| .. |
|
client
|
PY2 removal cleanup
|
2021-01-15 16:48:57 -08:00 |
|
experimental
|
PY2 removal cleanup
|
2021-01-15 16:48:57 -08:00 |
|
ops
|
Add option for input tensors for TPUEmbedding to have a first dimension which is a multiple of the batch_size.
|
2021-01-28 10:55:20 -08:00 |
|
profiler
|
PY2 removal cleanup
|
2021-01-15 16:48:57 -08:00 |
|
__init__.py
|
|
|
|
_tpu_estimator_embedding.py
|
|
|
|
api.py
|
|
|
|
async_checkpoint_test.py
|
|
|
|
async_checkpoint.py
|
More typing of TPU codebase.
|
2021-01-21 14:08:22 -08:00 |
|
bfloat16_test.py
|
|
|
|
bfloat16.py
|
More typing of TPU codebase.
|
2021-01-21 14:08:22 -08:00 |
|
BUILD
|
Enable async checkpoint test
|
2021-02-03 11:29:24 -08:00 |
|
datasets_test.py
|
|
|
|
datasets.py
|
More typing of TPU codebase.
|
2021-01-21 14:08:22 -08:00 |
|
device_assignment.py
|
More typing of TPU codebase.
|
2021-01-21 14:08:22 -08:00 |
|
error_handling.py
|
|
|
|
feature_column_test.py
|
Replace instances of "blacklist" with "denylist" where possible. See Google Developer guidelines at https://developers.google.com/style/word-list#blacklist for more information.
|
2020-07-20 16:05:25 -07:00 |
|
feature_column_v2_test.py
|
Change not exposed in public API
|
2020-07-31 13:16:45 -07:00 |
|
feature_column_v2.py
|
Change not exposed in public API
|
2020-07-31 13:16:45 -07:00 |
|
feature_column.py
|
Change not exposed in public API
|
2020-07-31 13:16:45 -07:00 |
|
functional.py
|
|
|
|
preempted_hook.py
|
|
|
|
session_support.py
|
Increase the ping timeout from 5 seconds to a minute. This gives a bit more breathing room in case the workers somehow become unresponsive .
|
2020-11-19 17:16:05 -08:00 |
|
tensor_tracer_flags.py
|
TensorTracer: Add an explicit error message when trace_dir is not provided.
|
2021-01-11 18:34:52 -08:00 |
|
tensor_tracer_report.py
|
Removing some unused options from tensor tracer.
|
2020-09-24 16:13:26 -07:00 |
|
tensor_tracer.proto
|
|
|
|
tensor_tracer.py
|
TensorTracer: Ignore calls for tracing tf function graphs.
|
2021-01-11 17:48:05 -08:00 |
|
topology_test.py
|
|
|
|
topology.py
|
|
|
|
tpu_config.py
|
|
|
|
tpu_context.py
|
|
|
|
tpu_embedding_gradient.py
|
Fix gradient computation bug in TPU embedding mid-level API.
|
2020-12-15 20:12:54 -08:00 |
|
tpu_embedding_v2_correctness_test.py
|
Add correctness test for ragged embeddings with the TPUEmbedding mid level API.
|
2021-01-29 12:09:52 -08:00 |
|
tpu_embedding_v2_cpu_test.py
|
Add sequence support for TPU Embeddings class on CPU.
|
2021-01-08 14:42:00 -08:00 |
|
tpu_embedding_v2_test.py
|
Update TPUEmbedding api to ensure that the table order only depends on the order of the tables in feature_config and not on the order of the table objects in a python set.
|
2021-02-11 11:12:24 -08:00 |
|
tpu_embedding_v2_utils_test.py
|
Split TPUEmbeddingConfiguration logging across multiple statements.
|
2020-10-22 14:46:54 -07:00 |
|
tpu_embedding_v2_utils.py
|
Split TPUEmbeddingConfiguration logging across multiple statements.
|
2020-10-22 14:46:54 -07:00 |
|
tpu_embedding_v2.py
|
Update TPUEmbedding api to ensure that the table order only depends on the order of the tables in feature_config and not on the order of the table objects in a python set.
|
2021-02-11 11:12:24 -08:00 |
|
tpu_embedding.py
|
Add support for FrequencyEstimator optimizer to the mid level tpu embedding API.
|
2020-12-01 17:02:39 -08:00 |
|
tpu_estimator.py
|
|
|
|
tpu_feed.py
|
Add types to tpu.py.
|
2020-11-03 14:01:22 -08:00 |
|
tpu_function.py
|
Adds a reminder to set export_saved_model_api_version=ExportSavedModelApiVersion.V2 when using TPUEstimator's inference_on_tpu.
|
2020-12-02 16:05:59 -08:00 |
|
tpu_infeed_test.py
|
|
|
|
tpu_name_util.py
|
Add types to tpu.py.
|
2020-11-03 14:01:22 -08:00 |
|
tpu_optimizer.py
|
|
|
|
tpu_outside_compilation_test.py
|
Add outside compilation test with TPUPartitionedCall.
|
2020-10-30 08:50:52 -07:00 |
|
tpu_sharding_test.py
|
Add number_of_partitions to the InfeedQueue API to allow infeed to be processed with pure data parallelism, but the partition/resharding happens inside the model function.
|
2020-07-30 13:10:17 -07:00 |
|
tpu_sharding.py
|
Add number_of_partitions to the InfeedQueue API to allow infeed to be processed with pure data parallelism, but the partition/resharding happens inside the model function.
|
2020-07-30 13:10:17 -07:00 |
|
tpu_strategy_util.py
|
Make multiple changes to enable TPU single host training e2e. The changes include:
|
2021-01-26 17:05:42 -08:00 |
|
tpu_system_metadata.py
|
|
|
|
tpu_test_wrapper_test.py
|
|
|
|
tpu_test_wrapper.bzl
|
Create no-src target for expanding visibility of TF deps.
|
2020-07-20 20:12:23 -07:00 |
|
tpu_test_wrapper.py
|
|
|
|
tpu_test.py
|
Do not cache external values in TPUReplicateContext.AddValue if there is no outer context since that is no-op. Doing so leads to name collisions in the present of nested graphs since op names are not necessarily unique across graphs.
|
2020-09-09 09:20:07 -07:00 |
|
tpu.bzl
|
|
|
|
tpu.py
|
Support passing and returning Nones in TPUStrategy.
|
2021-02-04 13:07:39 -08:00 |
|
training_loop.py
|
More typing of TPU codebase.
|
2021-01-21 14:08:22 -08:00 |
|
util.py
|
|
|
