experimental_experiment.torch_bench.dort_benchΒΆ

python -m experimental_experiment.torch_bench.dort_bench \
    -w 3                    \
    -r 5                    \
    --config medium         \
    --mixed 1               \
    --device cuda           \
    --verbose 0             \
    --backend custom        \
    --num_hidden_layers 1   \
    --dynamic 0
llama config={'input_dims': [(2, 1024), (2, 1024), (2, 1024), (2, 1024), (2, 1024), (2, 1024), (2, 1024), (2, 1024)], 'hidden_size': 1024, 'num_hidden_layers': 1, 'vocab_size': 1024, 'intermediate_size': 1024, 'max_position_embeddings': 1024, 'num_attention_heads': 2, '_attn_implementation': 'eager'}
backend=custom
verbose=0
implementation=eager
mixed=1
disable_pattern=[]
enable_pattern=['default']
CUDA no model: memory allocated=0, reserved=0
CUDA model loaded: memory allocated=37762048, reserved=44040192
Build the compile model with backend=custom
dynamic=False
warmup on device=cuda
CUDA memory allocated=37762048, reserved=44040192
torch/_functorch/_aot_autograd/utils.py:117: UserWarning: Your compiler for AOTAutograd is returning a function that doesn't take boxed arguments. Please wrap it with functorch.compile.make_boxed_func or handle the boxed arguments yourself. See https://github.com/pytorch/pytorch/pull/83137#issuecomment-1211320670 for rationale.
warnings.warn(
warmup done in 3.2293482000004587s.
memory allocated=37786624, reserved=65011712
measures
measures done.
dynamic=0
mixed=1
backend=custom
num_hidden_layers=1
mixed=1
repeat=5
device=cuda
avg=0.05740846000007878
times=[0.05616349999945669, 0.05769109999982902, 0.05737370000042574, 0.058259600000383216, 0.05755440000029921]
warmup_times=[3.1141115999998874, 0.05787190000046394, 0.057364700000107405]
-----------
:llama,2x1024-1024-1-1024-1024-1024-2-eager;
:config,medium;
:mixed,1;
:dynamic,False;
:backend,custom;
:repeat,5;
:warmup,3;
:torch,2.3.0.dev20240314+cu118;
:transformers,4.37.2;
:patterns,+default-;
:warmup_time,3.2293482000004587;
:time,0.05740846000007878;