@@ -15,11 +15,13 @@ Currently the following heuristics feature such integration:
15
15
16
16
This document is an outline of the tooling and APIs facilitating MLGO.
17
17
18
- Note that tools for orchestrating ML training are not part of LLVM, as they are
19
- dependency-heavy - both on the ML infrastructure choice, as well as choices of
20
- distributed computing. For the training scenario, LLVM only contains facilities
21
- enabling it, such as corpus extraction, training data extraction, and evaluation
22
- of models during training.
18
+ .. note ::
19
+
20
+ The tools for orchestrating ML training are not part of LLVM, as they are
21
+ dependency-heavy - both on the ML infrastructure choice, as well as choices of
22
+ distributed computing. For the training scenario, LLVM only contains facilities
23
+ enabling it, such as corpus extraction, training data extraction, and evaluation
24
+ of models during training.
23
25
24
26
25
27
.. contents ::
@@ -329,8 +331,10 @@ We currently feature 4 implementations:
329
331
the neural network, together with its weights (essentially, loops performing
330
332
matrix multiplications)
331
333
332
- NOTE: we are actively working on replacing this with an EmitC implementation
333
- requiring no out of tree build-time dependencies.
334
+ .. note ::
335
+
336
+ we are actively working on replacing this with an EmitC implementation
337
+ requiring no out of tree build-time dependencies.
334
338
335
339
- ``InteractiveModelRunner ``. This is intended for training scenarios where the
336
340
training algorithm drives compilation. This model runner has no special
@@ -526,9 +530,11 @@ implementation details.
526
530
Building with ML support
527
531
========================
528
532
529
- **NOTE ** For up to date information on custom builds, see the ``ml-* ``
530
- `build bots <http://lab.llvm.org >`_. They are set up using
531
- `like this <https://github.com/google/ml-compiler-opt/blob/main/buildbot/buildbot_init.sh >`_.
533
+ .. note ::
534
+
535
+ For up to date information on custom builds, see the ``ml-* ``
536
+ `build bots <http://lab.llvm.org >`_. They are set up using
537
+ `like this <https://github.com/google/ml-compiler-opt/blob/main/buildbot/buildbot_init.sh >`_.
532
538
533
539
Embed pre-trained models (aka "release" mode)
534
540
---------------------------------------------
@@ -562,9 +568,11 @@ You can also specify a URL for the path, and it is also possible to pre-compile
562
568
the header and object and then just point to the precompiled artifacts. See for
563
569
example ``LLVM_OVERRIDE_MODEL_HEADER_INLINERSIZEMODEL ``.
564
570
565
- **Note ** that we are transitioning away from the AOT compiler shipping with the
566
- tensorflow package, and to a EmitC, in-tree solution, so these details will
567
- change soon.
571
+ .. note ::
572
+
573
+ We are transitioning away from the AOT compiler shipping with the
574
+ tensorflow package, and to a EmitC, in-tree solution, so these details will
575
+ change soon.
568
576
569
577
Using TFLite (aka "development" mode)
570
578
-------------------------------------
0 commit comments