[AMD][AgentX] kimik2.7 TP8 3600s: none c[4,8], lmcache+mooncake c[16,32,48]; fix Mooncake worker import

Reshape the TP8 agentic sweep for the real 3600s comparison run:
- none (GPU-only) swept only at low conc [4,8] (it collapses under load)
- LMCache and Mooncake carry the high-conc [16,32,48] range
- duration 3600s (the prior 900s smoke was warmup-dominated for the
  ~90k-token agentic inputs, so per-request metrics were unreliable)

Fix Mooncake so the vLLM worker subprocesses can import it: the previous
bare `make install` only laid down C++ libs, so the launcher imported
`mooncake` but the workers hit "Please install mooncake ..." at KV-cache
init. Build the wheel via Mooncake's scripts/build_wheel.sh (auditwheel
bundles every .so) and pip install it into site-packages, and build with
-DWITH_STORE=ON so store.so (MooncakeStoreConnector) is present.

Drop the temporary configs/smoke-kimik2.7-mooncake.yaml overlay; all three
backends now run from one canonical config.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: hyukjlee

benchmarks/single_node/agentic/kimik2.7_fp4_mi355x.sh

@@ -211,18 +211,31 @@ case "$OFFLOAD_MODE" in
         # if the store module isn't importable. Clone to a container-local dir
         # (NOT bind-mounted /workspace) so the next job's checkout `clean: true`
         # won't trip over root-owned build artifacts.
+        #
+        # A bare `make install` only lays down the C++ libs, so the launcher can
+        # import `mooncake` but the vLLM worker SUBPROCESSES cannot (fresh
+        # site-packages) -> "Please install mooncake ..." at KV-cache init. Build
+        # the wheel via Mooncake's own scripts/build_wheel.sh (auditwheel bundles
+        # every .so into a self-contained package) and pip install it so the
+        # `mooncake` package lands in site-packages and is importable everywhere.
         if ! python3 -c "from mooncake.store import MooncakeDistributedStore" >/dev/null 2>&1; then
             MOONCAKE_SRC_DIR="${MOONCAKE_SRC_DIR:-/opt/mooncake-src}"
             MOONCAKE_GIT_REF="${MOONCAKE_GIT_REF:-main}"
+            MOONCAKE_PYVER="$(python3 -c 'import sys;print(f"{sys.version_info.major}.{sys.version_info.minor}")')"
+            pip install --quiet build auditwheel patchelf
             rm -rf "$MOONCAKE_SRC_DIR"
             git clone https://github.com/kvcache-ai/Mooncake.git "$MOONCAKE_SRC_DIR"
             ( cd "$MOONCAKE_SRC_DIR"
               git checkout "$MOONCAKE_GIT_REF"
               bash dependencies.sh
               mkdir -p build && cd build
-              cmake ..
+              cmake .. -DWITH_STORE=ON
               make -j
-              make install )
+              make install
+              cd ..
+              # Produces a self-contained wheel under mooncake-wheel/dist/.
+              bash scripts/build_wheel.sh "$MOONCAKE_PYVER"
+              pip install --force-reinstall mooncake-wheel/dist/*.whl )
             python3 -c "from mooncake.store import MooncakeDistributedStore" >/dev/null
         fi
 

configs/amd-master.yaml

@@ -558,14 +558,15 @@ kimik2.7-fp4-mi355x-vllm-agentic:
   multinode: false
   scenarios:
     agentic-coding:
-    # TP8 900s smoke to break the TP4 GPU-KV-capacity wall at higher concurrency.
-    # At TP4 the ~121 GiB KV pool held only ~12 resident ~90k-token contexts, so
-    # throughput plateaued (~160 tok/s) and preemption thrashed from conc16 up.
-    # TP8 doubles GPUs -> ~2x KV pool + faster per-token decode.
+    # TP8, 3600s. Compares three KV tiers on the agentic trace. At TP8 the
+    # GPU-only (none) path collapses under load (KV-capacity thrash), so it is
+    # only swept at low conc [4,8] where it is viable; the dram offload backends
+    # (LMCache, Mooncake) carry the high-conc [16,32,48] range where they hold
+    # throughput by keeping the GPU KV pool from thrashing.
     - dram-utilization: 0.80
-      duration: 900
+      duration: 3600
       search-space:
-      - { tp: 8, kv-offloading: none, conc-list: [16, 32, 48] }
+      - { tp: 8, kv-offloading: none, conc-list: [4, 8] }
       - { tp: 8, kv-offloading: dram, kv-offload-backend: { name: lmcache, version: "aaf7c0d3" }, conc-list: [16, 32, 48] }
       - { tp: 8, kv-offloading: dram, kv-offload-backend: { name: mooncake, version: "source-main" }, conc-list: [16, 32, 48] }