STT-tensorflow/tensorflow/compiler/xla/python/tpu_driver
Peter Hawkins 7df1f4ffc3 [XLA:Python] Add support for collecting Python tracebacks.
Adds a new xla_client.Traceback API that can collect Python tracebacks cheaply (~2us). This is several orders of magnitude cheaper than using the Python `inspect.stack` API.

Add a facility to attach a traceback to every buffer and executable object describing its creation context. To avoid paying a runtime cost when not debugging, tracebacks collection is optional and disabled by default.

PiperOrigin-RevId: 314793876
Change-Id: Ie5f509364065739c1da4d1a8a729c4c6f56e2d03
2020-06-04 13:39:07 -07:00
..
client [XLA:Python] Add support for collecting Python tracebacks. 2020-06-04 13:39:07 -07:00
platform/external Refactoring some dependencies within TPU Driver 2019-12-19 15:03:18 -08:00
BUILD Use tf_grpc_dependency in tensorflow/compiler/xla 2020-04-02 22:42:59 -07:00
direct_tpu_driver.cc Add dependency for the direct TPU driver back into the tpu_driver target 2020-02-18 13:29:57 -08:00
event_id.h TpuDriver: Improve handling of restarted clients/servers. 2019-11-15 14:04:48 -08:00
grpc_tpu_driver.cc This CL optimizes C++11 range-based for loops where the variable is copied in each iteration but it would suffice to obtain it by const reference. This is only applied to loop variables of types that are expensive to copy which means they are not trivially copyable or have a non-trivial copy constructor or destructor. 2020-04-03 11:53:09 -07:00
grpc_tpu_driver.h Don't use grpc_impl:: as it is not API 2019-12-02 09:07:16 -08:00
README.md
recording_tpu_driver.cc This CL optimizes C++11 range-based for loops where the variable is copied in each iteration but it would suffice to obtain it by const reference. This is only applied to loop variables of types that are expensive to copy which means they are not trivially copyable or have a non-trivial copy constructor or destructor. 2020-04-03 11:53:09 -07:00
tpu_driver.cc Add shape-based allocation and linearizer to external TPU driver 2020-01-09 11:43:21 -08:00
tpu_driver.h Prefixing TensorFlow thread annotation macros with TF_. 2020-03-05 08:42:01 -08:00
tpu_driver.proto Enable TPU POD for JAX/1VM by creating devices and local_devices topology. 2020-01-14 17:04:41 -08:00
tpu_service.proto Implement GRPC TPU driver reset. 2019-11-20 17:49:06 -08:00

TPU Driver API

This repository contains the TPU driver API and network (gRPC) transport implementation for high-performance access to TPU hardware.

Building

Bazel is used to build the driver library and tests. Remote tests will require access to a Cloud TPU.

Fetching Bazel

Download the latest copy of Bazel from https://github.com/bazelbuild/bazel/releases.

Building

bazel build ...

Testing

bazel test ...