Optimizing, Profiling, and Deploying TensorFlow AI Models with GPUs - San Francisco Python Meetup - Nov 8, 2017

HIGH PERFORMANCE TENSORFLOW IN
PRODUCTION WITH GPUS
SF PYTHON MEETUP NOV 8, 2017
SPECIAL THANKS TO YELP!! !!
CHRIS FREGLY, FOUNDER @PIPELINE.AI

INTRODUCTIONS: ME
§ Chris Fregly, Founder & Engineer @ PipelineAI
§ Formerly Netflix and Databricks
§ Advanced Spark and TensorFlow Meetup
Please Join Our 50,000+ Global Members!!
Contact Me
chris@pipeline.ai
@cfregly
* San Francisco
* Chicago
* Austin
* Washington DC
* Dusseldorf
* London

INTRODUCTIONS: YOU
§ Software Engineer, Data Scientist, Data Engineer, Data Analyst
§ Interested in Optimizing and Deploying TF Models to Production
§ Nice to Have a Working Knowledge of TensorFlow (Not Required)

CONTENT BREAKDOWN
50% Training Optimizations (GPUs, Training Pipeline, JIT)
50% Prediction Optimizations (AOT Compile, TF Serving)
Why Heavy Focus on Model Prediction vs. Just Training?
10s of Data Scientists <<< Millions of App Users
Training
Boring & Batch
Prediction
Exciting & Real-Time!!

AGENDA
Part 0: Latest PipelineAI Research
Part 1: Optimize TensorFlow Model Training
Part 2: Optimize TensorFlow Model Serving

100% OPEN SOURCE CODE
§ https://github.com/PipelineAI/pipeline/
§ Please 🌟 this GitHub Repo!
§ All slides, code, notebooks, and Docker images here:
https://github.com/PipelineAI/pipeline/tree/master/gpu.ml
https://github.com/rviscomi/red-dwarf

HANDS-ON EXERCISES
§ Combo of Jupyter Notebooks and Command Line
§ Command Line through Jupyter Terminal
§ Some Exercises Based on Experimental Features
You May See Errors. Stay Calm. You Will Be OK!!

PIPELINE.AI OVERVIEW
400,000 Docker Downloads
50,000 Users registered for
PipelineAI GA Release
2,000 GitHub Stars
15 Enterprise Beta Users

AGENDA
Part 0: Latest PipelineAI Research
§ Package, Deploy, and Tune Both Model + Runtime
§ Deploy Models and Experiments Safely to Prod
§ Compare Models Both Offline and Online
§ Auto-Shift Traffic to Winning Model or Cloud

PACKAGE MODEL + RUNTIME AS ONE
§ Package Model + Runtime into Immutable Docker Image
§ Same Environment: Local, Dev, and Prod
§ No Dependency Surprises in Production
§ Deploy and Tune Model + Runtime Together
pipeline predict-server-build --model-type=tensorflow
--model-name=mnist
--model-tag=”c"
--model-path=./models/tensorflow/mnist
Package Model
Server C Locally
pipeline predict-server-push --model-type=tensorflow
--model-name=mnist
--model-tag=”c”
Push Image C To
Docker Registry

TUNE MODEL + RUNTIME TOGETHER
§ Try Different Model Hyper-Parameters + Runtime Configs
§ Even Different Runtimes: TF Serving, TensorRT
§ Auto-Quantize Model Weights + Activations
§ Auto-Fuse Neural Network Layers Together
§ Generate Native CPU + GPU Code
pipeline predict-server-start --model-type=tensorflow
--model-name=mnist
--model-tag=”c"
Start Model
Server C Locally

LOAD TEST MODEL + RUNTIME LOCALLY
§ Perform Mini-Load Test on Local Model Server
§ Provides Immediate Feedback on Prediction Performance
§ Relative Performance Compared to Other Variations
§ No Need to Deploy to Test or Prod for Prediction Metrics
§ See Where Time is Being Spent During Prediction
pipeline predict --model-server-url=http://localhost:6969
--model-type=tensorflow
--model-name=mnist
--model-tag=”c”
--test-request-concurrency=1000
Load Test Model
Server C Locally

RUNTIME OPTION: NVIDIA TENSOR-RT
§ GPU-Optimized Prediction Runtime
§ Alternative to TensorFlow Serving
§ Post-Training Model Optimizations
§ Similar to TF Graph Transform Tool
§ PipelineAI Supports TensorRT!

DEPLOY MODELS SAFELY TO PROD
§ Deploy from Jupyter Notebook in 1-Click
§ Deploy to 1-2% Split or Shadowed Traffic
§ Tear-Down or Rollback Quickly
§ Command Line Interface (CLI)
pipeline predict-cluster-start --model-type=tensorflow
--model-name=mnist
--model-tag=”b”
--traffic-split=“0.02”
Start Model
Cluster B in Prod
--model-name=mnist
--model-tag=”c”
Start Model
Cluster C in Prod
--model-name=mnist
--model-tag=”a”
Start Model
Cluster A in Prod
Implementation Details…
No need to understand these.
We do, so you don’t have to! J

DEPLOY EXPERIMENTS SAFELY TO PROD
§ Create Experiments Directly from Jupyter or Command Line
§ Deploy Experiment
pipeline experiment-add --experiment-name=my_experiment
--model-name=mnist
--model-tag=“a”
--traffic-split=“97%”
CLI
Drag n’
Drop
pipeline experiment-start --experiment-name=my_experiment
--traffic-shadow=“20%”
--model-name=mnist
--model-tag=“b”
--model-name=mnist
--model-tag=“c”
1-Click
Start Experiment
with 20% Shadowed
of Production Traffic

COMPARE MODELS OFFLINE & ONLINE
§ Offline, Batch Metrics
§ Validation Accuracy
§ Training Accuracy
§ CPU/GPU Utilization
§ Actual, Live Predictions!
§ Relative Prediction Precision
§ Online, Real-Time Metrics
§ Response Time & Throughput
§ Cost Per Prediction!

PREDICTION PROFILING AND TUNING
§ Pinpoint Performance Bottlenecks
§ Fine-Grained Prediction Metrics
§ Three (3) Logic Prediction Steps
1. transform_request()
2. predict()
3. transform_response()

VIEW REAL-TIME PREDICTION STREAM
§ Visually Compare Real-Time Predictions
Feature
Inputs
Prediction
Confidence

CONTINUOUS MODEL TRAINING
§ Identify and Fix Borderline Predictions (~50-50% Confidence)
§ Fix Along Class Boundaries
§ Retrain on New Labeled Data
§ Game-ify Labeling Process
§ Enables Crowd Sourcing

SHIFT TRAFFIC TO MAX(REVENUE)
§ Shift Traffic to Winning Model using AI Bandit Algorithms

SHIFT TRAFFIC TO MIN(CLOUD COST)
§ Real-Time Cost Per Prediction
§ Across Clouds & On-Premise
§ Bandit-based Explore/Exploit

AGENDA
Part 1: Optimize TensorFlow Model Training
§ GPUs and TensorFlow
§ Train, Inspect, and Debug TensorFlow Models
§ TensorFlow Distributed Model Training on a Cluster
§ Optimize Training with JIT XLA Compiler

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 Jupyter Username/Password soon.
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/PipelineAI/pipeline/tree/master/
gpu.ml/notebooks

BREAK
Need Help?
Use the Chat!

SETTING UP TENSORFLOW WITH GPUS
§ Very Painful!
§ Especially inside Docker
§ Use nvidia-docker
§ Especially on Kubernetes!
§ Use Kubernetes 1.8+
§ http://pipeline.ai for GitHub + DockerHub Links

GPU HALF-PRECISION SUPPORT
§ FP32 is “Full Precision”, FP16 is “Half Precision”
§ Supported by Pascal P100 (2016) and Volta V100 (2017)
§ Half-Precision is OK for Approximate Deep Learning Use Cases
§ Fit Two(2) FP16’s into FP32 GPU Cores for 2x Throughput!
You Can Set
TF_FP16_MATMUL_USE_FP32_COMPUTE=0
on GPU w/ Compute Capability(CC) 5.3+

VOLTA V100 (2017) VS. PASCAL P100 (2016)
§ 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 Hardware Very Well
§ Hardware Changes are Painful
§ Use the Profilers & Debuggers

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!
01a_Explore_GPU
01b_Explore_Numba
gpu.ml/notebooks

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 Placeholder
§ Fetches: Fetch Output from Operation
§ Variables: What We Learn Through Training
§ aka “Weights”, “Parameters”
§ Devices: Hardware Device (GPU, CPU, TPU, ...)
-TensorFlow-
Trains
Variables
-User-
Fetches
Outputs
-User-
Feeds
Inputs
-TensorFlow-
Performs
Operations
-TensorFlow-
Flows
Tensors
with tf.device(“/cpu:0,/gpu:15”):

TENSORFLOW SESSION
Session
graph: GraphDef
Variables:
“W” : 0.328
“b” : -1.407
Variables are
Randomly
Initialized, then
Periodically
Checkpointed
GraphDef is
Created
During
Training, then
Frozen for
Inference

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 preparing for inference
GraphDef
x
W
mul add
b
MetaGraph
Metadata
Assets
SignatureDef
Tags
Version
Variables:
“W” : 0.328
“b” : -1.407

BATCH NORMALIZATION (2015)
§ Each Mini-Batch May Have Wildly Different Distributions
§ Normalize per Batch (and Layer)
§ Faster Training, Learns Quicker
§ Final Model is More Accurate
§ TensorFlow is already on 2nd Generation Batch Algorithm
§ First-Class Support for Fusing Batch Norm Layers
§ Final mean + variance Are Folded Into Our Graph Later
-- (Almost)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)

DROPOUT (2014)
§ Training Technique
§ Prevents Overfitting
§ Combines Exponential Set of Diff Neural Architectures
§ Inherent Ensembling, If You Will
§ Expressed as Probability Percentage (ie. 50%)
§ Boost Other Weights During Validation & Prediction
Training
Dropout
Validation &
Prediction
Dropout
0%
Dropout
50%
Dropout

OPTIMIZE GRAPH EXECUTION ORDER
§ https://github.com/yaroslavvb/stuff
"Linearize” Causes
TF to Minimize
Graph
Memory Usage.
This is Useful on
Single GPU with
Relatively Low
RAM.

EXTEND EXISTING DATA PIPELINES
§ 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

DON’T USE FEED_DICT!!
§ feed_dict Requires Python <-> C++ Serialization
§ Not Optimized for Production Ingestion Pipelines
§ Retrieves Next Batch After Current Batch is Done
§ Single-Threaded, Synchronous
§ CPUs/GPUs Not Fully Utilized!
§ Use Queue or Dataset API
sess.run(train_step, feed_dict={…}

QUEUES
§ More than traditional Queue
§ Uses CUDA Streams
§ Perform I/O, pre-processing, cropping, shuffling, …
§ Pull from HDFS, S3, Google Storage, Kafka, ...
§ Combine many small files into large TFRecord files
§ Use CPUs to free GPUs for compute
§ Helps saturate CPUs and GPUs

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
02_Feed_Queue_HDFS
gpu.ml/notebooks

LET’S TRAIN A MODEL (CPU)
03_Train_Model_CPU
gpu.ml/notebooks

LET’S TRAIN A MODEL (GPU)
03a_Train_Model_GPU
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
04_Debug_Model
gpu.ml/notebooks

SINGLE NODE, MULTI-GPU TRAINING
§ 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”):
GPU 0 GPU 1

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
Worker
0
Worker
1
Worker
2
gpu0 gpu1
gpu2 gpu3
gpu0 gpu1
gpu2 gpu3
gpu0 gpu1
gpu2 gpu3
gpu0
gpu1
gpu0
gpu0
Single
Node
Multiple
Nodes

DATA PARALLEL VS MODEL PARALLEL
§ Data Parallel (“Between-Graph Replication”)
§ Send exact same model to each device
§ Each device operates on 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!, But
Required for Large Models.
(GPU RAM Limitation)

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!

CHIEF WORKER
§ Worker Task 0 is Usually the Chief
§ Task 0 is guaranteed to exist
§ Performs Maintenance Tasks
§ Writes log summaries
§ Instructs PS to checkpoint vars
§ Performs PS health checks
§ (Re-)Initialize variables at (re-)start of training

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…

USE ESTIMATOR AND EXPERIMENT APIS
§ Simplify Model Building
§ Provide Clear Path to Production
§ Enable Rapid Model Experiments
§ Provide Flexible Parameter Tuning
§ Enable Downstream Optimizing & Serving Infra( )
§ Nudge Users to Best Practices Through Opinions
§ Provide Hooks/Callbacks to Override Opinions
§ Unified API for Local and Distributed TensorFlow
https://arxiv.org/pdf/1708.02637.pdf

ESTIMATOR API
§ “Train-to-Serve” Design
§ Create Custom - or Use a Canned Estimator
§ Hides Session, Graph, Layers, Iterative Loops (Train, Eval, Predict)
§ Hooks for All Phases of Model Training and Evaluation
§ Load Input: input_fn()
§ Train: model_fn() and train()
§ Evaluate: evaluate()
§ Save and Export: export_savedmodel()
§ Predict: predict() Uses sess.run() Slow Predictions!
Example:
https://github.com/GoogleCloudPlatform/
cloudml-samples/blob/master/census/
customestimator/

LAYERS API
§ Standalone Layer or Entire Sub-Graphs
§ Functions of Tensor Inputs & Outputs
§ Mix and Match with Operations
§ Assumes 1st Dimension is Batch Size
§ Handles One (1) to Many (*) Inputs
§ Special Types of Layers
§ Loss per Mini-Batch
§ Accuracy and MSE Track Across Mini-Batches

CANNED ESTIMATORS
§ Commonly-Used Estimators
§ Pre-Tested and Pre-Tuned
§ DNNClassifer, TensorForestEstimator
§ Always Use Canned Estimators If Possible
§ Reduced Lines of Code, Complexity, and Bugs
§ Use FeatureColumns to Define & Create Features
Custom vs. Canned
@ Google, August, 2017

FEATURECOLUMN ABSTRACTION
§ Used by Canned Estimator
§ Simplifies Input Ingestion
§ Declarative Way to Specify Model Training Inputs
§ Converts Sparse Features to Dense Tensors
§ Sparse Features: Query Keyword, Url, ProductID,…
§ Wide/Linear Models Use Feature-Crossing
§ Deep Models Use Embeddings

SINGLE VS. MULTI-OBJECTIVES + HEADS
§ Single-Objective Estimator
§ Single classification prediction
§ Multi-Objective Estimator
§ Two (2) classification predictions
§ Or One (1) classification prediction + One(1) final layer
§ Multiple Heads Are Used to Ensemble Models
§ Treats neural network as a feature engineering step!

EXPERIMENT API
§ Easier-to-Use Distributed TensorFlow
§ Combines Estimator with input_fn()
§ Used for Training, Evaluation, & Hyper-Parameter Tuning
§ Distributed Training Default to Data-Parallel & Async
§ Cluster Configuration is Fixed at Start of Training Job
§ No Auto-Scaling Allowed!!

ESTIMATOR & EXPERIMENT CONFIGS
§ TF_CONFIG
§ Special environment variable for config
§ Defines ClusterSpec in JSON incl. master, workers, PS’s
§ Must set ”{environment”:“cloud”} for distributed mode
§ RunConfig: Defines checkpoint interval, output directory, …
§ HParams: Hyper-parameter tuning parameters and ranges
§ learn_runner creates RunConfig before calling run() & tune()
§ schedule is set based on {”task”:{”type”}}
§ Set to train_and_evaluate for local, single-node training
TF_CONFIG=
'{
"environment": "cloud",
"cluster":
{
"master":["worker0:2222”],
"worker":["worker1:2222"],
"ps": ["ps0:2222"]
},
"task": {"type": "ps",
"index": "0"}
}'

SEPARATE TRAINING + EVALUATION
§ Separate Training and Evaluation Clusters
§ Evaluate Upon Checkpoint
§ Avoid Resource Contention
§ Let Training Continue in Parallel with Evaluation
Training
Cluster
Evaluation
Cluster
Parameter Server
Cluster

LET’S TRAIN DISTRIBUTED TENSORFLOW
05_Train_Model_Distributed_CPU
or 05a_Train_Model_Distributed_GPU
gpu.ml/notebooks

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
06_Train_Model_XLA_CPU
gpu.ml/notebooks

LET’S TRAIN WITH XLA GPU
06a_Train_Model_XLA_GPU
gpu.ml/notebooks

AGENDA
Part 2: Optimize TensorFlow Model Serving
§ AOT XLA Compiler and Graph Transform Tool
§ Key Components of TensorFlow Serving
§ Deploy Optimized TensorFlow Model
§ Optimize TensorFlow Serving Runtime

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)
§ Post-Training Optimization to Prepare 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 to improve compression (ie. 70%)
§ Quantize (FP32 -> INT8) to Speed Up Math Operations

GRAPH TRANSFORM TOOL
transform_graph
--in_graph=tensorflow_inception_graph.pb ß Original Graph
--out_graph=optimized_inception_graph.pb ß Transformed Graph
--inputs='Mul' ß Feed (Input)
--outputs='softmax' ß Fetch (Output)
--transforms=' ß List of Transforms
strip_unused_nodes
remove_nodes(op=Identity, op=CheckNumerics)
fold_constants(ignore_errors=true)
fold_batch_norms
fold_old_batch_norms
quantize_weights
quantize_nodes'

AFTER STRIPPING UNUSED NODES
§ Optimizations
§ strip_unused_nodes
§ Results
§ Graph much simpler
§ File size much smaller

AFTER REMOVING UNUSED NODES
§ Optimizations
§ remove_nodes
§ Results
§ Pesky nodes removed
§ File size a bit smaller

AFTER FOLDING CONSTANTS
§ Optimizations
§ remove_nodes
§ fold_constants
§ Results
§ Placeholders (feeds) -> Variables*
(*Why Variables and not Constants?)

AFTER FOLDING BATCH NORMS
§ Optimizations
§ remove_nodes
§ fold_constants
§ fold_batch_norms
§ Results
§ Graph remains the same
§ File size approximately the same

AFTER QUANTIZING WEIGHTS
§ Optimizations
§ remove_nodes
§ fold_constants
§ fold_batch_norms
§ quantize_weights
§ Results
§ Graph is same, file size is smaller, compute is faster

WEIGHT QUANTIZATION
§ FP16 and INT8 Are Smaller and Computationally Simpler
§ Weights/Variables are Constants
§ Easy to Linearly Quantize

LET’S OPTIMIZE FOR INFERENCE
07_Optimize_Model*
*Why just CPU version? Why not GPU?
gpu.ml/notebooks

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)

AFTER ACTIVATION QUANTIZATION
§ Optimizations
§ 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
08_Optimize_Model_Activations
gpu.ml/notebooks

FREEZING MODEL FOR DEPLOYMENT
§ Optimizations
§ remove_nodes
§ fold_constants
§ fold_batch_norms
§ quantize_weights
§ quantize_nodes
§ freeze_graph
§ Results
§ Variables -> Constants
Finally!
We’re Ready to Deploy!!

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
§ Inputs Pass Through Model Once
§ Model Returns Multiple Predictions or “Heads” including:
1. Human-readable prediction (ie. “penguin”, “church”,…)
2. Final layer of scores (float vector)
§ Final Layer of floats Pass to the Next Model in Ensemble
§ Optimizes Bandwidth, CPU/GPU, Latency, Memory
§ Enables Complex Model Composing and Ensembling

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 with
libtensorflow.so

NVIDIA TENSOR-RT RUNTIME
§ Post-Training Model Optimizations
§ Similar to TF Graph Transform Tool
§ GPU-Optimized Prediction Runtime
§ Alternative to TensorFlow Serving
§ PipelineAI Supports TensorRT!

SAVED MODEL FORMAT
09_Deploy_Optimized_Model
gpu.ml/notebooks

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

ADVANCED BATCHING & SERVING TIPS
§ Batch Just the GPU/TPU Portions of the Computation Graph
§ Batch Arbitrary Sub-Graphs using Batch / Unbatch Graph Ops
§ Distribute Large Models Into Shards Across TensorFlow Model Servers
§ Batch RNNs Used for Sequential and Time-Series Data
§ Find Best Batching Strategy For Your Data Through Experimentation
§ BasicBatchScheduler: Homogeneous requests (ie Regress or Classify)
§ SharedBatchScheduler: Mixed requests, multi-step, ensemble predict
§ StreamingBatchScheduler: Mixed CPU/GPU/IO-bound Workloads
§ Serve Only One (1) Model Inside One (1) TensorFlow Serving Process
§ Much Easier to Debug, Tune, Scale, and Manage Models in Production.

LET’S DEPLOY OPTIMIZED MODEL
10_Optimize_Model_Server
gpu.ml/notebooks

THANK YOU!! QUESTIONS?
§ https://github.com/PipelineAI/pipeline/
Contact Me
chris@pipeline.ai
@cfregly

Optimizing, Profiling, and Deploying TensorFlow AI Models with GPUs - San Francisco Python Meetup - Nov 8, 2017

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Viewers also liked

Viewers also liked (20)

Similar to Optimizing, Profiling, and Deploying TensorFlow AI Models with GPUs - San Francisco Python Meetup - Nov 8, 2017

Similar to Optimizing, Profiling, and Deploying TensorFlow AI Models with GPUs - San Francisco Python Meetup - Nov 8, 2017 (20)

More from Chris Fregly

More from Chris Fregly (15)

Recently uploaded

Recently uploaded (20)

Optimizing, Profiling, and Deploying TensorFlow AI Models with GPUs - San Francisco Python Meetup - Nov 8, 2017