fix: increase gpu-memory-utilization to 0.95 for dsr1_dep (#5144)

Signed-off-by: alec-flowers <aflowers@nvidia.com>

examples/backends/vllm/launch/dsr1_dep.sh

@@ -107,7 +107,7 @@ for ((i=0; i<GPUS_PER_NODE; i++)); do
         --max-model-len 4096 \
         --data-parallel-address $MASTER_ADDR \
         --data-parallel-rpc-port 13345 \
-        --gpu-memory-utilization 0.9 \
+        --gpu-memory-utilization 0.95 \
         --enforce-eager 2>&1 | tee $LOG_DIR/dsr1_dep_${dp_rank}.log &
 done