Download - High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference
![Page 1: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/1.jpg)
HIGH PERFORMANCE TENSORFLOW IN PRODUCTION WITH GPUS!CHRIS FREGLY,FOUNDER @ PIPELINE.AIML TRAIN, SYDNEY 2017
![Page 2: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/2.jpg)
INTRODUCTIONS: ME§ Chris Fregly, Research Engineer @
§ Formerly Netflix and Databricks
§ Advanced Spark and TensorFlow MeetupPlease Join Our 20,000+ Members Globally!
* San Francisco* Chicago* Washington DC* London
Please Join!!
![Page 3: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/3.jpg)
INTRODUCTIONS: YOU
§ Software Engineer or Data Scientist interested in optimizing and deploying TensorFlow models to production
§ Assume you have a working knowledge of TensorFlow
![Page 4: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/4.jpg)
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
![Page 5: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/5.jpg)
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!!
![Page 6: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/6.jpg)
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!
![Page 7: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/7.jpg)
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
![Page 8: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/8.jpg)
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
![Page 9: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/9.jpg)
EVERYBODY GETS A GPU!
![Page 10: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/10.jpg)
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!
![Page 11: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/11.jpg)
VERIFY SETUP
http://<ip-address>
Any username,Any password!
![Page 12: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/12.jpg)
LET’S EXPLORE OUR ENVIRONMENT§ Navigate to the following notebook:
01_Explore_Environment
§ https://github.com/fluxcapacitor/pipeline/gpu.ml/notebooks/
![Page 13: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/13.jpg)
PULSE CHECK
![Page 14: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/14.jpg)
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!
![Page 15: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/15.jpg)
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
![Page 16: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/16.jpg)
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
![Page 17: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/17.jpg)
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
![Page 18: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/18.jpg)
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
![Page 19: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/19.jpg)
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
![Page 20: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/20.jpg)
CUDA STREAMS
§ Asynchronous I/O Transfer§ Overlap Compute and I/O§ Keeps GPUs Saturated§ Fundamental to Queue Framework in TensorFlow
![Page 21: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/21.jpg)
LET’S SEE WHAT THIS THING CAN DO!§ Navigate to the following notebook:
01a_Explore_GPU01b_Explore_Numba
§ https://github.com/fluxcapacitor/pipeline/gpu.ml/notebooks/
![Page 22: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/22.jpg)
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
![Page 23: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/23.jpg)
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-FetchesOutputs
-User-FeedsInputs
-TensorFlow-Performs
Operations
-TensorFlow-Flows
Tensors
with tf.device(“worker:0/device/gpu:0,worker:1/device/gpu:0”)
![Page 24: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/24.jpg)
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”):
![Page 25: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/25.jpg)
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)
![Page 26: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/26.jpg)
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
![Page 27: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/27.jpg)
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
![Page 28: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/28.jpg)
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
![Page 29: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/29.jpg)
DATA MOVEMENT WITH QUEUES§ GPU Pulls Batch from Queue (CUDA Streams)§ GPU pulls next batch while processing current batch
GPUs Stay Fully Utilized!
![Page 30: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/30.jpg)
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)
![Page 31: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/31.jpg)
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))
![Page 32: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/32.jpg)
LET’S FEED DATA WITH A QUEUE§ Navigate to the following notebook:
02_Feed_Queue_HDFS
§ https://github.com/fluxcapacitor/pipeline/gpu.ml/notebooks/
![Page 33: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/33.jpg)
PULSE CHECK
![Page 34: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/34.jpg)
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!
![Page 35: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/35.jpg)
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
MetaGraphMetadata
AssetsSignatureDef
TagsVersion
Variables:“W” : 0.328“b” : -1.407
![Page 36: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/36.jpg)
TENSORFLOW SESSION
Session
graph: GraphDef
Variables:“W” : 0.328“b” : -1.407
Variables arePeriodically
Checkpointed
GraphDefis Static
![Page 37: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/37.jpg)
LET’S TRAIN A MODEL (CPU)§ Navigate to the following notebook:
03_Train_Model_CPU
§ https://github.com/fluxcapacitor/pipeline/gpu.ml/notebooks/
![Page 38: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/38.jpg)
LET’S TRAIN A MODEL (GPU)§ Navigate to the following notebook:
03a_Train_Model_GPU
§ https://github.com/fluxcapacitor/pipeline/gpu.ml/notebooks/
![Page 39: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/39.jpg)
TENSORFLOW DEBUGGER§ Step through Operations§ Inspect Inputs and Outputs§ Wrap Session in Debug Session
sess = tf.Session(config=config)sess =
tf_debug.LocalCLIDebugWrapperSession(sess)
![Page 40: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/40.jpg)
LET’S DEBUG A MODEL§ Navigate to the following notebook:
04_Debug_Model
§ https://github.com/fluxcapacitor/pipeline/gpu.ml/notebooks/
![Page 41: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/41.jpg)
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)
![Page 42: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/42.jpg)
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
![Page 43: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/43.jpg)
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 GPU2. CPU waits on all GPUs to finish batch3. CPU copies all gradients back from all GPUs4. CPU synchronizes + AVG all gradients from GPUs5. CPU updates GPUs with new variables/weights6. Repeat Step 1 until reaching stop condition (ie. max_epochs)
![Page 44: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/44.jpg)
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
![Page 45: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/45.jpg)
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!
![Page 46: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/46.jpg)
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)
![Page 47: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/47.jpg)
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
![Page 48: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/48.jpg)
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
![Page 49: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/49.jpg)
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…
![Page 50: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/50.jpg)
VALIDATING DISTRIBUTED MODEL
§ Separate Training and Validation Clusters § Validate using Saved Checkpoints from Parameter Servers§ Avoids Resource Contention
TrainingCluster
ValidationCluster
Parameter ServerCluster
![Page 51: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/51.jpg)
LET’S TRAIN WITH DISTRIBUTED CPU§ Navigate to the following notebook:
05_Train_Model_Distributed_CPU
§ https://github.com/fluxcapacitor/pipeline/gpu.ml/notebooks/
![Page 52: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/52.jpg)
LET’S TRAIN WITH DISTRIBUTED GPU§ Navigate to the following notebook:
05a_Train_Model_Distributed_GPU
§ https://github.com/fluxcapacitor/pipeline/gpu.ml/notebooks/
![Page 53: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/53.jpg)
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?!)
![Page 54: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/54.jpg)
PULSE CHECK
![Page 55: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/55.jpg)
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!
![Page 56: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/56.jpg)
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
![Page 57: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/57.jpg)
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
![Page 58: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/58.jpg)
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)
![Page 59: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/59.jpg)
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():`
![Page 60: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/60.jpg)
VISUALIZING JIT COMPILER IN ACTION
Before After
Google Web Tracing Framework:http://google.github.io/tracing-framework/
from tensorflow.python.client import timelinetrace = 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))
![Page 61: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/61.jpg)
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
![Page 62: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/62.jpg)
LET’S TRAIN WITH XLA CPU§ Navigate to the following notebook:
06_Train_Model_XLA_CPU
§ https://github.com/fluxcapacitor/pipeline/gpu.ml/notebooks/
![Page 63: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/63.jpg)
LET’S TRAIN WITH XLA GPU§ Navigate to the following notebook:
06a_Train_Model_XLA_GPU
§ https://github.com/fluxcapacitor/pipeline/gpu.ml/notebooks/
![Page 64: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/64.jpg)
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
![Page 65: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/65.jpg)
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
![Page 66: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/66.jpg)
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
![Page 67: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/67.jpg)
BEFORE OPTIMIZATIONS
![Page 68: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/68.jpg)
AFTER STRIPPING UNUSED NODES
§ Optimizations§ strip_unused_nodes
§ Results§ Graph much simpler§ File size much smaller
![Page 69: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/69.jpg)
AFTER REMOVING UNUSED NODES
§ Optimizations§ strip_unused_nodes§ remove_nodes
§ Results§ Pesky nodes removed§ File size a bit smaller
![Page 70: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/70.jpg)
AFTER FOLDING CONSTANTS
§ Optimizations§ strip_unused_nodes§ remove_nodes§ fold_constants
§ Results§ Placeholders (feeds) -> Variables*
(*Why Variables and not Constants?)
![Page 71: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/71.jpg)
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
![Page 72: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/72.jpg)
WEIGHT QUANTIZATION
§ FP16 and INT8 Are Smaller and Computationally Simpler§ Weights/Variables are Constants§ Easy to Linearly Quantize
![Page 73: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/73.jpg)
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
![Page 74: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/74.jpg)
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/
![Page 75: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/75.jpg)
BUT WAIT, THERE’S MORE!
![Page 76: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/76.jpg)
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)
![Page 77: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/77.jpg)
ACTIVATION QUANTIZATION GRAPH
CreateConversionSubgraph
ProducesQuantizedMatMul,
QuantizedRelu
Eliminate AdjacentDequantize +
Quantize
![Page 78: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/78.jpg)
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
![Page 79: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/79.jpg)
LET’S OPTIMIZE FOR INFERENCE§ Navigate to the following notebook:
08_Optimize_Model_Activations
§ https://github.com/fluxcapacitor/pipeline/gpu.ml/notebooks/
![Page 80: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/80.jpg)
LINEARIZE GRAPH EXECUTION ORDER
§ https://github.com/yaroslavvb/stuff
Linearize to minimize graphmemory usage
![Page 81: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/81.jpg)
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
![Page 82: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/82.jpg)
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, …
![Page 83: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/83.jpg)
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
![Page 84: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/84.jpg)
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)
![Page 85: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/85.jpg)
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 utilsfrom tensorflow.python.saved_model import signature_constantsfrom 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)
![Page 86: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/86.jpg)
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
![Page 87: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/87.jpg)
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 Linklibtensorflow.so
![Page 88: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/88.jpg)
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!!
![Page 89: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/89.jpg)
LET’S DEPLOY OPTIMIZED MODEL§ Navigate to the following notebook:
09_Deploy_Optimized_Model
§ https://github.com/fluxcapacitor/pipeline/gpu.ml/notebooks/
![Page 90: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/90.jpg)
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
![Page 91: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/91.jpg)
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 thresholdwill trigger a batch
![Page 92: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/92.jpg)
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
![Page 93: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/93.jpg)
LET’S DEPLOY OPTIMIZED MODEL§ Navigate to the following notebook:
10_Optimize_Model_Server
§ https://github.com/fluxcapacitor/pipeline/gpu.ml/notebooks/
![Page 94: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/94.jpg)
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
![Page 95: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/95.jpg)
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
![Page 96: High Performance TensorFlow in Production -- Sydney ML / AI Train Workshop @ UAI Conference](https://reader033.vdocuments.us/reader033/viewer/2022061307/5a654d377f8b9a182a8b4bc3/html5/thumbnails/96.jpg)
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: [email protected]
Twitter: @cfregly