yeandy commented on a change in pull request #16970: URL: https://github.com/apache/beam/pull/16970#discussion_r821798777
########## File path: sdks/python/apache_beam/ml/inference/base.py ########## @@ -0,0 +1,222 @@ +# +# Licensed to the Apache Software Foundation (ASF) under one or more +# contributor license agreements. See the NOTICE file distributed with +# this work for additional information regarding copyright ownership. +# The ASF licenses this file to You under the Apache License, Version 2.0 +# (the "License"); you may not use this file except in compliance with +# the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +# + +import logging +import platform +import resource +import sys +import time +from typing import Any +from typing import Iterable + +import apache_beam as beam +from apache_beam.utils import shared +from apache_beam.ml.inference.api import PredictionResult + +_MILLISECOND_TO_MICROSECOND = 1000 +_MICROSECOND_TO_NANOSECOND = 1000 +_SECOND_TO_MICROSECOND = 1000000 + + +def _unbatch(maybe_keyed_batches: Any): + keys, results = maybe_keyed_batches + if keys: + return zip(keys, results) + else: + return results + + +class ModelLoader: + """Has the ability to load an ML model.""" + def load_model(self): + """Loads an initializes a model for processing.""" + raise NotImplementedError(type(self)) + + +class InferenceRunner: + """Implements running inferences for a framework.""" + def run_inference(self, batch: Any, model: Any) -> Iterable[PredictionResult]: + """Runs inferences on a batch of examples and returns an Iterable of Predictions.""" + raise NotImplementedError(type(self)) + + +class MetricsCollector: + """A metrics collector that tracks ML related performance and memory usage.""" + def __init__(self, namespace: str): + # Metrics + self._inference_counter = beam.metrics.Metrics.counter( + namespace, 'num_inferences') + self._inference_request_batch_size = beam.metrics.Metrics.distribution( + namespace, 'inference_request_batch_size') + self._inference_request_batch_byte_size = ( + beam.metrics.Metrics.distribution( + namespace, 'inference_request_batch_byte_size')) + # Batch inference latency in microseconds. + self._inference_batch_latency_micro_secs = ( + beam.metrics.Metrics.distribution( + namespace, 'inference_batch_latency_micro_secs')) + self._model_byte_size = beam.metrics.Metrics.distribution( + namespace, 'model_byte_size') + # Model load latency in milliseconds. + self._load_model_latency_milli_secs = beam.metrics.Metrics.distribution( + namespace, 'load_model_latency_milli_secs') + + # Metrics cache + self.load_model_latency_milli_secs_cache = None + self.model_byte_size_cache = None + + def update_metrics_with_cache(self): + if self.load_model_latency_milli_secs_cache is not None: + self._load_model_latency_milli_secs.update( + self.load_model_latency_milli_secs_cache) + self.load_model_latency_milli_secs_cache = None + if self.model_byte_size_cache is not None: + self._model_byte_size.update(self.model_byte_size_cache) + self.model_byte_size_cache = None + + def update( + self, + examples_count: int, + examples_byte_size: int, + latency_micro_secs: int): + self._inference_batch_latency_micro_secs.update(latency_micro_secs) + self._inference_counter.inc(examples_count) + self._inference_request_batch_size.update(examples_count) + self._inference_request_batch_byte_size.update(examples_byte_size) + + +class RunInferenceDoFn(beam.DoFn): + def __init__(self, model_loader, inference_runner, clock=None): + self._model_loader = model_loader + self._inference_runner = inference_runner + self._shared_model_handle = shared.Shared() + # TODO: Compute a good metrics namespace + self._metrics_collector = MetricsCollector('default_namespace') + self._clock = clock + if not clock: + self._clock = _ClockFactory.make_clock() + self._model = None + + def _load_model(self): + def load(): + """Function for constructing shared LoadedModel.""" + memory_before = _get_current_process_memory_in_bytes() + start_time = self._clock.get_current_time_in_microseconds() + model = self._model_loader.load_model() + end_time = self._clock.get_current_time_in_microseconds() + memory_after = _get_current_process_memory_in_bytes() + self._metrics_collector.load_model_latency_milli_secs_cache = ( + (end_time - start_time) / _MILLISECOND_TO_MICROSECOND) + self._metrics_collector.model_byte_size_cache = ( + memory_after - memory_before) + return model + + return self._shared_model_handle.acquire(load) + + def setup(self): + super().setup() + self._model = self._load_model() + + def process(self, batch): + has_keys = isinstance(batch[0], tuple) + start_time = self._clock.get_current_time_in_microseconds() + if has_keys: + examples = [example for _, example in batch] + keys = [key for key, _ in batch] + else: + examples = batch + keys = None + inference_generator = self._inference_runner.run_inference( + examples, self._model) + predictions = [ + PredictionResult(e, r) for e, r in zip(examples, inference_generator) + ] + inference_latency = self._clock.get_current_time_in_microseconds( Review comment: 👍 -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
