feifeibear
commented
3 years ago Sorry I can not give any suggestions without more meaningful information provided? I made a docker with a lower torch version. Maybe you can have a try. thufeifeibear/turbo_transformers_cpu:bert_only_v0.1
In the prebuilt cpu docker container, I downgrade torch to 1.1.0, onnxruntime 1.4.0. date time ( uptime ) [ thread name/id ] file:line v| 2021-06-11 11:15:22.255 ( 0.000s) [main thread ] loguru.cpp:610 INFO| arguments: turbo_transformers_cxx 2021-06-11 11:15:22.255 ( 0.000s) [main thread ] loguru.cpp:613 INFO| Current dir: /workspace/benchmark 2021-06-11 11:15:22.255 ( 0.000s) [main thread ] loguru.cpp:615 INFO| stderr verbosity: 0 2021-06-11 11:15:22.255 ( 0.000s) [main thread ] loguru.cpp:616 INFO| ----------------------------------- {"QPS": 31.49370412192892, "elapsed": 4.762856710003689, "n": 150, "batch_size": 1, "seq_len": 10, "framework": "turbo", "thread_num": 4, "model_name": "bert"} 2021-06-11 11:15:28.936 ( 6.680s) [main thread ] loguru.cpp:489 INFO| atexit {"QPS": 19.366318455987148, "elapsed": 7.745406043017283, "n": 150, "batch_size": 1, "seq_len": 10, "framework": "torch", "thread_num": 4, "model_name": "bert"} {"QPS": 19.7451021386962, "elapsed": 7.596820667036809, "n": 150, "batch_size": 1, "seq_len": 10, "framework": "torch_jit", "n_threads": 4, "model_name": "bert"} run_cpu_fixed_benchmark.sh: line 28: 271 Segmentation fault (core dumped) env OMP_WAIT_POLICY=ACTIVE OMP_NUM_THREADS=${n_th} python benchmark.py ${MODEL} --seq_len=${seq_len} --batch_size=${batch_size} -n ${N} --framework=${framework} --num_threads=${n_th}