Open TT-billteng opened 3 months ago
TT_METAL_WATCHER=60 pytest models/experimental/bert_large_performant/unit_tests/test_bert_large_split_query_key_value_and_split_heads.py::test_split_query_key_value_and_split_heads -v
models.experimental.bert_large_performant.unit_tests.test_bert_large_split_query_key_value_and_split_heads:run_split_query_key_value_and_split_heads_test:52 - v: BufferType.L1 and DataType.BFLOAT8_B 2024-03-16T05:29:28.3021735Z [38;2;000;128;000m LLRuntime[0m | [1m[38;2;100;149;237mINFO [0m | Watcher checking device 0 2024-03-16T05:29:28.3024248Z [38;2;000;128;000m Always[0m | [1m[38;2;100;149;237mINFO [0m | While running kernels: 2024-03-16T05:29:28.3026580Z [38;2;000;128;000m Always[0m | [1m[38;2;100;149;237mINFO [0m | brisc : tt_eager/tt_dnn/op_library/transformer_tms/kernels/dataflow/writer_tm_tile_layout_create_qkv_heads.cpp 2024-03-16T05:29:28.3029402Z [38;2;000;128;000m Always[0m | [1m[38;2;100;149;237mINFO [0m | ncrisc: tt_eager/tt_dnn/op_library/transformer_tms/kernels/dataflow/reader_tm_tile_layout_create_qkv_heads.cpp 2024-03-16T05:29:28.3031777Z [38;2;000;128;000m Always[0m | [1m[38;2;100;149;237mINFO [0m | triscs: tt_eager/tt_dnn/kernels/compute/transpose_wh.cpp 2024-03-16T05:29:28.3033542Z [38;2;000;128;000m Always[0m | [1m[38;2;100;149;237mINFO [0m | Last waypoint: NWBD,W,W,W,W 2024-03-16T05:29:28.3035058Z terminate called after throwing an instance of 'std::runtime_error' 2024-03-16T05:29:28.3036170Z what(): TT_THROW @ tt_metal/impl/debug/watcher_server.cpp:291: tt::exception 2024-03-16T05:29:28.3037010Z info: 2024-03-16T05:29:28.3037703Z Watcher detected an assert: core {}, riscv {}, line {}. Current kernel: {}. {} 2024-03-16T05:29:28.3038577Z (x=1,y=1) 2024-03-16T05:29:28.3039747Z [38;2;000;128;000m LLRuntime[0m | [1m[38;2;100;149;237mINFO [0m | Watcher stopped the device due to tripped assert. 2024-03-16T05:29:28.3044278Z [38;2;000;128;000m Always[0m | [1m[38;2;255;000;000mFATAL [0m | Watcher detected an assert: core (x=1,y=1), riscv brisc, line 195. Current kernel: tt_eager/tt_dnn/op_library/transformer_tms/kernels/dataflow/writer_tm_tile_layout_create_qkv_heads.cpp. Note that file name reporting is not yet implemented, and the reported line number for the assert may be from a different file. 2024-03-16T05:29:28.3047638Z brisc 2024-03-16T05:29:28.3047998Z 195
hey @jliangTT not sure who should own the issue, can you help?
TT_METAL_WATCHER=60 pytest models/experimental/bert_large_performant/unit_tests/test_bert_large_split_query_key_value_and_split_heads.py::test_split_query_key_value_and_split_heads -v