[DOCS] Introduction to Relay IR. by tqchen · Pull Request #2185 · apache/tvm

tqchen · 2018-11-28T04:27:05Z

This is an introduction material to Relay IR for developers who have a background on data-flow and computational graphs. This tutorial is a result of discussion with @jroesch @yzhliu @junrushao1994 @MarisaKirisame @slyubomirsky @joshpoll and other folks. We try to blend the views from deep learning frameworks and PL

tqchen · 2018-11-28T04:29:16Z

cc @zheng-da @zhiics

junrushao · 2018-11-28T04:31:06Z

cc @szha

masahi · 2018-11-28T09:19:00Z

+framework developer choose the representation they are familiar with.
+This does, however, have some implications on how we write passes:
+
+- If you come from a data-flow background and want to handle let, keep a map of var to the expressions so you can perform lookup when encountering a var. This is a likely means a minimum change as we already need a map from expr-> transformed expression anyway. Note that this will effectively remove all the let in the program.


This is a likely means -> This likely means

zhiics · 2018-11-28T06:26:48Z

+The Module can be viewed as a ``Map<GlobalVar, Function>``. Here GlobalVar is just an id that is used to represent the functions
+in the module. ``@muladd`` and ``@myfunc`` are GlobalVars in the above example. When a CallNode is used to call another function,
+the corresponding GlobalVar is stored in the op field of the CallNode. It contains a level of indirection -- we need to look up
+body of the called function from the modele using the corresponding GlobalVar. In this particular case, we could also directly


s/modele/module

zhiics · 2018-11-28T06:36:32Z

+
+Different data structures will impact how you might write transformations, and we need to keep that in mind.
+So now, as a deep learning framework developer, you might ask, why do we need let-binding.
+Yours PL friends will always tell you that let is important -- as PL is a quite established field,


s/Yours/Your

zhiics · 2018-11-28T18:13:23Z

+This article introduces Relay IR -- the second generation of NNVM.
+We expect readers from two kinds of background -- those who have a programming language background and deep learning
+framework developers who are familiar with the computational graph representation.
+This article is mainly written for deep learning framework developers who are familiar with the computational graph representation.


Is this a little repetitive?

zhiics · 2018-11-28T18:13:54Z

+Build Computational Graph with Relay
+------------------------------------
+Traditional deep learning frameworks use computational graphs as their intermediate representation.
+A computational graph (or data-flow graph), is a directed acyclic graph (DAG) that represent the computation.


zhiics · 2018-11-28T18:31:59Z

+construct a simple two-node graph. You can find that the syntax of the example is not that different from existing
+computational graph IR like NNVMv1, with the only difference in terms of terminology:
+
+- Existing frameworks usually uses graph and subgraph


zhiics · 2018-11-28T18:38:30Z

+
+Each data-flow node is a CallNode in Relay. The relay python DSL allows you to construct a data-flow quickly.
+One thing we want to highlight in the above code -- is that we explicitly constructed an Add node with
+both input points to ``%1``.  When a deep learning framework evaluates the above program, it will compute


both inputs point to ?

zhiics · 2018-11-28T18:49:20Z

+  }
+
+Let binding solves this problem, as the computation of the value happens at the let node. In both programs,
+if we change ``%1 = log(%x)`` to ``let %v1 = log(%x)``, we clearly specifies the computation location to


s/specifies/specify

zhiics · 2018-11-28T18:51:22Z

+-- we don’t need to worry about where to put the let when we generate the code. The dataflow form also gives more freedom
+to the later passes to decide where to put the evaluation point. As a result, it might not be a bad idea to use data flow
+form of the program in the initial phases of optimizations when you find it is convenient.
+As a matter of fact, many optimizations in relay today are written to optimize dataflow programs.


s/relay/Relay ?

tqchen · 2018-11-29T05:14:54Z

Thanks @masahi @zhiics for comments, I have updated accordingly

zhiics · 2018-11-29T05:21:30Z

Cool. I think the tutorial is really helpful for people to understand Relay.

zhiics · 2018-11-29T05:46:36Z

+Since program optimizations take these AST data structures and transform them, the two different structure will
+affect the compiler code we are going to write. For example, if we want to detect a pattern ``add(log(x), y)``:
+
+- In the data-flow form, we can first access the add node, then directly look at its first arguments to see if it is a log


s/arguments/argument

zhiics · 2018-11-29T05:51:52Z

+One thing we want to highlight in the above code -- is that we explicitly constructed an Add node with
+both input point to ``%1``.  When a deep learning framework evaluates the above program, it will compute
+the nodes in topological order, and ``%1`` will only be computed once.
+While the this fact is very natural to deep learning framework builders, it is something that might


While this fact

zhiics · 2018-11-29T05:54:56Z

I found two more typos after another reading.

tqchen · 2018-11-29T18:02:24Z

Thanks @masahi @zhiics this is merged

tqchen force-pushed the master branch 2 times, most recently from 944c947 to 7c64e66 Compare November 28, 2018 04:55

[DOCS] Introduction to Relay IR.

80a615c

tqchen force-pushed the master branch from 7c64e66 to 80a615c Compare November 28, 2018 05:01

masahi reviewed Nov 28, 2018

View reviewed changes

zhiics reviewed Nov 28, 2018

View reviewed changes

Update per review comments

cba72f5

zhiics approved these changes Nov 29, 2018

View reviewed changes

masahi approved these changes Nov 29, 2018

View reviewed changes

tqchen added 2 commits November 28, 2018 21:38

minor fixes

0ee317e

minor fix

a87b156

zhiics reviewed Nov 29, 2018

View reviewed changes

fix thanks to zhiics

e0467a5

tqchen merged commit f522b0f into apache:master Nov 29, 2018

jroesch mentioned this pull request Dec 1, 2018

[RELAY][DOCS] Relay Documentation Sprint #2212

Closed

16 tasks

FrozenGene pushed a commit to FrozenGene/tvm that referenced this pull request Dec 27, 2018

[DOCS] Introduction to Relay IR. (apache#2185)

5cf3265

wweic pushed a commit to neo-ai/tvm that referenced this pull request Feb 20, 2019

[DOCS] Introduction to Relay IR. (apache#2185)

417d072

wweic pushed a commit to neo-ai/tvm that referenced this pull request Feb 20, 2019

[DOCS] Introduction to Relay IR. (apache#2185)

88e2b1d

Conversation

tqchen commented Nov 28, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

tqchen commented Nov 28, 2018

Uh oh!

junrushao commented Nov 28, 2018

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

zhiics Nov 28, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

tqchen commented Nov 29, 2018

Uh oh!

zhiics commented Nov 29, 2018

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

zhiics commented Nov 29, 2018

Uh oh!

tqchen commented Nov 29, 2018

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

tqchen commented Nov 28, 2018 •

edited

Loading

zhiics Nov 28, 2018 •

edited

Loading