- 26 2月, 2016 9 次提交
-
-
由 A. Unique TensorFlower 提交于
Fix an error message in tf.sparse_to_dense to include the possibility that indices are invalid because they are out of bounds. Change: 115522264
-
由 Eugene Brevdo 提交于
These tools are meant to allow recording of benchmark & unit test structured output to pbtxt files in a directory only when the environment variable TEST_REPORT_FILE_PREFIX is set. For now, only saving of C++ microbenchmark output is supported. Change: 115518303
-
由 Sherry Moore 提交于
Change: 115516426
-
由 Kiril Gorovoy 提交于
Change: 115515678
-
由 Josh Levenberg 提交于
Change: 115511835
-
由 Josh Levenberg 提交于
Change: 115511794
-
由 Vincent Vanhoucke 提交于
Helps with: https://github.com/tensorflow/tensorflow/issues/917 Also fixes https://github.com/tensorflow/tensorflow/issues/1162 The main benefit is that the computation of the sufficient statistics is now decoupled of the aggregation of the moments, which means that if you want to perform the accumulation incrementally, you don't have to keep all the inputs around, and can instead keep the much more compact sum and sum-of-squares. Accumulation could also be performed locally if you aggregate across multiple devices. Computing sum and sum-of-squares can also theoretically be performed in parallel now. Tested running inception: same performance, same step time. Batch normalization benchmark is a bit faster on CPU, a bit slower on GPU: Before: cpu shape:4/3 #layers:10 mode:py scale:True train:False - 1.139310 secs gpu shape:4/3 #layers:10 mode:py scale:True train:False - 0.021970 secs cpu shape:4/3 #layers:10 mode:py scale:True train:True - 2.767147 secs gpu shape:4/3 #layers:10 mode:py scale:True train:True - 0.074531 secs cpu shape:4/3 #layers:10 mode:py scale:True train:False - 0.742835 secs gpu shape:4/3 #layers:10 mode:py scale:True train:False - 0.013473 secs cpu shape:4/3 #layers:10 mode:py scale:True train:True - 1.738806 secs gpu shape:4/3 #layers:10 mode:py scale:True train:True - 0.052777 secs cpu shape:2/1 #layers:10 mode:py scale:True train:False - 0.119180 secs gpu shape:2/1 #layers:10 mode:py scale:True train:False - 0.011201 secs cpu shape:2/1 #layers:10 mode:py scale:True train:True - 0.218297 secs gpu shape:2/1 #layers:10 mode:py scale:True train:True - 0.048526 secs After: cpu shape:4/3 #layers:10 mode:py scale:True train:False - 0.998944 secs gpu shape:4/3 #layers:10 mode:py scale:True train:False - 0.025828 secs cpu shape:4/3 #layers:10 mode:py scale:True train:True - 2.657428 secs gpu shape:4/3 #layers:10 mode:py scale:True train:True - 0.086614 secs cpu shape:4/3 #layers:10 mode:py scale:True train:False - 0.603137 secs gpu shape:4/3 #layers:10 mode:py scale:True train:False - 0.017668 secs cpu shape:4/3 #layers:10 mode:py scale:True train:True - 1.519533 secs gpu shape:4/3 #layers:10 mode:py scale:True train:True - 0.055214 secs cpu shape:2/1 #layers:10 mode:py scale:True train:False - 0.071344 secs gpu shape:2/1 #layers:10 mode:py scale:True train:False - 0.016440 secs cpu shape:2/1 #layers:10 mode:py scale:True train:True - 0.222093 secs gpu shape:2/1 #layers:10 mode:py scale:True train:True - 0.039967 secs Change: 115507032
-
由 Josh Levenberg 提交于
RemoveNewDefaultAttrsFromGraphDef(). Change: 115506523
-
由 A. Unique TensorFlower 提交于
Change: 115505008
-
- 25 2月, 2016 18 次提交
-
-
由 Benoit Steiner 提交于
X1 crashes when attempting to compile them Change: 115500414
-
由 Eugene Brevdo 提交于
Change: 115496194
-
由 Geoffrey Irving 提交于
Change: 115495726
-
由 Eugene Brevdo 提交于
Change: 115494526
-
由 A. Unique TensorFlower 提交于
Support leaving the offset (beta) parameter out in batch_normalization, in which case no offset will be added after normalization. Change: 115489328
-
由 A. Unique TensorFlower 提交于
Change: 115472914
-
由 A. Unique TensorFlower 提交于
Rename map in control_flow_ops to map_fn, to avoid name conflict with Python's native 'map' function. This also fixes the bug with control_flow_ops.case Change: 115472163
-
由 David G. Andersen 提交于
Change: 115470945
-
由 A. Unique TensorFlower 提交于
Describe how to load many runs. Change: 115467346
-
由 Geoffrey Irving 提交于
Both gather and scatter now unconditionally validate indices in the inner loop, which prevents crashes if indices are changed asynchronously while the ops are running. For gather when validate_indices = true, the new code is within the noise of the old code speedwise or possibly slightly faster (unsurprising since the new code fuses two loops). Specifically, the geometric mean of int32 gather benchmarks goes from 4.05GB/s to 4.04-4.07GB/s. For gather when validate_indices = false, the old code and a version of the old code that supported validate_indices = false both get 1.5% slower. Xiaoqiang and I deem this difference insufficient to preserve the unsafe code path, so poof: it's gone. For scatter (which always validates), the new code is slightly faster than the old code: the geometric mean goes from 546-559M items/s to 573M items/s. Change: 115467091
-
由 Yuan Yu 提交于
Change: 115464489
-
由 Eugene Brevdo 提交于
Change: 115464229
-
由 Vijay Vasudevan 提交于
to be used to colocate based on attributes rather than either names of ops or devices (op names and devices aren't portable). A follow up change will add an ops.colocate_with() to Python that adds this attribute to nodes, and will be used to replace calls to 'with tf.device(foo.device)' in TF library code, which assumes that devices have been specified. Change: 115463464
-
由 A. Unique TensorFlower 提交于
Change: 115462062
-
由 A. Unique TensorFlower 提交于
Change: 115419426
-
由 A. Unique TensorFlower 提交于
Change: 115408162
-
由 Josh Levenberg 提交于
Change: 115384748
-
由 Josh Levenberg 提交于
Change: 115379524
-
- 24 2月, 2016 13 次提交
-
-
由 Manjunath Kudlur 提交于
-
由 James Wexler 提交于
Change: 115371065
-
由 A. Unique TensorFlower 提交于
Change: 115370821
-
由 A. Unique TensorFlower 提交于
Reason: tsd is deprecated (https://github.com/DefinitelyTyped/tsd/issues/269) and typings is the new standard. Also, tsd was behaving badly - running `tsd install` on a clean client was causing it to incorrectly depend on typing files from node_modules, which resulted in a broken build. This issue does not exist with typings. For convenience, and since typings is really fast when all deps are up-to-date, I made it a part of the standard gulp task. `npm install` so you have all the deps, and running `gulp` will keep the typing files synchronized - there no longer is a separate step for downloading them. The logical next step is to do the same for bower. I did wire that up, but I will not connect it to the gulp task until after the big bower dependency upgrade CL is through. If I add it right now, it will fail on unresolved dependency conflicts and make everyone sad. Change: 115370585
-
由 Vijay Vasudevan 提交于
Change: 115364038
-
由 A. Unique TensorFlower 提交于
The tutorial Python files are copied to a separate directory and run against Python installation on the system. The script performs basic quality checks, including timeout, accuracy / loss thresholding, and checks for generated checkpoint files and summary files. Change: 115362939
-
由 Yuan Yu 提交于
For gradient computation for loops, stacks are used to store the tensors that are computed in the forward but needed in backprop. This CL enables very long sequence training by swapping the stack tensors from GPU to CPU. Change: 115359847
-
由 Derek Murray 提交于
Change: 115358623
-
由 James Wexler 提交于
Change: 115354844
-
由 Yuan Yu 提交于
Change: 115351830
-
由 Geoffrey Irving 提交于
Change: 115347996
-
由 Derek Murray 提交于
The PickUnusedPortOrDie implementation is based on a simplified version of `grpc_pick_unused_port_or_die()` in gRPC. This utility will be necessary for tests of the distributed runtime (issue #23). Change: 115345502
-
由 Dan Smilkov 提交于
Also: - rename sceneBehavior -> sceneElement to make it clearer it is a polymer element. - improve the info card by showing the actual op node in the successors/predecessors list when the metaedge only contains one base edge (one tensor). Change: 115339805
-