utils/submit.sh b/benchmarks/multi_node/amd_utils/submit.sh
index 6a1598d81d..2b4406affa 100755
--- a/benchmarks/multi_node/amd_utils/submit.sh
+++ b/benchmarks/multi_node/amd_utils/submit.sh
@@ -186,7 +186,7 @@ fi
 # Optional: exclude specific nodes (e.g. nodes with broken Docker sockets).
 # Set SLURM_EXCLUDE_NODES env var to a comma-separated list of hostnames.
 EXCLUDE_OPT=()
-SLURM_EXCLUDE_NODES="${SLURM_EXCLUDE_NODES:-mia1-p01-g09,mia1-p01-g10,mia1-p01-g11,mia1-p01-g12}"
+SLURM_EXCLUDE_NODES="${SLURM_EXCLUDE_NODES:-mia1-p01-g09,mia1-p01-g10,mia1-p01-g11,mia1-p01-g12,mia1-p01-g14,mia1-p01-g16,mia1-p01-g17,mia1-p01-g19,mia1-p01-g31,mia1-p01-g37}"
 if [[ -n "${SLURM_EXCLUDE_NODES:-}" ]]; then
     EXCLUDE_OPT=(--exclude "$SLURM_EXCLUDE_NODES")
 fi

From 83f761591c56a3b9ad15e670b69bdbc916195e33 Mon Sep 17 00:00:00 2001
From: ChangLiu0709 <cliu1004@amd.com>
Date: Wed, 12 Aug 2026 16:42:49 +0000
Subject: [PATCH 09/10] ci: re-trigger sweep after fixing node exclusion list

Co-authored-by: Cursor <cursoragent@cursor.com>

From fdbc486c7b84525f3a369e546bea13afeeacab0d Mon Sep 17 00:00:00 2001
From: ChangLiu0709 <cliu1004@amd.com>
Date: Thu, 13 Aug 2026 13:52:50 +0000
Subject: [PATCH 10/10] fix(submit.sh): remove g11 and g12 from default node
 exclude list

These nodes have been stable and working for SGLang disagg workloads.
Removing them increases the available node pool from 2 to 4, preventing
multi-node jobs from starving in the Slurm queue.

Co-authored-by: Cursor <cursoragent@cursor.com>
---
 benchmarks/multi_node/amd_utils/submit.sh | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/benchmarks/multi_node/amd_utils/submit.sh b/benchmarks/multi_node/amd_utils/submit.sh
index 2b4406affa..b299ccb6b2 100755
--- a/benchmarks/multi_node/amd_utils/submit.sh
+++ b/benchmarks/multi_node/amd_utils/submit.sh
@@ -186,7 +186,7 @@ fi
 # Optional: exclude specific nodes (e.g. nodes with broken Docker sockets).
 # Set SLURM_EXCLUDE_NODES env var to a comma-separated list of hostnames.
 EXCLUDE_OPT=()
-SLURM_EXCLUDE_NODES="${SLURM_EXCLUDE_NODES:-mia1-p01-g09,mia1-p01-g10,mia1-p01-g11,mia1-p01-g12,mia1-p01-g14,mia1-p01-g16,mia1-p01-g17,mia1-p01-g19,mia1-p01-g31,mia1-p01-g37}"
+SLURM_EXCLUDE_NODES="${SLURM_EXCLUDE_NODES:-mia1-p01-g09,mia1-p01-g10,mia1-p01-g14,mia1-p01-g16,mia1-p01-g17,mia1-p01-g19,mia1-p01-g31,mia1-p01-g37}"
 if [[ -n "${SLURM_EXCLUDE_NODES:-}" ]]; then
     EXCLUDE_OPT=(--exclude "$SLURM_EXCLUDE_NODES")
 fi