| Commit message (Collapse) | Author | Age |
|
|
|
| |
PiperOrigin-RevId: 209653135
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
With the old behavior you'd have to supply a graph argument in order
to inherit state from the eager context, which doesn't make much
sense. To keep it simple, I removed the argument and made it
unconditionally inherit state from the current context or default
graph.
In addition, this reuses the current graph's collections instead of
copying them. This allows collections to be modified from inside a
FuncGraph. This is necessary for cond_v2 (at least until collections
go away?).
PiperOrigin-RevId: 209649802
|
|
|
|
|
|
|
|
| |
always available).
Previously, absl was not enabled for mobile or for makefile-based builds.
PiperOrigin-RevId: 209648505
|
|\
| |
| |
| |
| |
| | |
aaroey:fix_allocator_when_engine_device_not_specified
PiperOrigin-RevId: 209647689
|
| |
| |
| |
| |
| |
| |
| |
| | |
HeapSimulator.
The new modeling of subcomputations is still not entirely accurate, but probably not worth putting more work into, since TuplePointsToAnalysis will be removed from HeapSimulator soon.
PiperOrigin-RevId: 209646234
|
| |
| |
| |
| | |
PiperOrigin-RevId: 209645191
|
| |
| |
| |
| | |
PiperOrigin-RevId: 209644988
|
| |
| |
| |
| |
| |
| | |
function to get the precision mode.
PiperOrigin-RevId: 209641428
|
| |
| |
| |
| |
| |
| |
| |
| | |
HloEvaluator should be stable enough for reduce folding, but it shouldn't
crash when it encounters an instruction without a layout. Verify the layout on
every instruction that gets evaluated and return an error on failure.
PiperOrigin-RevId: 209641401
|
| |
| |
| |
| | |
PiperOrigin-RevId: 209640734
|
|\ \
| | |
| | |
| | | |
PiperOrigin-RevId: 209638931
|
| | |
| | |
| | |
| | | |
PiperOrigin-RevId: 209637025
|
| | |
| | |
| | |
| | | |
PiperOrigin-RevId: 209634537
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Previously, a function instantiation error (e.g. in `Dataset.map()`) would lead
to an error in each GetNext() call that attempted to use the function. Moving this
to iterator instantiation time has the benefit that the error will be reported
once when the initialization op is executed, which has a more helpful stack
trace, since it should not be conflated with other potential op failures.
PiperOrigin-RevId: 209633511
|
| | |
| | |
| | |
| | |
| | |
| | | |
2) Grow a new layer of an oblivious tree.
PiperOrigin-RevId: 209633300
|
|\ \ \
| | | |
| | | |
| | | | |
PiperOrigin-RevId: 209632789
|
| | | |
| | | |
| | | |
| | | | |
PiperOrigin-RevId: 209631832
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
BEGIN_PUBLIC
Reapply "Allow DT_INT64 input shapes for ReshapeOp." with fix.
END_PUBLIC
*** Reason for rollback ***
Reapply with fix.
*** Original change description ***
Automated rollback of commit 9f59beb67643953d87e7673fa0000cc775562693
PiperOrigin-RevId: 209629073
|
| | | |
| | | |
| | | |
| | | | |
PiperOrigin-RevId: 209627830
|
| | | |
| | | |
| | | |
| | | | |
PiperOrigin-RevId: 209627240
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Avoid the ASAN failure due to unitialized tensors.
PiperOrigin-RevId: 209624108
|
|\ \ \ \
| | | | |
| | | | |
| | | | | |
PiperOrigin-RevId: 209623532
|
| | | | |
| | | | |
| | | | |
| | | | | |
PiperOrigin-RevId: 209621853
|
| | | | |
| | | | |
| | | | |
| | | | | |
PiperOrigin-RevId: 209617148
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
* During bias_centering, the global_step on per-feature accumulators are not updated, so once the bias centering is done, we now update the accumulators to have in-sync global step.
* When a parameter server is restarted, the accumulators need to be synchronized with the stamp token after checkpoint is reloaded.
PiperOrigin-RevId: 209617008
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
ResourceStridedSliceAssign.
This will trigger a copy if there are concurrent writes to the variable.
PiperOrigin-RevId: 209615011
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
Also stop building a prioritized list every time a device is selected.
PiperOrigin-RevId: 209614803
|
| | | | |
| | | | |
| | | | |
| | | | | |
PiperOrigin-RevId: 209608984
|
| | | | |
| | | | |
| | | | |
| | | | | |
PiperOrigin-RevId: 209606318
|
| | | | |
| | | | |
| | | | |
| | | | | |
PiperOrigin-RevId: 209605597
|
| | | | |
| | | | |
| | | | |
| | | | | |
PiperOrigin-RevId: 209605066
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
The fix itself is simple but I'm not sure if this how we should be testing
auto-jit.
PiperOrigin-RevId: 209602427
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
job name.
PiperOrigin-RevId: 209597829
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
Send&recv instructions and cross-replica-sum instructions are imposing
extra dependencies via the channel id or all reduce id. This CL teaches
the reachability calculation logic in hlo computation to correctly
account for these "invisible" dependencies.
The main purpose is to stop multi output fusion from generating
dependency cyclies via communicating instructions.
PiperOrigin-RevId: 209593997
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
`docs_controls` was recently added to allow users to tag objects with `do_not_generate_docs` (it can be used as a decorator). This allows you to skip objects by _identity_, instead of path.
In my initial implementation I checked for the tag in the doc generation step.
This change fixes it to skip objects during the api-crawling step. This is necessary to avoid crawling the internals of all the excluded objects.
To enable this, the `_is_private` method of PublicAPIVisitor needs access to the object, not just the path to it. The changes in `public_api` allow this.
`generate_lib` uses the `_is_private` signature change to inspect the object.
PiperOrigin-RevId: 209587765
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
In the case where the stddev is small relative to the bounds, the sampler
is currently generating some outliers.
Added warning logging if the sampler terminates before sampling successfully.
Increased kMaxIterations from 100 to 1000 (this may not be a good long-term
solution but, in general, users probably prefer additional computation to
spurious outliers).
PiperOrigin-RevId: 209583394
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
This operation computes:
ref[i_1, ..., i_n, indices[i_1, ..., i_n, j]] = updates[i_1, ..., i_n, j]
That is, it assumes that `ref`, `indices` and `updates` have a series of leading dimensions that are the same for all of them, and the updates are performed on the last dimension of indices.
PiperOrigin-RevId: 209566652
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
Either --tf_xla_auto_jit=true or global_jit_level==ON in the session config will result in the auto jit being allowed to cluster across XlaScopes.
PiperOrigin-RevId: 209559970
|
| | | | |
| | | | |
| | | | |
| | | | | |
PiperOrigin-RevId: 209559959
|
| | | | |
| | | | |
| | | | |
| | | | | |
PiperOrigin-RevId: 209559380
|
|\ \ \ \ \
| | | | | |
| | | | | |
| | | | | | |
PiperOrigin-RevId: 209546715
|
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
PiperOrigin-RevId: 209544220
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
group_norm. Fix normalization test in test fusion.
PiperOrigin-RevId: 209534762
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
ParameterServerStrategy when using >1 device per machine. This means
wrapping the variable instances returned in that case in a class
that intercepts assign_*() method calls.
PiperOrigin-RevId: 209533673
|
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
PiperOrigin-RevId: 209533012
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
Same for WrapUnique.
PiperOrigin-RevId: 209531124
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
from registered Ops.
This initial change provides the very basics to start generating TypeScript. Non-deprecated and visible Ops are exported as a typescript function using internal functionality that is used the @tensorflow/tfjs-node repo (https://github.com/tensorflow/tfjs-node).
Future changes will introduce more code generation + tests. This initial change will help set the foundation for those upcoming changes.
PiperOrigin-RevId: 209528126
|
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
PiperOrigin-RevId: 209528032
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
rev_block.
PiperOrigin-RevId: 209524010
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
introducing a utility function for repeated code.
PiperOrigin-RevId: 209523944
|