- 25 6月, 2020 40 次提交
-
-
由 Marcello Maggioni 提交于
Adding an HLO pass that expands the kLogistic HLO into a desired sequence of different HLOs. Currently two different strategies are added. 1) A lowering through an expansion using TAHN (0.5 + 0.5 * tanh(0.5 * x)) 2) A lowering through an expansion using EXP (1.0 / (1.0 + exp(-x))) PiperOrigin-RevId: 318208462 Change-Id: Ibcfba8e95f76c85cdbffc42566f5cec5e663c72b
-
由 Haoyu Zhang 提交于
PiperOrigin-RevId: 318205034 Change-Id: I9dfceae9d9c9896931c99c4f39e0227b8da17df5
-
由 A. Unique TensorFlower 提交于
class. PiperOrigin-RevId: 318203984 Change-Id: I85e2aa5d4c498b67f3e9130fa45a38173f84b35b
-
由 Marcello Maggioni 提交于
It should have a cost similar to kSqrt. PiperOrigin-RevId: 318203675 Change-Id: Ic55aba543bd1760ca0d89c62401524f4ca4ac97b
-
由 Saurabh Saxena 提交于
Enabled unified API test to run with TFRT. Replace dyn_cast with tensorflow::dyn_cast. TF_ExecuteOperation no longer takes a TF_ExecutionContext arg since the current impls tie the op builder to the creating context and it is not clear if we will ever support that API. PiperOrigin-RevId: 318203001 Change-Id: If5048b1404b87c809606c236419e1869630bcd46
-
由 Brian Zhao 提交于
PiperOrigin-RevId: 318199820 Change-Id: I901124780f8687d0f572cff4546f8792f1120e47
-
由 Michael Gester 提交于
Also added required ops and MLIR unit tests. PiperOrigin-RevId: 318199300 Change-Id: I46bf921b5a14c1c4428bfdf51d3e3415a3af65bc
-
由 Saurabh Saxena 提交于
PiperOrigin-RevId: 318190892 Change-Id: I9770cc8689f1ad20c25b04ca2d5503bab9f0b33c
-
由 George Karpenkov 提交于
PiperOrigin-RevId: 318190091 Change-Id: I740286eb8a71ebcf493fa13f9f7ced45be24048b
-
由 Scott Zhu 提交于
PiperOrigin-RevId: 318189813 Change-Id: I7659dd47ddd91e698fc23edd5a527e05a382d513
-
由 Bruce Fontaine 提交于
Allow tf.distribute.TPUStrategy to be used with TPUEmbedding API and ensure that LossScaleOptimizer properly rejects it. PiperOrigin-RevId: 318186211 Change-Id: Id3b9cb8288e5d28ddbaec97d5b35627ab35bc08d
-
由 Henry Tan 提交于
PiperOrigin-RevId: 318186086 Change-Id: I2cf472a1d1b17cc27d07c7e085073620cc2c699e
-
由 Chenkai Kuang 提交于
PiperOrigin-RevId: 318185207 Change-Id: Iaafc17d4396adf918f5e6ed25009f86213f9591d
-
由 Jacques Pienaar 提交于
This should make it easier to split the generated ops files. On its own this doesn't help the compilation, it just moves the ops to their own lib where we could split those up again. PiperOrigin-RevId: 318184606 Change-Id: Ie0f26cb7d3e4c3da70c66e845312896afed81a90
-
https://github.com/llvm/llvm-project/commit/1c0bbe4341ac由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 318182263 Change-Id: I7643bd239e3e64f789a73632edf0bc3dde8d4203
-
由 Henry Tan 提交于
PiperOrigin-RevId: 318177458 Change-Id: I849fb7214fac4eebbfec3a929f030a3de3cbc9a5
-
由 Robert David 提交于
Remove redundant ternary operator: scratch3 is already nullptr if asymmetric_quantize_inputs is false. PiperOrigin-RevId: 318176946 Change-Id: Ia695d44cc11ea89460efefc0c9cebfafb6eb3b26
-
由 Pavithra Vijay 提交于
PiperOrigin-RevId: 318176105 Change-Id: I85a13341bf659d1930dc55762f8c8a11020678c2
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 318175201 Change-Id: I691d559c86d00fabb8f0c028c1e815f804f7d350
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 318173722 Change-Id: I3649bfa6804fd81252292470516872382612c23d
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 318171874 Change-Id: Ib1c1693f93e580e51515640295531edcb494e3b0
-
Imported from GitHub PR https://github.com/tensorflow/tensorflow/pull/39691 Integrated Cadence HiFi Mini optimized functions for fully_connected, softmax and svdf kernels. These optimized functions gives better performance for keyword_benchmark application. Copybara import of the project: -- 094b959873d4db062b65848c97fcff1990dcbe8f by bhanu prakash bandaru venkata <bhanup@cadence.com>: Cadence HiFi Mini NN Library: Optimized kernels integration Integrated fully_connected, softmax and svdf cadence HiFi Mini optimized kernels to tensorflow lite micro kernels. These optimized kernels gives better performance over reference kernels while running keyword_benchmark application. -- 172dc6c9ecc8289a984e786d0a254e397c984745 by bhanu prakash bandaru venkata <bhanup@cadence.com>: Cadence HiFi Mini NN Library: code cleanup and optimization Further optimized states update in svdf function. Code cleanup: keep the files which are used in keyword_benchmark application (keep the functions called in fully_connected, softmax and svdf). delete remaining files in NN library -- 898d39d526fd38aec04a217b2bc627297c95805d by bhanu prakash bandaru venkata <bhanup@cadence.com>: Cadence HiFi Mini Nn Library: updating copy rights Copyrights are updated in fully_connected, softmax and svdf files. indentation related change in dot_prod. COPYBARA_INTEGRATE_REVIEW=https://github.com/tensorflow/tensorflow/pull/39691 from bhanuprakashbv:master 898d39d526fd38aec04a217b2bc627297c95805d PiperOrigin-RevId: 318168368 Change-Id: I61f7812b3a9bcc3a327ed4047c5d72fb64b012e5
-
由 TensorFlower Gardener 提交于
PiperOrigin-RevId: 318167643 Change-Id: Id417d71f454eb1eadc3f41213c4887494ea810e7
-
由 Smit Hinsu 提交于
Currently test has an undefined behavior: SanitizerError UndefinedBehaviorSanitizer: null-pointer-use include/c++/v1/vector:1551:12 PiperOrigin-RevId: 318166647 Change-Id: Ie0c3cac1c2b13738ed9cb9942a0b567059a98876
-
由 Haoyu Zhang 提交于
Currently `UpdateRemoteMaster` first get a list of registered functions, and then register them to remote workers. If functions are removed at the same time, it might hit nullptr when looking up functions by names in the previously acquired list. In this change, the function_defs come from the ToProto(). It makes a copy of the function protos to avoid the race condition. PiperOrigin-RevId: 318166239 Change-Id: Ia13091ccc9577691c6508f8de7f7d2766ef81353
-
https://github.com/llvm/llvm-project/commit/cdd6a2788cac由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 318163539 Change-Id: I9a9fcff5cb03a49b46e474c050b78e0a51bf6657
-
由 Scott Zhu 提交于
PiperOrigin-RevId: 318163436 Change-Id: Ia0d60bdf7d47de0efe32dfa84778168db4cee0b5
-
由 A. Unique TensorFlower 提交于
1. Remove identity Reshape ops with control inputs. 2. Remove identity BroadcastTo ops. This op is common, as it appears in the gradient of Sum. PiperOrigin-RevId: 318163412 Change-Id: I63954d71c4ed4ff8636bdad651ac7f18d0518fc0
-
由 Pete Warden 提交于
PiperOrigin-RevId: 318163038 Change-Id: I3e509e2f9eb6a57760470934c8b08c567b0da47e
-
由 Nat Jeffries 提交于
PiperOrigin-RevId: 318160671 Change-Id: I70612a26d92b2f840f54fc808dddbf2279ab5c98
-
由 Bruce Fontaine 提交于
PiperOrigin-RevId: 318160598 Change-Id: I99ec2aa41a6e4e5fb4720b878fd82ad10802dcc9
-
由 Marcello Maggioni 提交于
This will enable target specific lowering for the logistic function to be performed when this HLO is used. PiperOrigin-RevId: 318159527 Change-Id: I453782fea99838fddd9039f63faa5c876cb7dec0
-
由 Robert David 提交于
PiperOrigin-RevId: 318157584 Change-Id: I9bbee585b18fc508e6afb166e394e8d4bf50940d
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 318153404 Change-Id: Iffce4d65c1b8b4c12c24a61094ab1b13fb6b9ee2
-
由 Vo Van Nghia 提交于
-
由 Christian Sigg 提交于
Two function declarations were added by accident. Macro was missing closing parenthesis. PiperOrigin-RevId: 318149363 Change-Id: I6363ec5f75719a594f0ab19732dd06704ead02c5
-
由 Vo Van Nghia 提交于
-
由 Francois Chollet 提交于
layers with a default of `training=True` called in e.g. a Sequential container. PiperOrigin-RevId: 318145694 Change-Id: I1af5286824e3a45e1a7d1b8a4fadd7ec223895dc
-
由 Saurabh Saxena 提交于
PiperOrigin-RevId: 318144766 Change-Id: Ie2bc05db0c8b022b085c6b60ad4bd54838407abe
-
由 Jaesung Chung 提交于
Since tf_saved_model passes now requires CPU op kernels to capture resource variables, it would be better to be split from the tensorflow passes to keep use cases, that are not required tf_saved_model dialect, small. PiperOrigin-RevId: 318143061 Change-Id: I0b30efd1dcb8945cdfe6184664f66cd757e0ea67
-