cdocutils.nodes document q)q}q(U nametypesq}qXmachine learningqNsUsubstitution_defsq}qUparse_messagesq ]q Ucurrent_sourceq NU decorationq NUautofootnote_startq KUnameidsq}qhUmachine-learningqsUchildrenq]qcdocutils.nodes section q)q}q(U rawsourceqUUparentqhUsourceqXh/home/gtbldadm/nightlybuilds/processor-sdk-doc/source/linux/Foundational_Components_Machine_Learning.rstqUtagnameqUsectionqU attributesq}q(Udupnamesq]Uclassesq]Ubackrefsq ]Uidsq!]q"haUnamesq#]q$hauUlineq%KUdocumentq&hh]q'(cdocutils.nodes title q()q)}q*(hXMachine Learningq+hhhhhUtitleq,h}q-(h]h]h ]h!]h#]uh%Kh&hh]q.cdocutils.nodes Text q/XMachine Learningq0q1}q2(hh+hh)ubaubcdocutils.nodes paragraph q3)q4}q5(hXSitara Machine Learning toolkit brings machine learning to the edge by enabling machine learning inference on all Sitara devices (Arm only, Arm + specialized hardware accelerators). It is provided as part of TI's Processor SDK Linux, free to download and use. Sitara machine learning today consists of TI Deep Learning (TIDL), Neo-AI-DLR, TVM runtime, TensorFlow Lite, Arm NN, and RNN library.q6hhhhhU paragraphq7h}q8(h]h]h ]h!]h#]uh%Kh&hh]q9h/XSitara Machine Learning toolkit brings machine learning to the edge by enabling machine learning inference on all Sitara devices (Arm only, Arm + specialized hardware accelerators). It is provided as part of TI's Processor SDK Linux, free to download and use. Sitara machine learning today consists of TI Deep Learning (TIDL), Neo-AI-DLR, TVM runtime, TensorFlow Lite, Arm NN, and RNN library.q:q;}q<(hh6hh4ubaubcdocutils.nodes figure q=)q>}q?(hUhhhhhUfigureq@h}qA(UalignqBXcenterh!]qCUid1qDah ]h]h]h#]uh%Nh&hh]qE(cdocutils.nodes image qF)qG}qH(hX.. figure:: ../images/Sitara_machine_learning_stack_diagram.png :width: 1120px :align: center :height: 576px Sitara Machine Learnining Offering h}qI(h]UuriX9linux/../images/Sitara_machine_learning_stack_diagram.pngqJh!]h ]UwidthX1120pxh]U candidatesqK}qLU*hJsh#]UheightX576pxuhh>h]hUimageqMubcdocutils.nodes caption qN)qO}qP(hX"Sitara Machine Learnining OfferingqQhh>hhhUcaptionqRh}qS(h]h]h ]h!]h#]uh%Kh]qTh/X"Sitara Machine Learnining OfferingqUqV}qW(hhQhhOubaubeubcsphinx.ext.ifconfig ifconfig qX)qY}qZ(hUhhhhhUifconfigq[h}q\(Uexprq]X(CONFIG_part_family in ('General_family')h!]h ]h]h]h#]uh%Kh&hh]q^(cdocutils.nodes rubric q_)q`}qa(hXP`TI Deep Learning (TIDL) `__qbhhYhhhUrubricqch}qd(h]h]h ]h!]h#]uh%Kh&hh]qecdocutils.nodes reference qf)qg}qh(hhbh}qi(UnameXTI Deep Learning (TIDL)UrefuriqjX2Foundational_Components/Machine_Learning/tidl.htmlh!]h ]h]h]h#]uhh`h]qkh/XTI Deep Learning (TIDL)qlqm}qn(hUhhgubahU referenceqoubaubcdocutils.nodes bullet_list qp)qq}qr(hUhhYhhhU bullet_listqsh}qt(UbulletquX*h!]h ]h]h]h#]uh%Kh&hh]qv(cdocutils.nodes list_item qw)qx}qy(hXhAccelerates deep learning inference on C66x DSP cores and/or on Embedded Vision Engine (EVE) subsystems.qzhhqhhhU list_itemq{h}q|(h]h]h ]h!]h#]uh%Nh&hh]q}h3)q~}q(hhzhhxhhhh7h}q(h]h]h ]h!]h#]uh%Kh]qh/XhAccelerates deep learning inference on C66x DSP cores and/or on Embedded Vision Engine (EVE) subsystems.qq}q(hhzhh~ubaubaubhw)q}q(hXAvailable on AM57x device only.qhhqhhhh{h}q(h]h]h ]h!]h#]uh%Nh&hh]qh3)q}q(hhhhhhhh7h}q(h]h]h ]h!]h#]uh%Kh]qh/XAvailable on AM57x device only.qq}q(hhhhubaubaubhw)q}q(hXKSupports CNN at the moment, and imports Caffe, ONNX, and Tensorflow models.qhhqhhhh{h}q(h]h]h ]h!]h#]uh%Nh&hh]qh3)q}q(hhhhhhhh7h}q(h]h]h ]h!]h#]uh%Kh]qh/XKSupports CNN at the moment, and imports Caffe, ONNX, and Tensorflow models.qq}q(hhhhubaubaubeubeubh_)q}q(hXZ`Neo-AI Deep Learning Runtime (DLR) `__qhhhhhhch}q(h]h]h ]h!]h#]uh%Kh&hh]qhf)q}q(hhh}q(UnameX"Neo-AI Deep Learning Runtime (DLR)hjX1Foundational_Components/Machine_Learning/neo.htmlh!]h ]h]h]h#]uhhh]qh/X"Neo-AI Deep Learning Runtime (DLR)qq}q(hUhhubahhoubaubhp)q}q(hUhhhhhhsh}q(huX*h!]h ]h]h]h#]uh%Kh&hh]q(hw)q}q(hXQNeo-AI-DLR is a new open source machine learning runtime for on-device inference.qhhhhhh{h}q(h]h]h ]h!]h#]uh%Nh&hh]qh3)q}q(hhhhhhhh7h}q(h]h]h ]h!]h#]uh%Kh]qh/XQNeo-AI-DLR is a new open source machine learning runtime for on-device inference.qq}q(hhhhubaubaubhw)q}q(hXSupports Keras, Tensorflow, TFLite, GluonCV, MXNet, Pytorch, ONNX, and XGBoost models optimized automatically by `Amazon SageMaker Neo `__ or `TVM compiler `__.hhhhhh{h}q(h]h]h ]h!]h#]uh%Nh&hh]qh3)q}q(hXSupports Keras, Tensorflow, TFLite, GluonCV, MXNet, Pytorch, ONNX, and XGBoost models optimized automatically by `Amazon SageMaker Neo `__ or `TVM compiler `__.hhhhhh7h}q(h]h]h ]h!]h#]uh%Kh]q(h/XqSupports Keras, Tensorflow, TFLite, GluonCV, MXNet, Pytorch, ONNX, and XGBoost models optimized automatically by qq}q(hXqSupports Keras, Tensorflow, TFLite, GluonCV, MXNet, Pytorch, ONNX, and XGBoost models optimized automatically by hhubhf)q}q(hX@`Amazon SageMaker Neo `__h}q(UnameXAmazon SageMaker NeohjX%https://aws.amazon.com/sagemaker/neo/h!]h ]h]h]h#]uhhh]qh/XAmazon SageMaker Neoqȅq}q(hUhhubahhoubh/X or q˅q}q(hX or hhubhf)q}q(hX;`TVM compiler `__h}q(UnameX TVM compilerhjX(https://github.com/apache/incubator-tvm/h!]h ]h]h]h#]uhhh]qh/X TVM compilerq҅q}q(hUhhubahhoubh/X.q}q(hX.hhubeubaubhw)q}q(hXHSupports all Cortex-A ARM cores (AM3x, AM4x, AM5x, AM6x Sitara devices).qhhhhhh{h}q(h]h]h ]h!]h#]uh%Nh&hh]qh3)q}q(hhhhhhhh7h}q(h]h]h ]h!]h#]uh%Kh]qh/XHSupports all Cortex-A ARM cores (AM3x, AM4x, AM5x, AM6x Sitara devices).qq}q(hhhhubaubaubhw)q}q(hXVOn AM5729 and AM5749 devices, uses TIDL to accelerate supported models automatically. hhhhhh{h}q(h]h]h ]h!]h#]uh%Nh&hh]qh3)q}q(hXUOn AM5729 and AM5749 devices, uses TIDL to accelerate supported models automatically.qhhhhhh7h}q(h]h]h ]h!]h#]uh%K h]qh/XUOn AM5729 and AM5749 devices, uses TIDL to accelerate supported models automatically.q셁q}q(hhhhubaubaubeubh_)q}q(hXC`TVM Runtime `__qhhhhhhch}q(h]h]h ]h!]h#]uh%K#h&hh]qhf)q}q(hhh}q(UnameX TVM RuntimehjX1Foundational_Components/Machine_Learning/tvm.htmlh!]h ]h]h]h#]uhhh]qh/X TVM Runtimeqq}q(hUhhubahhoubaubhp)q}q(hUhhhhhhsh}q(huX*h!]h ]h]h]h#]uh%K$h&hh]q(hw)q}r(hXOpen source deep learning runtime for on-device inference, supporting models compiled by `TVM compiler `__.hhhhhh{h}r(h]h]h ]h!]h#]uh%Nh&hh]rh3)r}r(hXOpen source deep learning runtime for on-device inference, supporting models compiled by `TVM compiler `__.hhhhhh7h}r(h]h]h ]h!]h#]uh%K$h]r(h/XYOpen source deep learning runtime for on-device inference, supporting models compiled by rr}r (hXYOpen source deep learning runtime for on-device inference, supporting models compiled by hjubhf)r }r (hX;`TVM compiler `__h}r (UnameX TVM compilerhjX(https://github.com/apache/incubator-tvm/h!]h ]h]h]h#]uhjh]r h/X TVM compilerrr}r(hUhj ubahhoubh/X.r}r(hX.hjubeubaubhw)r}r(hXMAvailable on all Cortex-A ARM cores (AM3x, AM4x, AM5x, AM6x Sitara devices). hhhhhh{h}r(h]h]h ]h!]h#]uh%Nh&hh]rh3)r}r(hXLAvailable on all Cortex-A ARM cores (AM3x, AM4x, AM5x, AM6x Sitara devices).rhjhhhh7h}r(h]h]h ]h!]h#]uh%K&h]rh/XLAvailable on all Cortex-A ARM cores (AM3x, AM4x, AM5x, AM6x Sitara devices).rr}r(hjhjubaubaubeubh_)r}r (hXJ`TensorFlow Lite `__r!hhhhhhch}r"(h]h]h ]h!]h#]uh%K)h&hh]r#hf)r$}r%(hj!h}r&(UnameXTensorFlow LitehjX4Foundational_Components/Machine_Learning/tflite.htmlh!]h ]h]h]h#]uhjh]r'h/XTensorFlow Liter(r)}r*(hUhj$ubahhoubaubhp)r+}r,(hUhhhhhhsh}r-(huX*h!]h ]h]h]h#]uh%K*h&hh]r.(hw)r/}r0(hX:Open source deep learning runtime for on-device inference.r1hj+hhhh{h}r2(h]h]h ]h!]h#]uh%Nh&hh]r3h3)r4}r5(hj1hj/hhhh7h}r6(h]h]h ]h!]h#]uh%K*h]r7h/X:Open source deep learning runtime for on-device inference.r8r9}r:(hj1hj4ubaubaubhw)r;}r<(hXGRuns on all Cortex-A ARM cores (AM3x, AM4x, AM5x, AM6x Sitara devices).r=hj+hhhh{h}r>(h]h]h ]h!]h#]uh%Nh&hh]r?h3)r@}rA(hj=hj;hhhh7h}rB(h]h]h ]h!]h#]uh%K+h]rCh/XGRuns on all Cortex-A ARM cores (AM3x, AM4x, AM5x, AM6x Sitara devices).rDrE}rF(hj=hj@ubaubaubhw)rG}rH(hXImports Tensorflow Lite models.rIhj+hhhh{h}rJ(h]h]h ]h!]h#]uh%Nh&hh]rKh3)rL}rM(hjIhjGhhhh7h}rN(h]h]h ]h!]h#]uh%K,h]rOh/XImports Tensorflow Lite models.rPrQ}rR(hjIhjLubaubaubhw)rS}rT(hXUses TIDL import tool to create TIDL offloadable Tensorflow Lite models, which can be executed via Tensorflow Lite runtime with TIDL acceleration on AM5729 and AM5749 devices. hj+hhhh{h}rU(h]h]h ]h!]h#]uh%Nh&hh]rVh3)rW}rX(hXUses TIDL import tool to create TIDL offloadable Tensorflow Lite models, which can be executed via Tensorflow Lite runtime with TIDL acceleration on AM5729 and AM5749 devices.rYhjShhhh7h}rZ(h]h]h ]h!]h#]uh%K-h]r[h/XUses TIDL import tool to create TIDL offloadable Tensorflow Lite models, which can be executed via Tensorflow Lite runtime with TIDL acceleration on AM5729 and AM5749 devices.r\r]}r^(hjYhjWubaubaubeubh_)r_}r`(hX@`Arm NN `__rahhhhhhch}rb(h]h]h ]h!]h#]uh%K1h&hh]rchf)rd}re(hjah}rf(UnameXArm NNhjX3Foundational_Components/Machine_Learning/armnn.htmlh!]h ]h]h]h#]uhj_h]rgh/XArm NNrhri}rj(hUhjdubahhoubaubhp)rk}rl(hUhhhhhhsh}rm(huX*h!]h ]h]h]h#]uh%K2h&hh]rn(hw)ro}rp(hX0Open source inference engine available from Arm.rqhjkhhhh{h}rr(h]h]h ]h!]h#]uh%Nh&hh]rsh3)rt}ru(hjqhjohhhh7h}rv(h]h]h ]h!]h#]uh%K2h]rwh/X0Open source inference engine available from Arm.rxry}rz(hjqhjtubaubaubhw)r{}r|(hXGRuns on all Cortex-A ARM cores (AM3x, AM4x, AM5x, AM6x Sitara devices).r}hjkhhhh{h}r~(h]h]h ]h!]h#]uh%Nh&hh]rh3)r}r(hj}hj{hhhh7h}r(h]h]h ]h!]h#]uh%K3h]rh/XGRuns on all Cortex-A ARM cores (AM3x, AM4x, AM5x, AM6x Sitara devices).rr}r(hj}hjubaubaubhw)r}r(hX=Imports Caffe, ONNX, TensorFlow, and TensorFlow Lite models. hjkhhhh{h}r(h]h]h ]h!]h#]uh%Nh&hh]rh3)r}r(hX<Imports Caffe, ONNX, TensorFlow, and TensorFlow Lite models.rhjhhhh7h}r(h]h]h ]h!]h#]uh%K4h]rh/X<Imports Caffe, ONNX, TensorFlow, and TensorFlow Lite models.rr}r(hjhjubaubaubeubh_)r}r(hXR`RNN Library `__rhhhhhhch}r(h]h]h ]h!]h#]uh%K7h&hh]rhf)r}r(hjh}r(UnameX RNN LibraryhjX@Examples_and_Demos_Application_Demos.html#deploy-and-integrationh!]h ]h]h]h#]uhjh]rh/X RNN Libraryrr}r(hUhjubahhoubaubhp)r}r(hUhhhhhhsh}r(huX*h!]h ]h]h]h#]uh%K8h&hh]r(hw)r}r(hXProvides Long Short-Term Memory (LSTM) and fully connected layers in a standalone library to allow for rapid prototyping of inference applications that require Recurrent Neural Networks.rhjhhhh{h}r(h]h]h ]h!]h#]uh%Nh&hh]rh3)r}r(hjhjhhhh7h}r(h]h]h ]h!]h#]uh%K8h]rh/XProvides Long Short-Term Memory (LSTM) and fully connected layers in a standalone library to allow for rapid prototyping of inference applications that require Recurrent Neural Networks.rr}r(hjhjubaubaubhw)r}r(hXGRuns on all Cortex-A ARM cores (AM3x, AM4x, AM5x, AM6x Sitara devices).rhjhhhh{h}r(h]h]h ]h!]h#]uh%Nh&hh]rh3)r}r(hjhjhhhh7h}r(h]h]h ]h!]h#]uh%K9h]rh/XGRuns on all Cortex-A ARM cores (AM3x, AM4x, AM5x, AM6x Sitara devices).rr}r(hjhjubaubaubhw)r}r(hXIntegrated into TI's Processor SDK Linux in an OOB demo for `Predictive Maintenance `__. hjhhhh{h}r(h]h]h ]h!]h#]uh%Nh&hh]rh3)r}r(hXIntegrated into TI's Processor SDK Linux in an OOB demo for `Predictive Maintenance `__.rhjhhhh7h}r(h]h]h ]h!]h#]uh%K:h]r(h/X<Integrated into TI's Processor SDK Linux in an OOB demo for rr}r(hX<Integrated into TI's Processor SDK Linux in an OOB demo for hjubhf)r}r(hXb`Predictive Maintenance `__h}r(UnameXPredictive MaintenancehjXEExamples_and_Demos_Application_Demos.html#predictive-maintenance-demoh!]h ]h]h]h#]uhjh]rh/XPredictive Maintenancerr}r(hUhjubahhoubh/X.r}r(hX.hjubeubaubeubcdocutils.nodes compound r)r}r(hUhhhhhUcompoundrh}r(h]h]rUtoctree-wrapperrah ]h!]h#]uh%Nh&hh]rcsphinx.addnodes toctree r)r}r(hUhjhhhUtoctreerh}r(UnumberedrKU includehiddenrhX.linux/Foundational_Components_Machine_LearningrhRNUglobrh!]h ]U titlesonlyrh]h]h#]Uentriesr]r(NX3linux/Foundational_Components/Machine_Learning/tidlrrNX2linux/Foundational_Components/Machine_Learning/neorrNX2linux/Foundational_Components/Machine_Learning/tvmrrNX5linux/Foundational_Components/Machine_Learning/tfliterrNX4linux/Foundational_Components/Machine_Learning/armnnrreUhiddenrU includefilesr]r(jjjjjeUmaxdepthrJuh%KU_sourcer?hUgettext_compactr@U generatorrANUdump_internalsrBNU smart_quotesrCU pep_base_urlrDU https://www.python.org/dev/peps/rEUsyntax_highlightrFUlongrGUinput_encoding_error_handlerrHj#Uauto_id_prefixrIUidrJUdoctitle_xformrKUstrip_elements_with_classesrLNU _config_filesrM]Ufile_insertion_enabledrNU raw_enabledrOKU dump_settingsrPNubUsymbol_footnote_startrQKUidsrR}rS(hhhDh>uUsubstitution_namesrT}rUhh&h}rV(h]h!]h ]Usourcehh]h#]uU footnotesrW]rXUrefidsrY}rZub.