Skip to content

Commit 461bbe7

Browse files
seungrokjclaude
andcommitted
fix(agentic): reduce Kimi FP4 B200 CPU DRAM limit to 1500 GB
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
1 parent 2441f1f commit 461bbe7

1 file changed

Lines changed: 2 additions & 2 deletions

File tree

benchmarks/single_node/agentic/kimik2.5_fp4_b200.sh

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -101,7 +101,7 @@ case "$OFFLOADING" in
101101
# the full eager sweep before.
102102
#(srok), internal node limitation
103103
#TOTAL_CPU_DRAM_GB=2500
104-
TOTAL_CPU_DRAM_GB=2500
104+
TOTAL_CPU_DRAM_GB=1500
105105
export VLLM_USE_SIMPLE_KV_OFFLOAD=1
106106
OFFLOAD_ARGS=(
107107
--kv_offloading_backend native
@@ -123,7 +123,7 @@ case "$OFFLOADING" in
123123
# cudaHostAlloc in LMCache 0.4.5's single-process local CPU backend.
124124
#(srok), internal node limitation
125125
#TOTAL_CPU_DRAM_GB=2500
126-
TOTAL_CPU_DRAM_GB=2500
126+
TOTAL_CPU_DRAM_GB=1500
127127
LMCACHE_HOST="${LMCACHE_HOST:-127.0.0.1}"
128128
LMCACHE_PORT="${LMCACHE_PORT:-5555}"
129129
LMCACHE_HTTP_PORT="${LMCACHE_HTTP_PORT:-8080}"

0 commit comments

Comments
 (0)