Raise ubatch benchmark context to 32K

scripts/bench-ubatch-rocmi4.sh

@@ -3,7 +3,7 @@
 # frank-default-large ROCmFPX standalone-MTP stack (dankclaw, gfx1151).
 #
 # Boots one llama-server instance per ubatch on a scratch port with a fixed
-# -b and -c 16384, runs a standardized ~75%-of-ctx prose workload, and reads
+# -b and -c 32768, runs a standardized ~75%-of-ctx prose workload, and reads
 # authoritative prefill/generation tok/s from the server's own print_timing
 # lines. Steps from UBATCH_MIN to UBATCH_MAX inclusive by UBATCH_STEP.
 #
@@ -30,8 +30,8 @@ fi
 : "${UBATCH_MAX:=6144}"
 : "${UBATCH_STEP:=256}"
 : "${UBATCH_VALUES:=256 512 768 1024 1280 1536 2048 3072 4096 5120 6144}"
-: "${CTX:=16384}"                # 16k contexts
-: "${BENCH_BATCH:=16384}"         # batch is fixed to the 16k context size
+: "${CTX:=32768}"                # 32k contexts
+: "${BENCH_BATCH:=32768}"         # batch is fixed to the 32k context size
 : "${OUT_TOKENS:=256}"
 : "${RUNS:=3}"                   # decode runs per ubatch
 : "${PORT_BASE:=7894}"           # scratch ports 7894..