OSSForks/tensorflow
1
0
Fork 0
mirror of https://github.com/zebrajr/tensorflow.git synced 2026-01-15 12:15:41 +00:00
Code Issues Actions 5 Packages Projects Releases Wiki Activity
52,884 Commits 50 Branches 144 Tags
60e892e4bc02274992d9baa0dba585eaa63fd701
Commit Graph

52884 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
serv-inc
60e892e4bc feat: update issue template command to 2.0.0 
still works with 1.13, but the previous version failed on 2.0
 2019-04-05 13:24:36 +02:00
Dan Moldovan
a1e3d4490d More consistently pass function arguments by omitting **kwargs when the target function did not define any and the caller did not include any keywords. 
PiperOrigin-RevId: 242066459
 2019-04-04 21:30:10 -07:00
A. Unique TensorFlower
f3c82c4646 Fixed alignment for uniform buffer. 
PiperOrigin-RevId: 242064978
 2019-04-04 21:12:46 -07:00
A. Unique TensorFlower
bc1700a6f0 Automated rollback of commit 82735fd7a6 
PiperOrigin-RevId: 242063285
 2019-04-04 20:55:41 -07:00
A. Unique TensorFlower
3bf4628d75 Qualify calls to some functions from <cmath>. 
PiperOrigin-RevId: 242057286
 2019-04-04 19:46:28 -07:00
Dong Lin
82735fd7a6 Add passed and description fields to BenchmarkEntry in test_log.proto 
PiperOrigin-RevId: 242054418
 2019-04-04 19:12:59 -07:00
A. Unique TensorFlower
0c4d4090c9 Add string support to slice. 
PiperOrigin-RevId: 242045915
 2019-04-04 18:00:53 -07:00
Eugene Zhulenev
51a3acd6a0 [Grappler] Do not copy device set when optimizing function library 
PiperOrigin-RevId: 242042808
 2019-04-04 17:40:20 -07:00
TensorFlower Gardener
3f56ae58ef Merge pull request #27522 from Intel-tensorflow:intel-mkl-602 
PiperOrigin-RevId: 242038689
 2019-04-04 17:21:15 -07:00
Tiezhen WANG
ff3a98d372 Update doc. contrib is going away in TF2 
PiperOrigin-RevId: 242037150
 2019-04-04 17:11:50 -07:00
Frank Chen
aba740379b Enable auto-sharding of datasets with distribution strategies 
PiperOrigin-RevId: 242034362
 2019-04-04 17:03:51 -07:00
Thomas O'Malley
621d18e896 Disable DistributionStrategy callable add_loss test. 
PiperOrigin-RevId: 242033785
 2019-04-04 16:59:08 -07:00
Xiao Yu
781286e952 Add a few more ScopedActivity for multi-worker and collective ops. Showing these activities in profile result can help debugging. 
PiperOrigin-RevId: 242033045
 2019-04-04 16:54:30 -07:00
A. Unique TensorFlower
e5688cb6bd c++11 is the default for all compilers (Xcode 10.1+) that Apple is currently allowing you to submit iOS apps to the app store with, so get rid of the unnecessary flag. 
PiperOrigin-RevId: 242032623
 2019-04-04 16:48:40 -07:00
Guangda Lai
74a62cef28 Fix the inputs/captured_inputs/variables of the frozen concrete function by 
removing the converted nodes/tensors. This also makes the returned concrete
function callable.

PiperOrigin-RevId: 242032292
 2019-04-04 16:42:05 -07:00
Russell Power
2d869c9086 Change timeout thresholds for querying TPU metadata. 
PiperOrigin-RevId: 242031057
 2019-04-04 16:36:00 -07:00
A. Unique TensorFlower
c77d651611 Add dependency to make xla.einsum run on Cloud TPU. 
PiperOrigin-RevId: 242030629
 2019-04-04 16:28:02 -07:00
TensorFlower Gardener
3678c5dd5e Merge pull request #27514 from frreiss:issue-unusued 
PiperOrigin-RevId: 242028806
 2019-04-04 16:19:03 -07:00
Rick Chao
def33ac186 Roll back the change to make ModelCheckpoint.epochs_since_last_save attribute private to unbreak tests. 
PiperOrigin-RevId: 242025639
 2019-04-04 16:13:05 -07:00
Tong Shen
61b1792470 If Const node in outside compilation at head has successor that's not outside compilation at head, do not move it out of TPU computation. 
Example: if we put a TopKV2 op in outside compilation, its "input" is from TPU and its "k" is a Const node in outside compilation, original logic will move "k" out of TPU computation and leave TopKV2 in outside compilation. So the shape inference graph for outside compilation will have an XlaRecvAtHost node as TopKV2's "k" input, and we cannot infer output shape for the TopKV2 node any more.

Solution is to keep the "k" Const in outside compilation with the TopKV2.

PiperOrigin-RevId: 242024072
 2019-04-04 16:08:09 -07:00
TensorFlower Gardener
81e8f5e1a2 Merge pull request #27350 from ANSHUMAN87:arena-planner-rephrase 
PiperOrigin-RevId: 242024012
 2019-04-04 16:01:44 -07:00
Sanjoy Das
a50e8e64df [TF:XLA] Bump open source abseil revision to 666fc1266bccfd8e6eaaa084e7b42580bb8eb199 
PiperOrigin-RevId: 242023659
 2019-04-04 15:56:38 -07:00
TensorFlower Gardener
66633c57f7 Merge pull request #27513 from mattn:fix-make-doc 
PiperOrigin-RevId: 242023037
 2019-04-04 15:49:11 -07:00
Thomas O'Malley
2f6edb4f0c Fix custom initializer and regularizer loading. 
PiperOrigin-RevId: 242022733
 2019-04-04 15:43:11 -07:00
A. Unique TensorFlower
210c75c448 Fix dependency script 
PiperOrigin-RevId: 242018962
 2019-04-04 15:23:52 -07:00
Sung Jin Hwang
bc7cb41c62 Remove checking if argument Tensor is initialized in call frame. 
In the existing code, when an argument Tensor is not initialized, it falls through the if-else chain and the first else branch catches it, and will reference out-of-range memory region in vector.

Looking at the code, there does not seem to be a strong reason to check whether an argument is initialized or not. So removing that condition in if cluase.

PiperOrigin-RevId: 242015796
 2019-04-04 15:08:42 -07:00
TensorFlower Gardener
e64922ff8f Merge pull request #27481 from amanp592:patch-1 
PiperOrigin-RevId: 242012088
 2019-04-04 14:51:15 -07:00
A. Unique TensorFlower
27590b0e09 Added optimized Reshapex4 
Improved work-group size picking.

PiperOrigin-RevId: 242010654
 2019-04-04 14:43:11 -07:00
Saurabh Saxena
3b16606bbf Enable testWhileAutoControlDeps in eager mode. 
PiperOrigin-RevId: 242007977
 2019-04-04 14:28:43 -07:00
Russell Power
1e73e6027a TPUEstimator: graceful shutdown improvements. 
PiperOrigin-RevId: 242004381
 2019-04-04 14:18:57 -07:00
Pete Warden
53b237d357 Internal debugging change 
PiperOrigin-RevId: 242003375
 2019-04-04 14:13:29 -07:00
Jared Duke
3fe8c6b43b Enable GPU in TFLite camera demo for bazel builds 
PiperOrigin-RevId: 242002157
 2019-04-04 14:08:01 -07:00
Eugene Zhulenev
3f8281e762 Add missing break to the switch statement. 
PiperOrigin-RevId: 242000242
 2019-04-04 14:02:24 -07:00
A. Unique TensorFlower
4591bad9f9 Automated rollback of commit fe30579d67 
PiperOrigin-RevId: 241998820
 2019-04-04 13:57:10 -07:00
Sanjoy Das
65bc758aee Expose the results of deadness analysis more directly 
Instead of a HasInputsWithMismatchingDeadness routine expose a way for clients
of DeadnessAnalysis to directly infer the predicate that tells us when a node is
dead.  Then the clustering constraint is: all nodes being clustered have the
same predicate.

Note that this is *not* an NFC change, and can actually change clustering.  For
instance if we have:

digraph {
  A -> C
  B -> C
  A -> D
  B -> D
  C -> E
  D -> E
}

with A and B with different liveness (let's assume neither A nor B are
compilable by XLA), we previously would not cluster C or D.  But now we would
because we can tell that C, D and E all have the same predicate: IsAlive(A) &
IsAlive(B).  I have added tests demonstrating this change in behavior.

This CL also fixes a bug in jit/xla_launch_util.cc where we were missing a
subtraction from missing_ctx_input_prefix.

PiperOrigin-RevId: 241998170
 2019-04-04 13:51:07 -07:00
A. Unique TensorFlower
862d326bff Change grad of SoftPlus to use the sigmoid function which works better for higher order gradients. 
PiperOrigin-RevId: 241997428
 2019-04-04 13:45:37 -07:00
Anna R
40e666c30b Automated rollback of commit 4ee64e012a 
PiperOrigin-RevId: 241991334
 2019-04-04 13:14:26 -07:00
A. Unique TensorFlower
359e02ef20 [XLA:CPU] Adds HLO-level tracing. 
PiperOrigin-RevId: 241989294
 2019-04-04 13:08:52 -07:00
Sanjoy Das
65ee4a62b8 Add a --xla_gpu_disable_autotune flag 
This is useful for checking whether a test failure was caused by a
non-deterministic tuning result.

PiperOrigin-RevId: 241988618
 2019-04-04 13:03:06 -07:00
Jared Duke
a0104b7a5a Tweak NDK API level parsing for Android 
Allow setting the level via $ANDROID_NDK_API_LEVEL, rather than
hardcoding to the current NDK version. Also increase the default
NDK API level to 18.

PiperOrigin-RevId: 241981236
 2019-04-04 12:36:28 -07:00
Xiao Yu
4b2d9127bc Add monitoring python APIs. These API will be used by Keras API to export metrics. 
PiperOrigin-RevId: 241979112
 2019-04-04 12:31:14 -07:00
Jared Duke
1b77aae959 Add GPU and NNAPI delegates to benchmark_model tool 
Delegates can be enabled via:
  --use_nnapi=true
  --use_gpu=true

Existing NNAPI functionality has been renamed to:
  --use_legacy_nnapi=true

PiperOrigin-RevId: 241979034
 2019-04-04 12:24:26 -07:00
TensorFlower Gardener
3ce688f5a7 Merge pull request #27358 from chenchc:fix-bfc-overflow 
PiperOrigin-RevId: 241976268
 2019-04-04 12:17:12 -07:00
TensorFlower Gardener
26dc36f2b1 Merge pull request #27391 from feihugis:Test_InterleaveDatasetOp 
PiperOrigin-RevId: 241976029
 2019-04-04 12:10:41 -07:00
Saurabh Saxena
4efd2019db Do not set on_host true when forwarding buffer in TensorLists. 
PiperOrigin-RevId: 241975751
 2019-04-04 12:03:56 -07:00
Pavithra Vijay
cc142356ab Automated rollback of commit cdd0dcc6f7 
PiperOrigin-RevId: 241968545
 2019-04-04 11:39:53 -07:00
Sachin Joglekar
998a3f9c45 Internal Change 
PiperOrigin-RevId: 241966695
 2019-04-04 11:25:25 -07:00
A. Unique TensorFlower
2e41f50d17 Switch centos remote platforms to centos6. 
PiperOrigin-RevId: 241963613
 2019-04-04 11:10:39 -07:00
nammbash
1ab8935f43 graph transform bug 2019-04-04 10:57:35 -07:00
Brian Lee
ee45529b5b Reword exception message for failed AutoGraph conversion 
PiperOrigin-RevId: 241957881
 2019-04-04 10:48:11 -07:00