Update cpp_graph_inference.sh

intel · Mar 7, 2024 · b204c08 · b204c08
1 parent a70ffc1
commit b204c08
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/tests/model-test/cpp_graph_inference.sh b/tests/model-test/cpp_graph_inference.sh
@@ -410,7 +410,7 @@ function main() {
                     else
                         real_ctx=$ctx # TODO(Zhenzhong): use same ctx for  chatglm & baichuan
                         [[ "${model}" == "chatglm2" || "${model}" == "chatglm-6b" ||
-                            "${model}" == "baichuan-13b" || "${model}" == "baichuan2-13b" ]] && real_ctx=1500
+                            "${model}" == "baichuan-13b" || "${model}" == "baichuan2-13b" ]] && real_ctx=1100
                         NEURAL_SPEED_VERBOSE=1 OMP_NUM_THREADS=$cores_per_instance numactl -m 0 -C 0-$(($cores_per_instance - 1)) \
                             $infer_cmd --seed 1234 -t $cores_per_instance -b 2047 -c $real_ctx -n ${output} -m ${model}-${precision}.bin $extension -p "$prompt" 2>&1 | tee ${WORKSPACE}/${logs_file} || true &
                         monitor