SlideShare a Scribd company logo

High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference

•
•
Chris Fregly
Chris Fregly

http://pipeline.ai Title PipelineAI Distributed Spark ML + Tensorflow AI + GPU Workshop *A GPU-based cloud instance will be provided to each attendee as part of this event Highlights We will each build an end-to-end, continuous Tensorflow AI model training and deployment pipeline on our own GPU-based cloud instance. At the end, we will combine our cloud instances to create the LARGEST Distributed Tensorflow AI Training and Serving Cluster in the WORLD! Agenda Spark ML Tensorflow AI Storing and Serving Models with HDFS Trade-offs of CPU vs. *GPU, Scale Up vs. Scale Out CUDA + cuDNN GPU Development Overview Tensorflow Model Checkpointing, Saving, Exporting, and Importing Distributed Tensorflow AI Model Training (Distributed Tensorflow) Centralized Logging and Visualizing of Distributed Tensorflow Training (Tensorboard) Distributed Tensorflow AI Model Serving/Predicting (Tensorflow Serving) Centralized Logging and Metrics Collection (Prometheus, Grafana) Continuous Tensorflow AI Model Deployment (Tensorflow, Airflow) Hybrid Cross-Cloud and On-Premise Deployments (Kubernetes) High-Performance and Fault-Tolerant Microsservices using Request Batching and Circuit Breakers (NetflixOSS) Github Repo https://github.com/fluxcapacitor/pipeline

1 of 96
Download to read offline
HIGH PERFORMANCE TENSORFLOW IN PRODUCTION WITH GPUS! CHRIS FREGLY, FOUNDER @ PIPELINE.AI ML TRAIN, SYDNEY 2017
INTRODUCTIONS: ME § Chris Fregly, Research Engineer @ § Formerly Netflix and Databricks § Advanced Spark and TensorFlow Meetup Please Join Our 20,000+ Members Globally! * San Francisco * Chicago * Washington DC * London Please Join!!
INTRODUCTIONS: YOU § Software Engineer or Data Scientist interested in optimizing and deploying TensorFlow models to production § Assume you have a working knowledge of TensorFlow
100% OPEN SOURCE CODE § https://github.com/fluxcapacitor/pipeline/ § Please Star this Repo! J § Slides, code, notebooks, Docker images available here: https://github.com/fluxcapacitor/pipeline/ gpu.ml
HANDS-ON EXERCISES § Combo of Jupyter Notebooks and Command Line § Command Line through Jupyter Terminal § Some Exercises Based on Experimental Features Warning: You Will See Errors. You will be OK!!
CONTENT NOTES § 50% Training Optimizations (GPUs, XLA, JIT) § 50% Predicting Optimizations (XLA, AOT, TF Serving) § Why Heavy Focus on Predicting? § Training: boring batch, O(num_data_scientists) § Inference: exciting realtime, O(num_users_of_app) § We Use Simple Models to Highlight Optimizations Warning: This is not intro material. You will be OK!
YOU WILL LEARN… § Part 1: TensorFlow Model Training § TensorFlow and GPUs § Inspect and Debug Models § Distributed Training Across a Cluster § Optimize Training with Queues, Dataset API, and JIT XLA Compiler § Part 2: TensorFlow Model Deploying and Serving § Optimize Predicting with AOT XLA and Graph Transform Tool (GTT) § Deploy Model and Predict § Key Components of TensorFlow Serving § Optimize TensorFlow Serving Runtime
AGENDA § GPUs and TensorFlow § Train and Debug TensorFlow Model § Train with Distributed TensorFlow Cluster § Optimize Model Training with XLA JIT Compiler § Optimize Model Predicting with XLA AOT and Graph Transforms § Deploy Model to TensorFlow Serving Runtime § Optimize TensorFlow Serving Runtime § Wrap-up and Q&A
EVERYBODY GETS A GPU!
SETUP ENVIRONMENT § Step 1: Browse to the following: http://allocator.community.pipeline.ai/allocate § Step 2: Browse to the following: http://<ip-address> § Step 3: Browse around. I will provide a username/password in a bit! Need Help? Use the Chat!
VERIFY SETUP http://<ip-address> Any username, Any password!
LET’S EXPLORE OUR ENVIRONMENT § Navigate to the following notebook: 01_Explore_Environment § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
PULSE CHECK
BREAK § https://github.com/fluxcapacitor/pipeline/ § Slides, code, notebooks, Docker images available here: https://github.com/fluxcapacitor/pipeline/ gpu.ml Need Help? Use the Chat!
SETTING UP TENSORFLOW WITH GPUS § Very Painful! § Especially inside Docker § Use nvidia-docker § Especially on Kubernetes! § Use Kubernetes 1.7+ § http://pipeline.ai for GitHub + DockerHub Links
GPU HALF-PRECISION SUPPORT § FP16, INT8 are “Half Precision” § Supported by Pascal P100 (2016) and Volta V100 (2017) § Flexible FP32 GPU Cores Can Fit 2 FP16’s for 2x Throughput! § Half-Precision is OK for Approximate Deep Learning Use Cases
VOLTA V100 RECENTLY ANNOUNCED § 84 Streaming Multiprocessors (SM’s) § 5,376 GPU Cores § 672 Tensor Cores (ie. Google TPU) § Mixed FP16/FP32 Precision § More Shared Memory § New L0 Instruction Cache § Faster L1 Data Cache § V100 vs. P100 Performance § 12x TFLOPS @ Peak Training § 6x Inference Throughput
V100 AND CUDA 9 § Independent Thread Scheduling - Finally!! § Similar to CPU fine-grained thread synchronization semantics § Allows GPU to yield execution of any thread § Still Optimized for SIMT (Same Instruction Multiple Thread) § SIMT units automatically scheduled together § Explicit Synchronization P100 V100
GPU CUDA PROGRAMMING § Barbaric, But Fun Barbaric! § Must Know Underlying Hardware Very Well § Many Great Debuggers/Profilers § Hardware Changes are Painful! § Newer CUDA versions automatically JIT-compile old CUDA code to new NVPTX § Not optimal, of course
CUDA STREAMS § Asynchronous I/O Transfer § Overlap Compute and I/O § Keeps GPUs Saturated § Fundamental to Queue Framework in TensorFlow
LET’S SEE WHAT THIS THING CAN DO! § Navigate to the following notebook: 01a_Explore_GPU 01b_Explore_Numba § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
AGENDA § GPUs and TensorFlow § Train and Debug TensorFlow Model § Train with Distributed TensorFlow Cluster § Optimize Model Training with XLA JIT Compiler § Optimize Model Predicting with XLA AOT and Graph Transforms § Deploy Model to TensorFlow Serving Runtime § Optimize TensorFlow Serving Runtime § Wrap-up and Q&A
TRAINING TERMINOLOGY § Tensors: N-Dimensional Arrays § ie. Scalar, Vector, Matrix § Operations: MatMul, Add, SummaryLog,… § Graph: Graph of Operations (DAG) § Session: Contains Graph(s) § Feeds: Feed inputs into Operation § Fetches: Fetch output from Operation § Variables: What we learn through training § aka “weights”, “parameters” § Devices: Hardware device on which we train -TensorFlow- Trains Variables -User- Fetches Outputs -User- Feeds Inputs -TensorFlow- Performs Operations -TensorFlow- Flows Tensors with tf.device(“worker:0/device/gpu:0,worker:1/device/gpu:0”)
TRAINING DEVICES § cpu:0 § By default, all CPUs § Requires extra config to target a CPU § gpu:0..n § Each GPU has a unique id § TF usually prefers a single GPU § xla_cpu:0, xla_gpu:0..n § “JIT Compiler Device” § Hints TensorFlow to attempt JIT Compile with tf.device(“/cpu:0”): with tf.device(“/gpu:0”): with tf.device(“/gpu:1”):
TRAINING METRICS: TENSORBOARD § Summary Ops § Event Files /root/tensorboard/linear/<version>/events… § Tags § Organize data within Tensorboard UI loss_summary_op = tf.summary.scalar('loss', loss) merge_all_summary_op = tf.summary.merge_all() summary_writer = tf.summary.FileWriter( '/root/tensorboard/linear/<version>', graph=sess.graph)
TRAINING ON EXISTING INFRASTRUCTURE § Data Processing § HDFS/Hadoop § Spark § Containers § Docker § Schedulers § Kubernetes § Mesos <dependency> <groupId>org.tensorflow</groupId> <artifactId>tensorflow-hadoop</artifactId> </dependency> https://github.com/tensorflow/ecosystem
TRAINING PIPELINES: QUEUE + DATASET § Don’t Use feed_dict for Production Workloads!! § feed_dict Requires Python <-> C++ Serialization § Retrieval is Single-threaded, Synchronous, SLOW! § Can’t Retrieve Until Current Batch is Complete § CPUs/GPUs Not Fully Utilized! § Use Queue or Dataset API
QUEUES § More than Just a Traditional Queue § Perform I/O, pre-processing, cropping, shuffling § Pulls from HDFS, S3, Google Storage, Kafka, ... § Combine many small files into large TFRecord files § Typically use CPUs to focus GPUs on compute § Uses CUDA Streams
DATA MOVEMENT WITH QUEUES § GPU Pulls Batch from Queue (CUDA Streams) § GPU pulls next batch while processing current batch GPUs Stay Fully Utilized!
QUEUE CAPACITY PLANNING § batch_size § # examples / batch (ie. 64 jpg) § Limited by GPU RAM § num_processing_threads § CPU threads pull and pre-process batches of data § Limited by CPU Cores § queue_capacity § Limited by CPU RAM (ie. 5 * batch_size)
DETECT UNDERUTILIZED CPUS, GPUS § Instrument training code to generate “timelines” § Analyze with Google Web Tracing Framework (WTF) § Monitor CPU with `top`, GPU with `nvidia-smi` http://google.github.io/tracing-framework/ from tensorflow.python.client import timeline trace = timeline.Timeline(step_stats=run_metadata.step_stats) with open('timeline.json', 'w') as trace_file: trace_file.write( trace.generate_chrome_trace_format(show_memory=True))
LET’S FEED DATA WITH A QUEUE § Navigate to the following notebook: 02_Feed_Queue_HDFS § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
PULSE CHECK
BREAK § https://github.com/fluxcapacitor/pipeline/ § Slides, code, notebooks, Docker images available here: https://github.com/fluxcapacitor/pipeline/ gpu.ml Need Help? Use the Chat!
TENSORFLOW MODEL § MetaGraph § Combines GraphDef and Metadata § GraphDef § Architecture of your model (nodes, edges) § Metadata § Asset: Accompanying assets to your model § SignatureDef: Maps external : internal tensors § Variables § Stored separately during training (checkpoint) § Allows training to continue from any checkpoint § Variables are “frozen” into Constants when deployed for inference GraphDef x W mul add b MetaGraph Metadata Assets SignatureDef Tags Version Variables: “W” : 0.328 “b” : -1.407
TENSORFLOW SESSION Session graph: GraphDef Variables: “W” : 0.328 “b” : -1.407 Variables are Periodically Checkpointed GraphDef is Static
LET’S TRAIN A MODEL (CPU) § Navigate to the following notebook: 03_Train_Model_CPU § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
LET’S TRAIN A MODEL (GPU) § Navigate to the following notebook: 03a_Train_Model_GPU § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
TENSORFLOW DEBUGGER § Step through Operations § Inspect Inputs and Outputs § Wrap Session in Debug Session sess = tf.Session(config=config) sess = tf_debug.LocalCLIDebugWrapperSession(sess)
LET’S DEBUG A MODEL § Navigate to the following notebook: 04_Debug_Model § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
BATCH NORMALIZATION § Each Mini-Batch May Have Wildly Different Distributions § Normalize per batch (and layer) § Speeds up Training!! § Weights are Learned Quicker § Final Model is More Accurate § Final mean and variance will be folded into Graph later -- Always Use Batch Normalization! -- z = tf.matmul(a_prev, W) a = tf.nn.relu(z) a_mean, a_var = tf.nn.moments(a, [0]) scale = tf.Variable(tf.ones([depth/channels])) beta = tf.Variable(tf.zeros ([depth/channels])) bn = tf.nn.batch_normalizaton(a, a_mean, a_var, beta, scale, 0.001)
AGENDA § GPU Environment § Train and Debug TensorFlow Model § Train with Distributed TensorFlow Cluster § Optimize Model Training with XLA JIT Compiler § Optimize Model Predicting with XLA AOT and Graph Transforms § Deploy Model to TensorFlow Serving Runtime § Optimize TensorFlow Serving Runtime § Wrap-up and Q&A
MULTI-GPU TRAINING (SINGLE NODE) § Variables stored on CPU (cpu:0) § Model graph (aka “replica”, “tower”) is copied to each GPU(gpu:0, gpu:1, …) Multi-GPU Training Steps: 1. CPU transfers model to each GPU 2. CPU waits on all GPUs to finish batch 3. CPU copies all gradients back from all GPUs 4. CPU synchronizes + AVG all gradients from GPUs 5. CPU updates GPUs with new variables/weights 6. Repeat Step 1 until reaching stop condition (ie. max_epochs)
DISTRIBUTED, MULTI-NODE TRAINING § TensorFlow Automatically Inserts Send and Receive Ops into Graph § Parameter Server Synchronously Aggregates Updates to Variables § Nodes with Multiple GPUs will Pre-Aggregate Before Sending to PS Worker0 Worker0 Worker1 Worker0 Worker1 Worker2 gpu0 gpu1 gpu2 gpu3 gpu0 gpu1 gpu2 gpu3 gpu0 gpu1 gpu2 gpu3 gpu0 gpu1 gpu0 gpu0
SYNCHRONOUS VS. ASYNCHRONOUS § Synchronous § Nodes compute gradients § Nodes update Parameter Server (PS) § Nodes sync on PS for latest gradients § Asynchronous § Some nodes delay in computing gradients § Nodes don’t update PS § Nodes get stale gradients from PS § May not converge due to stale reads!
DATA PARALLEL VS MODEL PARALLEL § Data Parallel (“Between-Graph Replication”) § Send exact same model to each device § Each device operates on its partition of data § ie. Spark sends same function to many workers § Each worker operates on their partition of data § Model Parallel (“In-Graph Replication”) § Send different partition of model to each device § Each device operates on all data Very Difficult!! Required for Large Models. (GPU RAM Limitation)
DISTRIBUTED TENSORFLOW CONCEPTS § Client § Program that builds a TF Graph, constructs a session, interacts with the cluster § Written in Python, C++ § Cluster § Set of distributed nodes executing a graph § Nodes can play any role § Jobs (“Roles”) § Parameter Server (“ps”) stores and updates variables § Worker (“worker”) performs compute-intensive tasks (stateless) § Assigned 0..* tasks § Task (“Server Process”) “ps” and “worker” are conventional names
CHIEF WORKER § Worker Task 0 is Chosen by Default § Task 0 is guaranteed to exist § Implements Maintenance Tasks § Writes checkpoints § Initializes parameters at start of training § Writes log summaries § Parameter Server health checks
NODE AND PROCESS FAILURES § Checkpoint to Persistent Storage (HDFS, S3) § Use MonitoredTrainingSession and Hooks § Use a Good Cluster Orchestrator (ie. Kubernetes,Mesos) § Understand Failure Modes and Recovery States Stateless, Not Bad: Training Continues Stateful, Bad: Training Must Stop Dios Mio! Long Night Ahead…
VALIDATING DISTRIBUTED MODEL § Separate Training and Validation Clusters § Validate using Saved Checkpoints from Parameter Servers § Avoids Resource Contention Training Cluster Validation Cluster Parameter Server Cluster
LET’S TRAIN WITH DISTRIBUTED CPU § Navigate to the following notebook: 05_Train_Model_Distributed_CPU § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
LET’S TRAIN WITH DISTRIBUTED GPU § Navigate to the following notebook: 05a_Train_Model_Distributed_GPU § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
NEW(‘ISH): EXPERIMENT + ESTIMATOR § Higher-Level APIs Simplify Distributed Training § Picks Up Configuration from Environment § Supports Custom Models (ie. Keras) § Used for Training, Validation, and Prediction § API is Changing, but Patterns Remain the Same § Works Well with Google Cloud ML (Surprised?!)
PULSE CHECK
BREAK § https://github.com/fluxcapacitor/pipeline/ § Slides, code, notebooks, Docker images available here: https://github.com/fluxcapacitor/pipeline/ gpu.ml Need Help? Use the Chat!
AGENDA § GPUs and TensorFlow § Train and Debug TensorFlow Model § Train with Distributed TensorFlow Cluster § Optimize Model Training with XLA JIT Compiler § Optimize Model Predicting with XLA AOT and Graph Transforms § Deploy Model to TensorFlow Serving Runtime § Optimize TensorFlow Serving Runtime § Wrap-up and Q&A
XLA FRAMEWORK § Accelerated Linear Algebra (XLA) § Goals: § Reduce reliance on custom operators § Improve execution speed § Improve memory usage § Reduce mobile footprint § Improve portability § Helps TF Stay Flexible and Performant
XLA HIGH LEVEL OPTIMIZER (HLO) § Compiler Intermediate Representation (IR) § Independent of source and target language § Define Graphs using HLO Language § XLA Step 1 Emits Target-Independent HLO § XLA Step 2 Emits Target-Dependent LLVM § LLVM Emits Native Code Specific to Target § Supports x86-64, ARM64 (CPU), and NVPTX (GPU)
JIT COMPILER § Just-In-Time Compiler § Built on XLA Framework § Goals: § Reduce memory movement – especially useful on GPUs § Reduce overhead of multiple function calls § Similar to Spark Operator Fusing in Spark 2.0 § Unroll Loops, Fuse Operators, Fold Constants, … § Scope to session, device, or `with jit_scope():`
VISUALIZING JIT COMPILER IN ACTION Before After Google Web Tracing Framework: http://google.github.io/tracing-framework/ from tensorflow.python.client import timeline trace = timeline.Timeline(step_stats=run_metadata.step_stats) with open('timeline.json', 'w') as trace_file: trace_file.write( trace.generate_chrome_trace_format(show_memory=True))
VISUALIZING FUSING OPERATORS pip install graphviz dot -Tpng /tmp/hlo_graph_1.w5LcGs.dot -o hlo_graph_1.png GraphViz: http://www.graphviz.org hlo_*.dot files generated by XLA
LET’S TRAIN WITH XLA CPU § Navigate to the following notebook: 06_Train_Model_XLA_CPU § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
LET’S TRAIN WITH XLA GPU § Navigate to the following notebook: 06a_Train_Model_XLA_GPU § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
AGENDA § GPUs and TensorFlow § Train and Debug TensorFlow Model § Train with Distributed TensorFlow Cluster § Optimize Model Training with XLA JIT Compiler § Optimize Model Predicting with XLA AOT and Graph Transforms § Deploy Model to TensorFlow Serving Runtime § Optimize TensorFlow Serving Runtime § Wrap-up and Q&A
AOT COMPILER § Standalone, Ahead-Of-Time (AOT) Compiler § Built on XLA framework § tfcompile § Creates executable with minimal TensorFlow Runtime needed § Includes only dependencies needed by subgraph computation § Creates functions with feeds (inputs) and fetches (outputs) § Packaged as cc_libary header and object files to link into your app § Commonly used for mobile device inference graph § Currently, only CPU x86-64 and ARM are supported - no GPU
GRAPH TRANSFORM TOOL (GTT) § Optimize Trained Models for Inference § Remove training-only Ops (checkpoint, drop out, logs) § Remove unreachable nodes between given feed -> fetch § Fuse adjacent operators to improve memory bandwidth § Fold final batch norm mean and variance into variables § Round weights/variables improves compression (ie. 70%) § Quantize weights and activations simplifies model § FP32 down to INT8
BEFORE OPTIMIZATIONS
AFTER STRIPPING UNUSED NODES § Optimizations § strip_unused_nodes § Results § Graph much simpler § File size much smaller
AFTER REMOVING UNUSED NODES § Optimizations § strip_unused_nodes § remove_nodes § Results § Pesky nodes removed § File size a bit smaller
AFTER FOLDING CONSTANTS § Optimizations § strip_unused_nodes § remove_nodes § fold_constants § Results § Placeholders (feeds) -> Variables* (*Why Variables and not Constants?)
AFTER FOLDING BATCH NORMS § Optimizations § strip_unused_nodes § remove_nodes § fold_constants § fold_batch_norms § Results § Graph remains the same § File size approximately the same
WEIGHT QUANTIZATION § FP16 and INT8 Are Smaller and Computationally Simpler § Weights/Variables are Constants § Easy to Linearly Quantize
AFTER QUANTIZING WEIGHTS § Optimizations § strip_unused_nodes § remove_nodes § fold_constants § fold_batch_norms § quantize_weights § Results § Graph is same, file size is smaller, compute is faster
LET’S OPTIMIZE FOR INFERENCE § Navigate to the following notebook: 07_Optimize_Model* (*Why just CPU version? Why not both CPU and GPU?) § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
BUT WAIT, THERE’S MORE!
ACTIVATION QUANTIZATION § Activations Not Known Ahead of Time § Depends on input, not easy to quantize § Requires Additional Calibration Step § Use a “representative” dataset § Per Neural Network Layer… § Collect histogram of activation values § Generate many quantized distributions with different saturation thresholds § Choose threshold to minimize… KL_divergence(ref_distribution, quant_distribution) § Not Much Time or Data is Required (Minutes on Commodity Hardware)
ACTIVATION QUANTIZATION GRAPH Create Conversion Subgraph Produces QuantizedMatMul, QuantizedRelu Eliminate Adjacent Dequantize + Quantize
AFTER ACTIVATION QUANTIZATION § Optimizations § strip_unused_nodes § remove_nodes § fold_constants § fold_batch_norms § quantize_weights § quantize_nodes (activations) § Results § Larger graph, needs calibration! Requires additional freeze_requantization_ranges
LET’S OPTIMIZE FOR INFERENCE § Navigate to the following notebook: 08_Optimize_Model_Activations § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
LINEARIZE GRAPH EXECUTION ORDER § https://github.com/yaroslavvb/stuff Linearize to minimize graph memory usage
AGENDA § GPUs and TensorFlow § Train and Debug TensorFlow Model § Train with Distributed TensorFlow Cluster § Optimize Model Training with XLA JIT Compiler § Optimize Model Predicting with XLA AOT and Graph Transforms § Deploy Model to TensorFlow Serving Runtime § Optimize TensorFlow Serving Runtime § Wrap-up and Q&A
MODEL SERVING TERMINOLOGY § Inference § Only Forward Propagation through Network § Predict, Classify, Regress, … § Bundle § GraphDef, Variables, Metadata, … § Assets § ie. Map of ClassificationID -> String § {9283: “penguin”, 9284: “bridge”} § Version § Every Model Has a Version Number (Integer) § Version Policy § ie. Serve Only Latest (Highest), Serve Both Latest and Previous, …
TENSORFLOW SERVING FEATURES § Supports Auto-Scaling § Custom Loaders beyond File-based § Tune for Low-latency or High-throughput § Serve Diff Models/Versions in Same Process § Customize Models Types beyond HashMap and TensorFlow § Customize Version Policies for A/B and Bandit Tests § Support Request Draining for Graceful Model Updates § Enable Request Batching for Diff Use Cases and HW § Supports Optimized Transport with GRPC and Protocol Buffers
PREDICTION SERVICE § Predict (Original, Generic) § Input: List of Tensor § Output: List of Tensor § Classify § Input: List of tf.Example (key, value) pairs § Output: List of (class_label: String, score: float) § Regress § Input: List of tf.Example (key, value) pairs § Output: List of (label: String, score: float)
PREDICTION INPUTS + OUTPUTS § SignatureDef § Defines inputs and outputs § Maps external (logical) to internal (physical) tensor names § Allows internal (physical) tensor names to change from tensorflow.python.saved_model import utils from tensorflow.python.saved_model import signature_constants from tensorflow.python.saved_model import signature_def_utils graph = tf.get_default_graph() x_observed = graph.get_tensor_by_name('x_observed:0') y_pred = graph.get_tensor_by_name('add:0') inputs_map = {'inputs': x_observed} outputs_map = {'outputs': y_pred} predict_signature = signature_def_utils.predict_signature_def(inputs=inputs_map, outputs=outputs_map)
MULTI-HEADED INFERENCE § Multiple “Heads” of Model § Return class and scores to be fed into another model § Inputs Propagated Forward Only Once § Optimizes Bandwidth, CPU, Latency, Memory, Coolness
BUILD YOUR OWN MODEL SERVER (?!) § Adapt GRPC(Google) <-> HTTP (REST of the World) § Perform Batch Inference vs. Request/Response § Handle Requests Asynchronously § Support Mobile, Embedded Inference § Customize Request Batching § Add Circuit Breakers, Fallbacks § Control Latency Requirements § Reduce Number of Moving Parts #include “tensorflow_serving/model_servers/server_core.h” class MyTensorFlowModelServer { ServerCore::Options options; // set options (model name, path, etc) std::unique_ptr<ServerCore> core; TF_CHECK_OK( ServerCore::Create(std::move(options), &core) ); } Compile and Link libtensorflow.so
FREEZING MODEL FOR DEPLOYMENT § Optimizations § strip_unused_nodes § remove_nodes § fold_constants § fold_batch_norms § quantize_weights § quantize_nodes § freeze_graph § Results § Variables -> Constants Finally! We’re Ready to Deploy!!
LET’S DEPLOY OPTIMIZED MODEL § Navigate to the following notebook: 09_Deploy_Optimized_Model § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
AGENDA § GPUs and TensorFlow § Train and Debug TensorFlow Model § Train with Distributed TensorFlow Cluster § Optimize Model Training with XLA JIT Compiler § Optimize Model Predicting with XLA AOT and Graph Transforms § Deploy Model to TensorFlow Serving Runtime § Optimize TensorFlow Serving Runtime § Wrap-up and Q&A
REQUEST BATCH TUNING § max_batch_size § Enables throughput/latency tradeoff § Bounded by RAM § batch_timeout_micros § Defines batch time window, latency upper-bound § Bounded by RAM § num_batch_threads § Defines parallelism § Bounded by CPU cores § max_enqueued_batches § Defines queue upper bound, throttling § Bounded by RAM Reaching either threshold will trigger a batch
BATCH SCHEDULER STRATEGIES § BasicBatchScheduler § Best for homogeneous request types (ie. always classify or always regress) § Async callback upon max_batch_size or batch_timeout_micros § BatchTask encapsulates unit of work to be batched § SharedBatchScheduler § Best for heterogeneous request types, multi-step inference, ensembles, … § Groups BatchTasks into separate queues to form homogenous batches § Processes batches fairly through interleaving § StreamingBatchScheduler § Mixed CPU/GPU/IO-bound workloads § Provides fine-grained control for complex, multi-phase inference logic You Must Experiment to Find the Best Strategy for You!! Co-locate and Isolate Homogenous Workloads
LET’S DEPLOY OPTIMIZED MODEL § Navigate to the following notebook: 10_Optimize_Model_Server § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
AGENDA § GPUs and TensorFlow § Train and Debug TensorFlow Model § Train with Distributed TensorFlow Cluster § Optimize Model Training with XLA JIT Compiler § Optimize Model Predicting with XLA AOT and Graph Transforms § Deploy Model to TensorFlow Serving Runtime § Optimize TensorFlow Serving Runtime § Wrap-up and Q&A
YOU HAVE JUST LEARNED… § Part 1: TensorFlow Model Training § TensorFlow and GPUs § Inspect and Debug Models § Distributed Training Across a Cluster § Optimize Training with Queues, Dataset API, and JIT XLA Compiler § Part 2: TensorFlow Model Deploying and Serving § Optimize Predicting with AOT XLA and Graph Transform Tool (GTT) § Deploy Model and Predict § Key Components of TensorFlow Serving § Optimize TensorFlow Serving Runtime
Q&A § Thank you!! § https://github.com/fluxcapacitor/pipeline/ § Slides, code, notebooks, Docker images available here: https://github.com/fluxcapacitor/pipeline/ gpu.ml Contact Me @ Email: chris@pipeline.ai Twitter: @cfregly

More Related Content

High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference

  • 1. HIGH PERFORMANCE TENSORFLOW IN PRODUCTION WITH GPUS! CHRIS FREGLY, FOUNDER @ PIPELINE.AI ML TRAIN, SYDNEY 2017
  • 2. INTRODUCTIONS: ME § Chris Fregly, Research Engineer @ § Formerly Netflix and Databricks § Advanced Spark and TensorFlow Meetup Please Join Our 20,000+ Members Globally! * San Francisco * Chicago * Washington DC * London Please Join!!
  • 3. INTRODUCTIONS: YOU § Software Engineer or Data Scientist interested in optimizing and deploying TensorFlow models to production § Assume you have a working knowledge of TensorFlow
  • 4. 100% OPEN SOURCE CODE § https://github.com/fluxcapacitor/pipeline/ § Please Star this Repo! J § Slides, code, notebooks, Docker images available here: https://github.com/fluxcapacitor/pipeline/ gpu.ml
  • 5. HANDS-ON EXERCISES § Combo of Jupyter Notebooks and Command Line § Command Line through Jupyter Terminal § Some Exercises Based on Experimental Features Warning: You Will See Errors. You will be OK!!
  • 6. CONTENT NOTES § 50% Training Optimizations (GPUs, XLA, JIT) § 50% Predicting Optimizations (XLA, AOT, TF Serving) § Why Heavy Focus on Predicting? § Training: boring batch, O(num_data_scientists) § Inference: exciting realtime, O(num_users_of_app) § We Use Simple Models to Highlight Optimizations Warning: This is not intro material. You will be OK!
  • 7. YOU WILL LEARN… § Part 1: TensorFlow Model Training § TensorFlow and GPUs § Inspect and Debug Models § Distributed Training Across a Cluster § Optimize Training with Queues, Dataset API, and JIT XLA Compiler § Part 2: TensorFlow Model Deploying and Serving § Optimize Predicting with AOT XLA and Graph Transform Tool (GTT) § Deploy Model and Predict § Key Components of TensorFlow Serving § Optimize TensorFlow Serving Runtime
  • 8. AGENDA § GPUs and TensorFlow § Train and Debug TensorFlow Model § Train with Distributed TensorFlow Cluster § Optimize Model Training with XLA JIT Compiler § Optimize Model Predicting with XLA AOT and Graph Transforms § Deploy Model to TensorFlow Serving Runtime § Optimize TensorFlow Serving Runtime § Wrap-up and Q&A
  • 10. SETUP ENVIRONMENT § Step 1: Browse to the following: http://allocator.community.pipeline.ai/allocate § Step 2: Browse to the following: http://<ip-address> § Step 3: Browse around. I will provide a username/password in a bit! Need Help? Use the Chat!
  • 12. LET’S EXPLORE OUR ENVIRONMENT § Navigate to the following notebook: 01_Explore_Environment § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
  • 14. BREAK § https://github.com/fluxcapacitor/pipeline/ § Slides, code, notebooks, Docker images available here: https://github.com/fluxcapacitor/pipeline/ gpu.ml Need Help? Use the Chat!
  • 15. SETTING UP TENSORFLOW WITH GPUS § Very Painful! § Especially inside Docker § Use nvidia-docker § Especially on Kubernetes! § Use Kubernetes 1.7+ § http://pipeline.ai for GitHub + DockerHub Links
  • 16. GPU HALF-PRECISION SUPPORT § FP16, INT8 are “Half Precision” § Supported by Pascal P100 (2016) and Volta V100 (2017) § Flexible FP32 GPU Cores Can Fit 2 FP16’s for 2x Throughput! § Half-Precision is OK for Approximate Deep Learning Use Cases
  • 17. VOLTA V100 RECENTLY ANNOUNCED § 84 Streaming Multiprocessors (SM’s) § 5,376 GPU Cores § 672 Tensor Cores (ie. Google TPU) § Mixed FP16/FP32 Precision § More Shared Memory § New L0 Instruction Cache § Faster L1 Data Cache § V100 vs. P100 Performance § 12x TFLOPS @ Peak Training § 6x Inference Throughput
  • 18. V100 AND CUDA 9 § Independent Thread Scheduling - Finally!! § Similar to CPU fine-grained thread synchronization semantics § Allows GPU to yield execution of any thread § Still Optimized for SIMT (Same Instruction Multiple Thread) § SIMT units automatically scheduled together § Explicit Synchronization P100 V100
  • 19. GPU CUDA PROGRAMMING § Barbaric, But Fun Barbaric! § Must Know Underlying Hardware Very Well § Many Great Debuggers/Profilers § Hardware Changes are Painful! § Newer CUDA versions automatically JIT-compile old CUDA code to new NVPTX § Not optimal, of course
  • 20. CUDA STREAMS § Asynchronous I/O Transfer § Overlap Compute and I/O § Keeps GPUs Saturated § Fundamental to Queue Framework in TensorFlow
  • 21. LET’S SEE WHAT THIS THING CAN DO! § Navigate to the following notebook: 01a_Explore_GPU 01b_Explore_Numba § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
  • 22. AGENDA § GPUs and TensorFlow § Train and Debug TensorFlow Model § Train with Distributed TensorFlow Cluster § Optimize Model Training with XLA JIT Compiler § Optimize Model Predicting with XLA AOT and Graph Transforms § Deploy Model to TensorFlow Serving Runtime § Optimize TensorFlow Serving Runtime § Wrap-up and Q&A
  • 23. TRAINING TERMINOLOGY § Tensors: N-Dimensional Arrays § ie. Scalar, Vector, Matrix § Operations: MatMul, Add, SummaryLog,… § Graph: Graph of Operations (DAG) § Session: Contains Graph(s) § Feeds: Feed inputs into Operation § Fetches: Fetch output from Operation § Variables: What we learn through training § aka “weights”, “parameters” § Devices: Hardware device on which we train -TensorFlow- Trains Variables -User- Fetches Outputs -User- Feeds Inputs -TensorFlow- Performs Operations -TensorFlow- Flows Tensors with tf.device(“worker:0/device/gpu:0,worker:1/device/gpu:0”)
  • 24. TRAINING DEVICES § cpu:0 § By default, all CPUs § Requires extra config to target a CPU § gpu:0..n § Each GPU has a unique id § TF usually prefers a single GPU § xla_cpu:0, xla_gpu:0..n § “JIT Compiler Device” § Hints TensorFlow to attempt JIT Compile with tf.device(“/cpu:0”): with tf.device(“/gpu:0”): with tf.device(“/gpu:1”):
  • 25. TRAINING METRICS: TENSORBOARD § Summary Ops § Event Files /root/tensorboard/linear/<version>/events… § Tags § Organize data within Tensorboard UI loss_summary_op = tf.summary.scalar('loss', loss) merge_all_summary_op = tf.summary.merge_all() summary_writer = tf.summary.FileWriter( '/root/tensorboard/linear/<version>', graph=sess.graph)
  • 26. TRAINING ON EXISTING INFRASTRUCTURE § Data Processing § HDFS/Hadoop § Spark § Containers § Docker § Schedulers § Kubernetes § Mesos <dependency> <groupId>org.tensorflow</groupId> <artifactId>tensorflow-hadoop</artifactId> </dependency> https://github.com/tensorflow/ecosystem
  • 27. TRAINING PIPELINES: QUEUE + DATASET § Don’t Use feed_dict for Production Workloads!! § feed_dict Requires Python <-> C++ Serialization § Retrieval is Single-threaded, Synchronous, SLOW! § Can’t Retrieve Until Current Batch is Complete § CPUs/GPUs Not Fully Utilized! § Use Queue or Dataset API
  • 28. QUEUES § More than Just a Traditional Queue § Perform I/O, pre-processing, cropping, shuffling § Pulls from HDFS, S3, Google Storage, Kafka, ... § Combine many small files into large TFRecord files § Typically use CPUs to focus GPUs on compute § Uses CUDA Streams
  • 29. DATA MOVEMENT WITH QUEUES § GPU Pulls Batch from Queue (CUDA Streams) § GPU pulls next batch while processing current batch GPUs Stay Fully Utilized!
  • 30. QUEUE CAPACITY PLANNING § batch_size § # examples / batch (ie. 64 jpg) § Limited by GPU RAM § num_processing_threads § CPU threads pull and pre-process batches of data § Limited by CPU Cores § queue_capacity § Limited by CPU RAM (ie. 5 * batch_size)
  • 31. DETECT UNDERUTILIZED CPUS, GPUS § Instrument training code to generate “timelines” § Analyze with Google Web Tracing Framework (WTF) § Monitor CPU with `top`, GPU with `nvidia-smi` http://google.github.io/tracing-framework/ from tensorflow.python.client import timeline trace = timeline.Timeline(step_stats=run_metadata.step_stats) with open('timeline.json', 'w') as trace_file: trace_file.write( trace.generate_chrome_trace_format(show_memory=True))
  • 32. LET’S FEED DATA WITH A QUEUE § Navigate to the following notebook: 02_Feed_Queue_HDFS § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
  • 34. BREAK § https://github.com/fluxcapacitor/pipeline/ § Slides, code, notebooks, Docker images available here: https://github.com/fluxcapacitor/pipeline/ gpu.ml Need Help? Use the Chat!
  • 35. TENSORFLOW MODEL § MetaGraph § Combines GraphDef and Metadata § GraphDef § Architecture of your model (nodes, edges) § Metadata § Asset: Accompanying assets to your model § SignatureDef: Maps external : internal tensors § Variables § Stored separately during training (checkpoint) § Allows training to continue from any checkpoint § Variables are “frozen” into Constants when deployed for inference GraphDef x W mul add b MetaGraph Metadata Assets SignatureDef Tags Version Variables: “W” : 0.328 “b” : -1.407
  • 36. TENSORFLOW SESSION Session graph: GraphDef Variables: “W” : 0.328 “b” : -1.407 Variables are Periodically Checkpointed GraphDef is Static
  • 37. LET’S TRAIN A MODEL (CPU) § Navigate to the following notebook: 03_Train_Model_CPU § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
  • 38. LET’S TRAIN A MODEL (GPU) § Navigate to the following notebook: 03a_Train_Model_GPU § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
  • 39. TENSORFLOW DEBUGGER § Step through Operations § Inspect Inputs and Outputs § Wrap Session in Debug Session sess = tf.Session(config=config) sess = tf_debug.LocalCLIDebugWrapperSession(sess)
  • 40. LET’S DEBUG A MODEL § Navigate to the following notebook: 04_Debug_Model § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
  • 41. BATCH NORMALIZATION § Each Mini-Batch May Have Wildly Different Distributions § Normalize per batch (and layer) § Speeds up Training!! § Weights are Learned Quicker § Final Model is More Accurate § Final mean and variance will be folded into Graph later -- Always Use Batch Normalization! -- z = tf.matmul(a_prev, W) a = tf.nn.relu(z) a_mean, a_var = tf.nn.moments(a, [0]) scale = tf.Variable(tf.ones([depth/channels])) beta = tf.Variable(tf.zeros ([depth/channels])) bn = tf.nn.batch_normalizaton(a, a_mean, a_var, beta, scale, 0.001)
  • 42. AGENDA § GPU Environment § Train and Debug TensorFlow Model § Train with Distributed TensorFlow Cluster § Optimize Model Training with XLA JIT Compiler § Optimize Model Predicting with XLA AOT and Graph Transforms § Deploy Model to TensorFlow Serving Runtime § Optimize TensorFlow Serving Runtime § Wrap-up and Q&A
  • 43. MULTI-GPU TRAINING (SINGLE NODE) § Variables stored on CPU (cpu:0) § Model graph (aka “replica”, “tower”) is copied to each GPU(gpu:0, gpu:1, …) Multi-GPU Training Steps: 1. CPU transfers model to each GPU 2. CPU waits on all GPUs to finish batch 3. CPU copies all gradients back from all GPUs 4. CPU synchronizes + AVG all gradients from GPUs 5. CPU updates GPUs with new variables/weights 6. Repeat Step 1 until reaching stop condition (ie. max_epochs)
  • 44. DISTRIBUTED, MULTI-NODE TRAINING § TensorFlow Automatically Inserts Send and Receive Ops into Graph § Parameter Server Synchronously Aggregates Updates to Variables § Nodes with Multiple GPUs will Pre-Aggregate Before Sending to PS Worker0 Worker0 Worker1 Worker0 Worker1 Worker2 gpu0 gpu1 gpu2 gpu3 gpu0 gpu1 gpu2 gpu3 gpu0 gpu1 gpu2 gpu3 gpu0 gpu1 gpu0 gpu0
  • 45. SYNCHRONOUS VS. ASYNCHRONOUS § Synchronous § Nodes compute gradients § Nodes update Parameter Server (PS) § Nodes sync on PS for latest gradients § Asynchronous § Some nodes delay in computing gradients § Nodes don’t update PS § Nodes get stale gradients from PS § May not converge due to stale reads!
  • 46. DATA PARALLEL VS MODEL PARALLEL § Data Parallel (“Between-Graph Replication”) § Send exact same model to each device § Each device operates on its partition of data § ie. Spark sends same function to many workers § Each worker operates on their partition of data § Model Parallel (“In-Graph Replication”) § Send different partition of model to each device § Each device operates on all data Very Difficult!! Required for Large Models. (GPU RAM Limitation)
  • 47. DISTRIBUTED TENSORFLOW CONCEPTS § Client § Program that builds a TF Graph, constructs a session, interacts with the cluster § Written in Python, C++ § Cluster § Set of distributed nodes executing a graph § Nodes can play any role § Jobs (“Roles”) § Parameter Server (“ps”) stores and updates variables § Worker (“worker”) performs compute-intensive tasks (stateless) § Assigned 0..* tasks § Task (“Server Process”) “ps” and “worker” are conventional names
  • 48. CHIEF WORKER § Worker Task 0 is Chosen by Default § Task 0 is guaranteed to exist § Implements Maintenance Tasks § Writes checkpoints § Initializes parameters at start of training § Writes log summaries § Parameter Server health checks
  • 49. NODE AND PROCESS FAILURES § Checkpoint to Persistent Storage (HDFS, S3) § Use MonitoredTrainingSession and Hooks § Use a Good Cluster Orchestrator (ie. Kubernetes,Mesos) § Understand Failure Modes and Recovery States Stateless, Not Bad: Training Continues Stateful, Bad: Training Must Stop Dios Mio! Long Night Ahead…
  • 50. VALIDATING DISTRIBUTED MODEL § Separate Training and Validation Clusters § Validate using Saved Checkpoints from Parameter Servers § Avoids Resource Contention Training Cluster Validation Cluster Parameter Server Cluster
  • 51. LET’S TRAIN WITH DISTRIBUTED CPU § Navigate to the following notebook: 05_Train_Model_Distributed_CPU § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
  • 52. LET’S TRAIN WITH DISTRIBUTED GPU § Navigate to the following notebook: 05a_Train_Model_Distributed_GPU § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
  • 53. NEW(‘ISH): EXPERIMENT + ESTIMATOR § Higher-Level APIs Simplify Distributed Training § Picks Up Configuration from Environment § Supports Custom Models (ie. Keras) § Used for Training, Validation, and Prediction § API is Changing, but Patterns Remain the Same § Works Well with Google Cloud ML (Surprised?!)
  • 55. BREAK § https://github.com/fluxcapacitor/pipeline/ § Slides, code, notebooks, Docker images available here: https://github.com/fluxcapacitor/pipeline/ gpu.ml Need Help? Use the Chat!
  • 56. AGENDA § GPUs and TensorFlow § Train and Debug TensorFlow Model § Train with Distributed TensorFlow Cluster § Optimize Model Training with XLA JIT Compiler § Optimize Model Predicting with XLA AOT and Graph Transforms § Deploy Model to TensorFlow Serving Runtime § Optimize TensorFlow Serving Runtime § Wrap-up and Q&A
  • 57. XLA FRAMEWORK § Accelerated Linear Algebra (XLA) § Goals: § Reduce reliance on custom operators § Improve execution speed § Improve memory usage § Reduce mobile footprint § Improve portability § Helps TF Stay Flexible and Performant
  • 58. XLA HIGH LEVEL OPTIMIZER (HLO) § Compiler Intermediate Representation (IR) § Independent of source and target language § Define Graphs using HLO Language § XLA Step 1 Emits Target-Independent HLO § XLA Step 2 Emits Target-Dependent LLVM § LLVM Emits Native Code Specific to Target § Supports x86-64, ARM64 (CPU), and NVPTX (GPU)
  • 59. JIT COMPILER § Just-In-Time Compiler § Built on XLA Framework § Goals: § Reduce memory movement – especially useful on GPUs § Reduce overhead of multiple function calls § Similar to Spark Operator Fusing in Spark 2.0 § Unroll Loops, Fuse Operators, Fold Constants, … § Scope to session, device, or `with jit_scope():`
  • 60. VISUALIZING JIT COMPILER IN ACTION Before After Google Web Tracing Framework: http://google.github.io/tracing-framework/ from tensorflow.python.client import timeline trace = timeline.Timeline(step_stats=run_metadata.step_stats) with open('timeline.json', 'w') as trace_file: trace_file.write( trace.generate_chrome_trace_format(show_memory=True))
  • 61. VISUALIZING FUSING OPERATORS pip install graphviz dot -Tpng /tmp/hlo_graph_1.w5LcGs.dot -o hlo_graph_1.png GraphViz: http://www.graphviz.org hlo_*.dot files generated by XLA
  • 62. LET’S TRAIN WITH XLA CPU § Navigate to the following notebook: 06_Train_Model_XLA_CPU § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
  • 63. LET’S TRAIN WITH XLA GPU § Navigate to the following notebook: 06a_Train_Model_XLA_GPU § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
  • 64. AGENDA § GPUs and TensorFlow § Train and Debug TensorFlow Model § Train with Distributed TensorFlow Cluster § Optimize Model Training with XLA JIT Compiler § Optimize Model Predicting with XLA AOT and Graph Transforms § Deploy Model to TensorFlow Serving Runtime § Optimize TensorFlow Serving Runtime § Wrap-up and Q&A
  • 65. AOT COMPILER § Standalone, Ahead-Of-Time (AOT) Compiler § Built on XLA framework § tfcompile § Creates executable with minimal TensorFlow Runtime needed § Includes only dependencies needed by subgraph computation § Creates functions with feeds (inputs) and fetches (outputs) § Packaged as cc_libary header and object files to link into your app § Commonly used for mobile device inference graph § Currently, only CPU x86-64 and ARM are supported - no GPU
  • 66. GRAPH TRANSFORM TOOL (GTT) § Optimize Trained Models for Inference § Remove training-only Ops (checkpoint, drop out, logs) § Remove unreachable nodes between given feed -> fetch § Fuse adjacent operators to improve memory bandwidth § Fold final batch norm mean and variance into variables § Round weights/variables improves compression (ie. 70%) § Quantize weights and activations simplifies model § FP32 down to INT8
  • 68. AFTER STRIPPING UNUSED NODES § Optimizations § strip_unused_nodes § Results § Graph much simpler § File size much smaller
  • 69. AFTER REMOVING UNUSED NODES § Optimizations § strip_unused_nodes § remove_nodes § Results § Pesky nodes removed § File size a bit smaller
  • 70. AFTER FOLDING CONSTANTS § Optimizations § strip_unused_nodes § remove_nodes § fold_constants § Results § Placeholders (feeds) -> Variables* (*Why Variables and not Constants?)
  • 71. AFTER FOLDING BATCH NORMS § Optimizations § strip_unused_nodes § remove_nodes § fold_constants § fold_batch_norms § Results § Graph remains the same § File size approximately the same
  • 72. WEIGHT QUANTIZATION § FP16 and INT8 Are Smaller and Computationally Simpler § Weights/Variables are Constants § Easy to Linearly Quantize
  • 73. AFTER QUANTIZING WEIGHTS § Optimizations § strip_unused_nodes § remove_nodes § fold_constants § fold_batch_norms § quantize_weights § Results § Graph is same, file size is smaller, compute is faster
  • 74. LET’S OPTIMIZE FOR INFERENCE § Navigate to the following notebook: 07_Optimize_Model* (*Why just CPU version? Why not both CPU and GPU?) § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
  • 76. ACTIVATION QUANTIZATION § Activations Not Known Ahead of Time § Depends on input, not easy to quantize § Requires Additional Calibration Step § Use a “representative” dataset § Per Neural Network Layer… § Collect histogram of activation values § Generate many quantized distributions with different saturation thresholds § Choose threshold to minimize… KL_divergence(ref_distribution, quant_distribution) § Not Much Time or Data is Required (Minutes on Commodity Hardware)
  • 78. AFTER ACTIVATION QUANTIZATION § Optimizations § strip_unused_nodes § remove_nodes § fold_constants § fold_batch_norms § quantize_weights § quantize_nodes (activations) § Results § Larger graph, needs calibration! Requires additional freeze_requantization_ranges
  • 79. LET’S OPTIMIZE FOR INFERENCE § Navigate to the following notebook: 08_Optimize_Model_Activations § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
  • 80. LINEARIZE GRAPH EXECUTION ORDER § https://github.com/yaroslavvb/stuff Linearize to minimize graph memory usage
  • 81. AGENDA § GPUs and TensorFlow § Train and Debug TensorFlow Model § Train with Distributed TensorFlow Cluster § Optimize Model Training with XLA JIT Compiler § Optimize Model Predicting with XLA AOT and Graph Transforms § Deploy Model to TensorFlow Serving Runtime § Optimize TensorFlow Serving Runtime § Wrap-up and Q&A
  • 82. MODEL SERVING TERMINOLOGY § Inference § Only Forward Propagation through Network § Predict, Classify, Regress, … § Bundle § GraphDef, Variables, Metadata, … § Assets § ie. Map of ClassificationID -> String § {9283: “penguin”, 9284: “bridge”} § Version § Every Model Has a Version Number (Integer) § Version Policy § ie. Serve Only Latest (Highest), Serve Both Latest and Previous, …
  • 83. TENSORFLOW SERVING FEATURES § Supports Auto-Scaling § Custom Loaders beyond File-based § Tune for Low-latency or High-throughput § Serve Diff Models/Versions in Same Process § Customize Models Types beyond HashMap and TensorFlow § Customize Version Policies for A/B and Bandit Tests § Support Request Draining for Graceful Model Updates § Enable Request Batching for Diff Use Cases and HW § Supports Optimized Transport with GRPC and Protocol Buffers
  • 84. PREDICTION SERVICE § Predict (Original, Generic) § Input: List of Tensor § Output: List of Tensor § Classify § Input: List of tf.Example (key, value) pairs § Output: List of (class_label: String, score: float) § Regress § Input: List of tf.Example (key, value) pairs § Output: List of (label: String, score: float)
  • 85. PREDICTION INPUTS + OUTPUTS § SignatureDef § Defines inputs and outputs § Maps external (logical) to internal (physical) tensor names § Allows internal (physical) tensor names to change from tensorflow.python.saved_model import utils from tensorflow.python.saved_model import signature_constants from tensorflow.python.saved_model import signature_def_utils graph = tf.get_default_graph() x_observed = graph.get_tensor_by_name('x_observed:0') y_pred = graph.get_tensor_by_name('add:0') inputs_map = {'inputs': x_observed} outputs_map = {'outputs': y_pred} predict_signature = signature_def_utils.predict_signature_def(inputs=inputs_map, outputs=outputs_map)
  • 86. MULTI-HEADED INFERENCE § Multiple “Heads” of Model § Return class and scores to be fed into another model § Inputs Propagated Forward Only Once § Optimizes Bandwidth, CPU, Latency, Memory, Coolness
  • 87. BUILD YOUR OWN MODEL SERVER (?!) § Adapt GRPC(Google) <-> HTTP (REST of the World) § Perform Batch Inference vs. Request/Response § Handle Requests Asynchronously § Support Mobile, Embedded Inference § Customize Request Batching § Add Circuit Breakers, Fallbacks § Control Latency Requirements § Reduce Number of Moving Parts #include “tensorflow_serving/model_servers/server_core.h” class MyTensorFlowModelServer { ServerCore::Options options; // set options (model name, path, etc) std::unique_ptr<ServerCore> core; TF_CHECK_OK( ServerCore::Create(std::move(options), &core) ); } Compile and Link libtensorflow.so
  • 88. FREEZING MODEL FOR DEPLOYMENT § Optimizations § strip_unused_nodes § remove_nodes § fold_constants § fold_batch_norms § quantize_weights § quantize_nodes § freeze_graph § Results § Variables -> Constants Finally! We’re Ready to Deploy!!
  • 89. LET’S DEPLOY OPTIMIZED MODEL § Navigate to the following notebook: 09_Deploy_Optimized_Model § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
  • 90. AGENDA § GPUs and TensorFlow § Train and Debug TensorFlow Model § Train with Distributed TensorFlow Cluster § Optimize Model Training with XLA JIT Compiler § Optimize Model Predicting with XLA AOT and Graph Transforms § Deploy Model to TensorFlow Serving Runtime § Optimize TensorFlow Serving Runtime § Wrap-up and Q&A
  • 91. REQUEST BATCH TUNING § max_batch_size § Enables throughput/latency tradeoff § Bounded by RAM § batch_timeout_micros § Defines batch time window, latency upper-bound § Bounded by RAM § num_batch_threads § Defines parallelism § Bounded by CPU cores § max_enqueued_batches § Defines queue upper bound, throttling § Bounded by RAM Reaching either threshold will trigger a batch
  • 92. BATCH SCHEDULER STRATEGIES § BasicBatchScheduler § Best for homogeneous request types (ie. always classify or always regress) § Async callback upon max_batch_size or batch_timeout_micros § BatchTask encapsulates unit of work to be batched § SharedBatchScheduler § Best for heterogeneous request types, multi-step inference, ensembles, … § Groups BatchTasks into separate queues to form homogenous batches § Processes batches fairly through interleaving § StreamingBatchScheduler § Mixed CPU/GPU/IO-bound workloads § Provides fine-grained control for complex, multi-phase inference logic You Must Experiment to Find the Best Strategy for You!! Co-locate and Isolate Homogenous Workloads
  • 93. LET’S DEPLOY OPTIMIZED MODEL § Navigate to the following notebook: 10_Optimize_Model_Server § https://github.com/fluxcapacitor/pipeline/gpu.ml/ notebooks/
  • 94. AGENDA § GPUs and TensorFlow § Train and Debug TensorFlow Model § Train with Distributed TensorFlow Cluster § Optimize Model Training with XLA JIT Compiler § Optimize Model Predicting with XLA AOT and Graph Transforms § Deploy Model to TensorFlow Serving Runtime § Optimize TensorFlow Serving Runtime § Wrap-up and Q&A
  • 95. YOU HAVE JUST LEARNED… § Part 1: TensorFlow Model Training § TensorFlow and GPUs § Inspect and Debug Models § Distributed Training Across a Cluster § Optimize Training with Queues, Dataset API, and JIT XLA Compiler § Part 2: TensorFlow Model Deploying and Serving § Optimize Predicting with AOT XLA and Graph Transform Tool (GTT) § Deploy Model and Predict § Key Components of TensorFlow Serving § Optimize TensorFlow Serving Runtime
  • 96. Q&A § Thank you!! § https://github.com/fluxcapacitor/pipeline/ § Slides, code, notebooks, Docker images available here: https://github.com/fluxcapacitor/pipeline/ gpu.ml Contact Me @ Email: chris@pipeline.ai Twitter: @cfregly