[AMD][AgentX] kimik3: add fp8retry arm (none c[1,16] + vllm-simple-fp8 c8)

Retries the two cells that died in run 30453589555 and adds none c16.

Both deaths were GPU faults (hipErrorUnknown), not timeouts. No guard was
close to firing: VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS=1200 against a ~180s
stall, warmup-grace-period 1800 against 420s elapsed, and the only "timeout"
strings in the server log are source lines inside the traceback. A timeout
would also not surface as hipErrorUnknown.

GPU faults have hit every arm on this cluster today -- bf16 kvnone on
g11/g17/g18/g19, fp8 kvnone on g10, fp8 + offload connector on g11 and g19 --
so both cells are retried unchanged rather than "fixed". The offload cell was
measurably slower before it died (2.7x longer to 5 completions than the
no-connector run at the same concurrency), but slowness did not kill it, and
that is a throughput question to settle with a number rather than a config
change.

none c16 is new. The no-offload arm has never been measured above c8, and fp8
roughly halves KV bytes per token, so the concurrency ceiling should have
moved; bf16 collapsed at c16 (588 tok/s/GPU, TTFT 295s) and this checks
whether fp8 changes that.

Also corrects the KV_CACHE_DTYPE header comment, which still claimed a
default of auto after acd6d30 made fp8 the default.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

Author: hyukjlee

benchmarks/single_node/agentic/kimik3_fp4_mi355x.sh

@@ -45,7 +45,7 @@ set -x
 #   MAX_NUM_BATCHED_TOKENS   4096   (reference)
 #   AITER_A8W4               1      (reference; 0 = aiter a16w4 MoE path)
 #   LANGUAGE_MODEL_ONLY      false  (reference loads the vision tower)
-#   KV_CACHE_DTYPE           auto   (unset -> flag not passed at all)
+#   KV_CACHE_DTYPE           fp8    (default for every arm; =auto for a bf16 A/B)
 #   MAX_MODEL_LEN            unset  (unset -> vLLM derives K3's 1M context)
 #   SPEC_DECODE              false  (DSpark; UNVALIDATED on ROCm)
 

configs/amd-master.yaml

@@ -729,6 +729,33 @@ kimik3-fp4-mi355x-vllm-agentic-fp8sweep:
       # fp8 pushes the concurrency ceiling past the bf16 collapse at c16.
       - { tp: 8, kv-offloading: dram, kv-offload-backend: { name: vllm-simple-fp8 }, conc-list: [8, 16] }
 
+kimik3-fp4-mi355x-vllm-agentic-fp8retry:
+  # Retry of the two cells that died in run 30453589555, plus a new none c16.
+  #
+  # Both deaths were GPU faults (hipErrorUnknown), not timeouts -- no guard was
+  # anywhere near firing (VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS=1200 vs a ~180s
+  # stall; warmup-grace-period 1800 vs 420s elapsed). GPU faults have hit every
+  # arm on this cluster today (bf16 kvnone on g11/g17/g18/g19, fp8 kvnone on
+  # g10, fp8+connector on g11/g19), so these are retried unchanged on the
+  # assumption they are the cluster's background fault rate rather than config.
+  #
+  # none c16 is new: the no-offload arm has never been measured above c8, and
+  # fp8 roughly halves KV per token, so the concurrency ceiling should have
+  # moved up.
+  image: vllm/vllm-openai-rocm:kimi-k3
+  model: moonshotai/Kimi-K3
+  model-prefix: kimik3
+  runner: cluster:mi355x-amds
+  precision: fp4
+  framework: vllm
+  multinode: false
+  scenarios:
+    agentic-coding:
+    - dram-utilization: 0.80
+      search-space:
+      - { tp: 8, kv-offloading: none, conc-list: [1, 16] }
+      - { tp: 8, kv-offloading: dram, kv-offload-backend: { name: vllm-simple-fp8 }, conc-list: [8] }
+
 kimik3-fp4-mi355x-vllm-agentic-fp8kv:
   # fp8 KV on the OFFLOAD arm. This is the fp8kvtest arm (run 30442578333:
   # 696 total tok/s/GPU, TTFT 2.6s, the first clean fp8 result) plus the