diff --git a/.github/configs/amd-master.yaml b/.github/configs/amd-master.yaml
index 1faf3682b..3c6c7e815 100644
--- a/.github/configs/amd-master.yaml
+++ b/.github/configs/amd-master.yaml
@@ -1497,8 +1497,8 @@ dsr1-fp4-mi355x-sglang-disagg-mtp:
         - "DECODE_MTP_SIZE=1"
 
 dsv4-fp8-mi355x-sglang:
-  image: rocm/sgl-dev:deepseek-v4-mi35x
-  model: sgl-project/DeepSeek-V4-Pro-FP8
+  image: rocm/sgl-dev:v0.5.10.post1-rocm700-mi35x-20260428
+  model: deepseek-ai/DeepSeek-V4-Pro
   model-prefix: dsv4
   runner: mi355x
   precision: fp8
@@ -1514,6 +1514,29 @@ dsv4-fp8-mi355x-sglang:
     search-space:
     - { tp: 8, conc-start: 4, conc-end: 64 }
 
+# FP4-experts variant of dsv4-fp8-mi355x-sglang. Same image and sglang overlay
+# (amd/deepseek_v4 branch pinned at runtime by dsv4_fp4_mi355x_sglang.sh);
+# differs only in SGLANG_DSV4_FP4_EXPERTS=True / SGLANG_FORCE_TRITON_MOE_FP8=0
+# and uses the bf16 DeepSeek-V4-Pro checkpoint (FP4 quantization is applied
+# at expert-runtime by sglang).
+dsv4-fp4-mi355x-sglang:
+  image: rocm/sgl-dev:v0.5.10.post1-rocm700-mi35x-20260428
+  model: deepseek-ai/DeepSeek-V4-Pro
+  model-prefix: dsv4
+  runner: mi355x
+  precision: fp4
+  framework: sglang
+  multinode: false
+  seq-len-configs:
+  - isl: 1024
+    osl: 1024
+    search-space:
+    - { tp: 8, conc-start: 4, conc-end: 64 }
+  - isl: 8192
+    osl: 1024
+    search-space:
+    - { tp: 8, conc-start: 4, conc-end: 64 }
+
 # vLLM with AITER MLA decode for DSv4 on MI355X (vllm-project/vllm#40889,
 # stacked on #40871). Uses the ATOM MI355X image (ROCm 7.2.2, aiter with
 # MLA decode, MI355X GPU detection); vLLM is rebuilt from the PR branch
diff --git a/benchmarks/single_node/dsv4_fp4_mi355x_sglang.sh b/benchmarks/single_node/dsv4_fp4_mi355x_sglang.sh
new file mode 100755
index 000000000..e6ed57b5c
--- /dev/null
+++ b/benchmarks/single_node/dsv4_fp4_mi355x_sglang.sh
@@ -0,0 +1,153 @@
+#!/usr/bin/env bash
+
+source "$(dirname "$0")/../benchmark_lib.sh"
+
+check_env_vars \
+    MODEL \
+    TP \
+    CONC \
+    ISL \
+    OSL \
+    RANDOM_RANGE_RATIO \
+    RESULT_FILENAME
+
+if [[ -n "$SLURM_JOB_ID" ]]; then
+  echo "JOB $SLURM_JOB_ID running on $SLURMD_NODENAME"
+fi
+
+hf download "$MODEL"
+
+# Overlay sglang from the amd/deepseek_v4 branch on top of whatever the
+# rocm/sgl-dev:v0.5.10.post1-rocm700-mi35x-20260428 image ships with. We
+# stay on the rocm700 (ROCm 7.0.0a) line because rocm720 hit
+# hipErrorInvalidConfiguration on use_symmetric_memory-allocated dp_attention
+# buffers (RCCL symmetric-memory bug; SGLANG_USE_ROCM700A WA only covers the
+# cuda-graph path, not eager mode that we use via --disable-cuda-graph).
+# Bump SGL_PR_SHA when the branch advances.
+SGL_PR_SHA="18afbf151a2992b06a089191769b299629ed73dd"
+SGL_PR_DIR="/tmp/sglang-amd-dsv4"
+
+if [ ! -d "$SGL_PR_DIR/.git" ]; then
+    git clone --filter=blob:none https://github.com/sgl-project/sglang.git "$SGL_PR_DIR"
+fi
+(
+    cd "$SGL_PR_DIR"
+    git fetch --depth=1 origin "$SGL_PR_SHA" 2>/dev/null \
+        || git fetch --depth=1 origin amd/deepseek_v4
+    git checkout --force "$SGL_PR_SHA"
+    test "$(git rev-parse HEAD)" = "$SGL_PR_SHA"
+
+    # Reinstall just the Python package; the image already has the ROCm
+    # kernel deps (aiter, triton, tilelang, torch) at versions matched to
+    # this branch, so --no-deps avoids pip resolving them against PyPI.
+    pip install --no-build-isolation --no-deps --force-reinstall -e python/
+)
+
+python3 -c "import sglang; print(f'sglang {sglang.__version__} from {sglang.__path__[0]}')"
+
+# Transformers in the container doesn't recognize the `deepseek_v4` model_type.
+# PR #23608's fallback in hf_transformers_utils.get_config tries to handle this
+# by writing a patched config to /tmp, but in practice isn't catching the error
+# in this image. Patch the cached config.json directly instead: set model_type
+# to `deepseek_v3` so AutoConfig.from_pretrained succeeds, and keep
+# architectures=['DeepseekV4ForCausalLM'] so SGLang dispatches to its native
+# DSv4 model class (python/sglang/srt/models/deepseek_v4.py).
+python3 << PYEOF
+import json
+from huggingface_hub import hf_hub_download
+path = hf_hub_download(repo_id="$MODEL", filename="config.json")
+with open(path) as f:
+    config = json.load(f)
+if config.get("model_type") == "deepseek_v4":
+    config["model_type"] = "deepseek_v3"
+    with open(path, "w") as f:
+        json.dump(config, f, indent=2)
+    print(f"Patched {path}: model_type deepseek_v4 -> deepseek_v3")
+else:
+    print(f"No patch needed: model_type is {config.get('model_type')!r}")
+PYEOF
+
+# DSv4 FP4-experts path. Mirrors the active path of python/run_dsv4.sh on
+# the amd/deepseek_v4 branch at SGL_PR_SHA:
+#   SGLANG_DSV4_FP4_EXPERTS=True   -> route experts through the FP4 kernels
+#   SGLANG_FORCE_TRITON_MOE_FP8=0  -> dispatch MoE through aiter (gating
+#                                    switch added in commit 33de1e64);
+#                                    also enables swiglu_limit clamp in the
+#                                    triton MoE fallback path.
+export SGLANG_REASONING_EFFORT=max
+export SGLANG_OPT_USE_FUSED_COMPRESS=false
+export SGLANG_OPT_USE_OLD_COMPRESSOR=true
+export SGLANG_OPT_USE_TILELANG_SWA_PREPARE=false
+export SGLANG_OPT_USE_JIT_KERNEL_FUSED_TOPK=false
+export SGLANG_OPT_USE_FUSED_HASH_TOPK=false
+export SGLANG_HACK_FLASHMLA_BACKEND=torch
+export SGLANG_OPT_DEEPGEMM_HC_PRENORM=false
+export SGLANG_OPT_USE_TILELANG_MHC_PRE=false
+export SGLANG_OPT_USE_TILELANG_MHC_POST=false
+export SGLANG_ENABLE_THINKING=1
+export SGLANG_USE_AITER=1
+export SGLANG_USE_ROCM700A=1
+export SGLANG_TOPK_TRANSFORM_512_TORCH=1
+export SGLANG_FP8_PAGED_MQA_LOGITS_TORCH=1
+export SGLANG_DSV4_FP4_EXPERTS=True
+export SGLANG_OPT_DPSK_V4_RADIX=0
+export SGLANG_OPT_USE_OVERLAP_STORE_CACHE=false
+export SGLANG_OPT_USE_FUSED_STORE_CACHE=false
+export SGLANG_FORCE_TRITON_MOE_FP8=0
+
+SERVER_LOG=/workspace/server.log
+PORT=${PORT:-8888}
+
+EVAL_CONTEXT_ARGS=""
+if [ "${EVAL_ONLY}" = "true" ]; then
+    setup_eval_context
+    EVAL_CONTEXT_ARGS="--context-length $EVAL_MAX_MODEL_LEN"
+fi
+# Start GPU monitoring (power, temperature, clocks every second)
+start_gpu_monitor
+
+python3 -m sglang.launch_server \
+    --model-path $MODEL \
+    --host=0.0.0.0 \
+    --port $PORT \
+    --tensor-parallel-size $TP \
+    --dp $TP \
+    --enable-dp-attention \
+    --trust-remote-code \
+    --disable-radix-cache \
+    --attention-backend compressed \
+    --max-running-request 256 \
+    --page-size 256 \
+    --chunked-prefill-size 8192 \
+    --disable-shared-experts-fusion \
+    --disable-cuda-graph \
+    --tool-call-parser deepseekv4 \
+    --reasoning-parser deepseek-v4 \
+    --watchdog-timeout 1800 $EVAL_CONTEXT_ARGS > $SERVER_LOG 2>&1 &
+
+SERVER_PID=$!
+
+# Wait for server to be ready
+wait_for_server_ready --port "$PORT" --server-log "$SERVER_LOG" --server-pid "$SERVER_PID"
+
+run_benchmark_serving \
+    --model "$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 "$RESULT_FILENAME" \
+    --result-dir /workspace/
+
+# After throughput, run evaluation only if RUN_EVAL is true
+if [ "${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/benchmarks/single_node/dsv4_fp8_mi355x.sh b/benchmarks/single_node/dsv4_fp8_mi355x.sh
index 971b18b6a..afb6d0fef 100755
--- a/benchmarks/single_node/dsv4_fp8_mi355x.sh
+++ b/benchmarks/single_node/dsv4_fp8_mi355x.sh
@@ -17,6 +17,34 @@ fi
 
 hf download "$MODEL"
 
+# Overlay sglang from the amd/deepseek_v4 branch on top of whatever the
+# rocm/sgl-dev:v0.5.10.post1-rocm700-mi35x-20260428 image ships with. We
+# stay on the rocm700 (ROCm 7.0.0a) line because rocm720 hit
+# hipErrorInvalidConfiguration on use_symmetric_memory-allocated dp_attention
+# buffers (RCCL symmetric-memory bug; SGLANG_USE_ROCM700A WA only covers the
+# cuda-graph path, not eager mode that we use via --disable-cuda-graph).
+# Bump SGL_PR_SHA when the branch advances.
+SGL_PR_SHA="18afbf151a2992b06a089191769b299629ed73dd"
+SGL_PR_DIR="/tmp/sglang-amd-dsv4"
+
+if [ ! -d "$SGL_PR_DIR/.git" ]; then
+    git clone --filter=blob:none https://github.com/sgl-project/sglang.git "$SGL_PR_DIR"
+fi
+(
+    cd "$SGL_PR_DIR"
+    git fetch --depth=1 origin "$SGL_PR_SHA" 2>/dev/null \
+        || git fetch --depth=1 origin amd/deepseek_v4
+    git checkout --force "$SGL_PR_SHA"
+    test "$(git rev-parse HEAD)" = "$SGL_PR_SHA"
+
+    # Reinstall just the Python package; the image already has the ROCm
+    # kernel deps (aiter, triton, tilelang, torch) at versions matched to
+    # this branch, so --no-deps avoids pip resolving them against PyPI.
+    pip install --no-build-isolation --no-deps --force-reinstall -e python/
+)
+
+python3 -c "import sglang; print(f'sglang {sglang.__version__} from {sglang.__path__[0]}')"
+
 # Transformers in the container doesn't recognize the `deepseek_v4` model_type.
 # PR #23608's fallback in hf_transformers_utils.get_config tries to handle this
 # by writing a patched config to /tmp, but in practice isn't catching the error
@@ -39,7 +67,12 @@ else:
     print(f"No patch needed: model_type is {config.get('model_type')!r}")
 PYEOF
 
-# DSv4-specific SGLang env vars (from sgl-project/sglang#23608)
+# DSv4-specific SGLang env vars. Mirrors python/run_dsv4.sh on the
+# amd/deepseek_v4 branch (commented FP8 path) at SGL_PR_SHA. The branch's
+# FP4 Models integration commit (33de1e64) flipped SGLANG_FORCE_TRITON_MOE_FP8
+# from 1 to 0; with it set to 0, FP8 MoE dispatches through aiter (shuffled
+# weights + aiter fused_moe) instead of the triton MoE fallback.
+export SGLANG_REASONING_EFFORT=max
 export SGLANG_OPT_USE_FUSED_COMPRESS=false
 export SGLANG_OPT_USE_OLD_COMPRESSOR=true
 export SGLANG_OPT_USE_TILELANG_SWA_PREPARE=false
@@ -58,7 +91,7 @@ export SGLANG_DSV4_FP4_EXPERTS=false
 export SGLANG_OPT_DPSK_V4_RADIX=0
 export SGLANG_OPT_USE_OVERLAP_STORE_CACHE=false
 export SGLANG_OPT_USE_FUSED_STORE_CACHE=false
-export SGLANG_FORCE_TRITON_MOE_FP8=1
+export SGLANG_FORCE_TRITON_MOE_FP8=0
 
 SERVER_LOG=/workspace/server.log
 PORT=${PORT:-8888}
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 422d5347f..85bca5bd0 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -2006,3 +2006,15 @@
     - "Change image to vllm/vllm-openai:v0.20.0-cu130"
     - "Use Mega MoE for DEP configs"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1221
+
+- config-keys:
+    - dsv4-fp8-mi355x-sglang
+    - dsv4-fp4-mi355x-sglang
+  description:
+    - "Pin sglang at runtime to amd/deepseek_v4 branch SHA 18afbf15 via clone+reinstall in the benchmark scripts"
+    - "Bump dsv4-fp8-mi355x-sglang image to rocm/sgl-dev:v0.5.10.post1-rocm700-mi35x-20260428"
+    - "Switch dsv4-fp8-mi355x-sglang model from sgl-project/DeepSeek-V4-Pro-FP8 to deepseek-ai/DeepSeek-V4-Pro"
+    - "Set SGLANG_FORCE_TRITON_MOE_FP8=0 in dsv4_fp8_mi355x.sh so FP8 MoE dispatches through aiter (matches branch run_dsv4.sh after FP4 Models commit 33de1e64)"
+    - "Add SGLANG_REASONING_EFFORT=max to both scripts"
+    - "Add dsv4-fp4-mi355x-sglang variant: SGLANG_DSV4_FP4_EXPERTS=True, SGLANG_FORCE_TRITON_MOE_FP8=0, model deepseek-ai/DeepSeek-V4-Pro"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1231