Skip to content

Commit 10a4f03

Browse files
seungrokjclaude
andcommitted
[AMD][AgentX] dsv4-fp4-mi355x-sglang-agentic-mtp: reduce mem_fraction_static to 0.85 and chunked_prefill_size to 4096 to avoid runtime OOM
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
1 parent 2fca704 commit 10a4f03

1 file changed

Lines changed: 3 additions & 3 deletions

File tree

benchmarks/single_node/agentic/dsv4_fp4_mi355x_sglang_mtp.sh

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -82,8 +82,8 @@ export AIPERF_HTTP_TCP_USER_TIMEOUT=1000000
8282
USE_SGLANG_ROUTER=false
8383
SGLANG_BACKEND_PORT="$PORT"
8484
ROUTER_LOG="$RESULT_DIR/router.log"
85-
MEM_FRACTION_STATIC=0.90
86-
CHUNKED_PREFILL_SIZE=8192
85+
MEM_FRACTION_STATIC=0.85
86+
CHUNKED_PREFILL_SIZE=4096
8787
PARALLEL_ARGS=(--tensor-parallel-size "$TP")
8888
if [ "$DP_ATTENTION" = "true" ]; then
8989
USE_SGLANG_ROUTER=true
@@ -98,7 +98,7 @@ if [ "$DP_ATTENTION" = "true" ]; then
9898
export SGLANG_DP_USE_REDUCE_SCATTER=1
9999
export GPU_MAX_HW_QUEUES=5
100100

101-
CHUNKED_PREFILL_SIZE=$((8192 * TP))
101+
CHUNKED_PREFILL_SIZE=$((4096 * TP))
102102
PARALLEL_ARGS+=(
103103
--dp "$TP"
104104
--enable-dp-attention

0 commit comments

Comments
 (0)