- 31 10月, 2020 22 次提交
-
-
由 Raman Sarokin 提交于
PiperOrigin-RevId: 339913277 Change-Id: If4a92b3fafe922b5abf00bc8dc04deef1ef6ca6d
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 339911119 Change-Id: Iddf765ff3e58ea67c0cdf3b1ffe3065a92e8dcfc
-
由 Advait Jain 提交于
This change avoids an implicit conversion from enum BuiltinOperator to int8_t for calls to CreateOperatorCodeDirect. This should mean that once the BuiltinOperator enum is larger than a byte, the TFLM code does not get tripped. PiperOrigin-RevId: 339910848 Change-Id: Ief978f2bb9ea818f8211ade3927baa35dc7a8512
-
由 Katherine Wu 提交于
Add property that allows layers to specify that the input_spec can also be used as the layer call function's input_signature. By default,all Keras exported layers will have this property return True, since Keras more rigidly defines the input_spec shape (compared to user-defined models, which may only have `ndims` set). PiperOrigin-RevId: 339909813 Change-Id: I1b486747aa1e413e6f24f6809fb88846ad4712ab
-
由 Scott Zhu 提交于
PiperOrigin-RevId: 339909803 Change-Id: I7cbcc1d74c0de2cd7f86e10e2b44f0d4b81274a0
-
由 Jose Baiocchi 提交于
PiperOrigin-RevId: 339909494 Change-Id: I1a9352342e9b79b55b5ba1f324fecf16a34b7647
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 339906858 Change-Id: I0ad0e7fef95d634a75ddb22a5b08dee9b66c2ec1
-
由 Jiri Simsa 提交于
PiperOrigin-RevId: 339906797 Change-Id: Ieff06a2d26a3b7bfc90987bba8457f1ff866f8af
-
由 TensorFlower Gardener 提交于
PiperOrigin-RevId: 339899776 Change-Id: Iff5a37227f189607727f3e24d75b41ee8b906af9
-
由 TensorFlower Gardener 提交于
PiperOrigin-RevId: 339899425 Change-Id: If9bbb9c68787dc484129102c716da2203c209dab
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 339893950 Change-Id: I6d3bf00d3296f5008ce2bd9f208dc0600a515665
-
由 Jian Li 提交于
PiperOrigin-RevId: 339893949 Change-Id: I43c4d0c58a19efb52b66a2ccbf9f792f05e0c73a
-
由 Thomas O'Malley 提交于
properties to help distinguish between metrics that should and shouldn't be passed sample_weight argument. Note these properties are set to None before Model.fit is called, since metrics are potentially broadcast to match the structure of data seen in Model.fit. PiperOrigin-RevId: 339892649 Change-Id: I0abffae08efde2b8adc58014ef205d318d66a9ab
-
由 Tomer Kaftan 提交于
PiperOrigin-RevId: 339892095 Change-Id: Ic6c9cebe68aad3d2b388bcdfa3e7249628dcf8ac
-
由 Tres Popp 提交于
PiperOrigin-RevId: 339890070 Change-Id: I8144220b5ffc20132e1bb2fd0b099266d2a0391c
-
由 Stephan Herhut 提交于
This adds back functionality to the unranked case that was already present (albeit manual) in the ranked case. PiperOrigin-RevId: 339889861 Change-Id: I6426ac5553c75bebd829135104d1b7802ff2a254
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 339888534 Change-Id: I60fb13fad42ec5938d7276151afdce2dc3a6926f
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 339887960 Change-Id: I536c60018b983c115c27d664b57fb268ea90ee98
-
由 Kuangyuan Chen 提交于
PiperOrigin-RevId: 339887058 Change-Id: I2decd73c27e6c3dc1a79cc723075598025827852
-
由 Marat Dukhan 提交于
Pull a work-around for compilation of x86 versions with legacy compilers PiperOrigin-RevId: 339886664 Change-Id: I5035f38cff401564061f092f59d5dea91313f5bf
-
由 Stephan Herhut 提交于
This will replace the existing, hardwired pass that was used for ranked kernels. PiperOrigin-RevId: 339886355 Change-Id: I362f04a806a0079733caa91edff73d2547fca221
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 339883324 Change-Id: I9b8cbe92bfa2bd279be4d72ca9a40b894f7fcaa7
-
- 30 10月, 2020 18 次提交
-
-
由 Ken Franko 提交于
PiperOrigin-RevId: 339877462 Change-Id: Ic52ed793da9b3d48c79df173fd000c9930036839
-
由 TensorFlower Gardener 提交于
PiperOrigin-RevId: 339875033 Change-Id: I46c3d5f23d9656b7d1d8ed27838048763b537bd3
-
由 Jian Li 提交于
PiperOrigin-RevId: 339873117 Change-Id: I40734689dac29e3466240f9f22d11d84171d9eae
-
由 Jacques Pienaar 提交于
The return type of function cannot be refined without considering the callee of the function. The difference is fine for TF dialect ops, but not for std.call (for example). Given this currently doesn't use a worklist across the Module and the pass refines functions in single loop, don't update return types unless known safe. Also use "can refine" in checking if can fold (rather than static type check) and enable refining shape even where element type not castable. This does incur an additional uses lookup per function refinement, which is not ideal and could be avoided by moving to worklist & operating on module internally instead. PiperOrigin-RevId: 339870845 Change-Id: Ie9fdafa0e961dbf66878fd9767d5a3685fbd00f6
-
由 Adrian Kuegel 提交于
This removes ConstWitnessOps which otherwise would block the lowering of tf.IsInf. PiperOrigin-RevId: 339864990 Change-Id: I9296481df4b4c029621e095ea7a7e13afa7b22ed
-
由 TensorFlower Gardener 提交于
PiperOrigin-RevId: 339861908 Change-Id: Idb3fc788031f646cc8f9f4977a28959f0ede7c32
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 339856428 Change-Id: I08042e653e665195a6a7b83e960d9597ee2811bb
-
https://github.com/tensorflow/tensorflow/pull/44270由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 339855048 Change-Id: Idfec167e701770053c1c481ff9e88538c03ea801
-
由 Adrian Kuegel 提交于
They return a i1 tensor. PiperOrigin-RevId: 339840456 Change-Id: I7b25d8b7c370f31a7c794117912e0e27636eee36
-
由 Renjie Liu 提交于
PiperOrigin-RevId: 339840441 Change-Id: I292db092143c9341026a3b2bae9f1eee7b64e657
-
由 Tres Popp 提交于
Additionally: - Forward listeners through new if/else op builders. This corrects an error that led to incomplete legalization of broadcasted op lowering. - Use OpConversionPattern to ensure up to date operand values are used. PiperOrigin-RevId: 339838833 Change-Id: Ib7ae51c8fd0d1e493cec0fad6d8aff2d9375d4e7
-
由 Stefano Galarraga 提交于
Fix error in NNAPI delegate ignoring the result of ANeuralNetworksModel_addOperation. This causes failures when adding other operations because of the delegate internal mapping of inputs and outputs of added operations become inconsistent with NNAPI. PiperOrigin-RevId: 339834284 Change-Id: Iede50ecdc6181e296c823b352a3f122030a31385
-
由 TensorFlower Gardener 提交于
PiperOrigin-RevId: 339833238 Change-Id: I8741a57de5b1adda350b59bb376c60449b3f08be
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 339833071 Change-Id: Ia8616679a736bd19e73a47a4df1d989919e7d9b2
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 339833070 Change-Id: I62e60687fcc5fb5b7a8aad5ba5c7a7e6e4db093f
-
由 Tres Popp 提交于
PiperOrigin-RevId: 339832292 Change-Id: Ic79dde9062fa050868ad61781d5e7e5a61661465
-
由 River Riddle 提交于
The DialectInlinerInterface now has builtin support for providing legality of call operations, as well as detecting when inlined operations will be cloned vs moved in-place. PiperOrigin-RevId: 339821908 Change-Id: I1b9200768adbe9b3c7dbcdeaed60f6176b0600a4
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 339820260 Change-Id: Ic704d3ec6c1d5e0a4155b6a49c14b977bf264716
-