o i e@s~dZddlZddlZddlmZddlmZddZddZ dd d Z Gd d d ej Z           dddZ dS)zCommon utils for benchmarks.N) callbacks)distribution_utilcCs,d|vsd|vr td|dddS)alSplit the suffix of the benchmark name. For example, for the name = 'benchmark_layer_call__Conv2D_small_shape', the return value is ['Conv2D', 'small', 'shape']. This is to generate the metadata of the benchmark test. Args: name: A string, the benchmark name. Returns: A list of strings of the suffix in the benchmark name. ___z*The format of the benchmark name is wrong.) ValueErrorsplit)namer R/var/www/myenv/lib/python3.10/site-packages/keras/src/benchmarks/benchmark_util.pyget_benchmark_namesr cGs:g}|D]}|dd|D|dd|Dq|S)zExtend the benchmark names with CPU and GPU suffix. Args: *params_list: A list of tuples represents the benchmark parameters. Returns: A list of strings with the benchmark name extended with CPU and GPU suffix. cS&g|]}|ddf|ddqS)r_CPUNr .0paramr r r 9&z5generate_benchmark_params_cpu_gpu..cSr )r_GPUrNr rr r r r<r)extend) params_listbenchmark_paramsparamsr r r !generate_benchmark_params_cpu_gpu,s   r.keras.cfit_graphcCsd||dt|dS)Nkeras_examplesbs_) model_nameimplementation parameters)str) keras_model batch_sizeimplr r r get_keras_examples_metadataAs r%c@s0eZdZdZddZddZddZdd Zd S) TimerCallBackz1Callback for logging time in each epoch or batch.cCs"g|_tj|_t|_d|_dS)NF)timestimeit default_timertimer startup_timerecorded_startup)selfr r r __init__Ns  zTimerCallBack.__init__cCs||_dSN)r*epoch_start_timer-elogsr r r on_epoch_beginTszTimerCallBack.on_epoch_begincCs|j||jdSr/)r'appendr*r0r1r r r on_epoch_endWszTimerCallBack.on_epoch_endcCs$|js||j|_d|_dSdS)NT)r,r*r+r1r r r on_batch_endZs zTimerCallBack.on_batch_endN)__name__ __module__ __qualname____doc__r.r4r6r7r r r r r&Ks  r& offc # Cs,|durtd|durtd|durtd| dkr td|jd} ggg} }}ggg}}}|| }tj| | d}t|D]}tj}|}t|}|#|}|}||}|}|j|||d||}Wdn1sywY|j |||d d t }|}|j |||||g| d |} | || || |j | t |j| | || || |qDg}| d t | d | dt |d | dt |d | dt |d | dt |d | d|d t |}!| | d}"||!|"fS)aRun models and measure the performance. Args: model_fn: Model function to be benchmarked. x: Input data. See `x` in the `fit()` method of `keras.Model`. y: Target data. See `y` in the `fit()` method of `keras.Model`. epochs: Integer. Number of epochs to train the model. If unspecified, `epochs` will default to 2. batch_size: Integer. Number of samples per gradient update. If unspecified, `batch_size` will default to 32. run_iters: Integer. Number of iterations to run the performance measurement. If unspecified, `run_iters` will default to 4. optimizer: String (name of optimizer) or optimizer instance. See `tf.keras.optimizers`. loss: String (name of objective function), objective function or `tf.keras.losses.Loss` instance. See `tf.keras.losses`. metrics: Lists of metrics to be evaluated by the model during training. See `metrics` in the `compile()` method of `keras.Model`. verbose: 0, 1, 2. Verbosity mode. See `verbose` in the `fit()` method of `keras.Model`. If unspecified, `verbose` will default to 0. num_gpus: Number of GPUs to run the model. distribution_strategy: Distribution strategies. It could be `multi_worker_mirrored`, `one_device`, `mirrored`. If unspecified, `distribution_strategy` will default to 'off'. Note that, `TPU` and `parameter_server` are not supported yet. Returns: Performance summary, which contains build_time, compile_time, startup_time, avg_epoch_time, wall_time, exp_per_sec, epochs, distribution_strategy. Raise: ValueError: If `x` is none or if `optimizer` is not provided or if `loss` is not provided or if `num_gpus` is negative. NzInput data is required.zOptimizer is required.zLoss function is required.rz`num_gpus` cannot be negative)distribution_strategynum_gpus) optimizerlossmetricsr)xyr#epochs)rErFr#rGrverbose build_time)r value compile_timer+avg_epoch_time exp_per_secrG)rshaperget_distribution_strategyranger(r)get_strategy_scopecompilefitr&r5r+npmeanr')#model_fnrErFrGr# run_itersrBrCrDrHrAr@ num_examplesbuild_time_listcompile_time_liststartup_time_listavg_epoch_time_listwall_time_listexp_per_sec_listtotal_num_examplesstrategyrr* start_timestrategy_scopet0modelrIt1rKcbkt2end_time wall_timeextrasr r r measure_performance`s1           rk)r) NNr<r=r>NNNrrr?)r;r(numpyrT keras.srcrkeras.src.benchmarksrr rr%Callbackr&rkr r r r s,