Set BLS size to 256 and turn off restart

triton-inference-server · May 30, 2024 · a06e090 · a06e090
1 parent 0dcae3f
commit a06e090
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 2 deletions.
diff --git a/qa/L0_backend_python/bls/test.sh b/qa/L0_backend_python/bls/test.sh
@@ -100,7 +100,7 @@ if [[ ${TEST_WINDOWS} == 0 ]]; then
         echo "instance_group [ { kind: KIND_CPU} ]" >> models/libtorch_cpu/config.pbtxt
 
     # Test with different sizes of CUDA memory pool
-    for CUDA_MEMORY_POOL_SIZE_MB in 64 128 ; do
+    for CUDA_MEMORY_POOL_SIZE_MB in 64 256 ; do
         CUDA_MEMORY_POOL_SIZE_BYTES=$((CUDA_MEMORY_POOL_SIZE_MB * 1024 * 1024))
         SERVER_ARGS="--model-repository=${MODELDIR}/bls/models --backend-directory=${BACKEND_DIR} --log-verbose=1 --cuda-memory-pool-byte-size=0:${CUDA_MEMORY_POOL_SIZE_BYTES}"
         for TRIAL in non_decoupled decoupled ; do

diff --git a/qa/L0_backend_python/test.sh b/qa/L0_backend_python/test.sh
@@ -448,7 +448,7 @@ SUBTESTS="lifecycle argument_validation logging custom_metrics"
 # [DLIS-6122] Disable model_control & request_rescheduling tests for Windows since they require load/unload
 # [DLIS-6123] Disable examples test for Windows since it requires updates to the example clients
 if [[ ${TEST_WINDOWS} == 0 ]]; then
-    SUBTESTS+=" restart model_control examples request_rescheduling"
+    SUBTESTS+=" model_control examples request_rescheduling"
 fi
 for TEST in ${SUBTESTS}; do
     # Run each subtest in a separate virtual environment to avoid conflicts