Update TF compat protos.

(cherry picked from commit 9531583)

Author: arcra

tensorboard/compat/proto/config.proto

@@ -2,14 +2,14 @@ syntax = "proto3";
 
 package tensorboard;
 
+import "tensorboard/compat/proto/coordination_config.proto";
 import "tensorboard/compat/proto/cost_graph.proto";
 import "tensorboard/compat/proto/graph.proto";
 import "tensorboard/compat/proto/step_stats.proto";
 import "tensorboard/compat/proto/cluster.proto";
 import "tensorboard/compat/proto/debug.proto";
 import "tensorboard/compat/proto/rewriter_config.proto";
 import "tensorboard/compat/proto/rpc_options.proto";
-import "tensorboard/compat/proto/coordination_config.proto";
 
 option cc_enable_arenas = true;
 option java_outer_classname = "ConfigProtos";
@@ -77,6 +77,11 @@ message GPUOptions {
   //    name "/device:GPU:<id>") are also called "TF GPU id"s. Please
   //    refer to third_party/tensorflow/core/common_runtime/gpu/gpu_id.h
   //    for more information.
+  // 3. The visible_device_list is also used for PluggableDevice. And
+  //    different types of PluggableDevices share this field. In that case,
+  //    the pluggable_device_type is used to distinguish them, making the
+  //    visible_device_list a list of <pluggable_device_type>:<device_index>,
+  //    e.g. "PluggableDeviceA:0,PluggableDeviceA:1,PluggableDeviceB:0".
   string visible_device_list = 5;
 
   // In the event polling loop sleep this many microseconds between
@@ -306,7 +311,7 @@ message GPUOptions {
 
   // Everything inside experimental is subject to change and is not subject
   // to API stability guarantees in
-  // https://www.tensorflow.org/guide/version_compat.
+  // https://www.tensorflow.org/guide/versions.
   Experimental experimental = 9;
 }
 
@@ -578,7 +583,7 @@ message ConfigProto {
 
   // Everything inside Experimental is subject to change and is not subject
   // to API stability guarantees in
-  // https://www.tensorflow.org/guide/version_compat.
+  // https://www.tensorflow.org/guide/versions.
   message Experimental {
     // Task name for group resolution.
     string collective_group_leader = 1;
@@ -750,7 +755,7 @@ message ConfigProto {
     bool disable_functional_ops_lowering = 21;
 
     // Provides a hint to XLA auto clustering to prefer forming a single large
-    // cluster that encompases most of the graph.
+    // cluster that encompasses most of the graph.
     bool xla_prefer_single_graph_cluster = 22;
 
     // Distributed coordination service configurations.
@@ -983,3 +988,28 @@ message CallableOptions {
 
   // Next: 9
 }
+
+message BatchingOptions {
+  // Number of scheduling threads for processing batches of work. Determines
+  // the number of batches processed in parallel. This should be roughly in line
+  // with the number of TPU cores available.
+  int32 num_batch_threads = 1;
+
+  // The maximum allowed batch size. Can be larger than allowed_batch_sizes to
+  // utilize large batch splitting.
+  int32 max_batch_size = 2;
+
+  // Maximum number of microseconds to wait before outputting an incomplete
+  // batch.
+  int32 batch_timeout_micros = 3;
+
+  // Optional list of allowed batch sizes. If left empty, does nothing.
+  // Otherwise, supplies a list of batch sizes, causing the op to pad batches up
+  // to one of those sizes. The entries must increase monotonically, and the
+  // final entry must be equal or less than the max_batch_size.
+  repeated int32 allowed_batch_sizes = 4;
+
+  // Maximum number of batches enqueued for processing before requests are
+  // failed fast.
+  int32 max_enqueued_batches = 5;
+}

tensorboard/compat/proto/coordination_config.proto

@@ -29,6 +29,10 @@ message CoordinationServiceConfig {
   // Maximum wait time for all members in the cluster to be registered.
   int64 cluster_register_timeout_in_ms = 4;
 
+  // Denotes if we should synchronize the agents' register attempts by blocking
+  // on a barrier. This is useful for synchronized restarts.
+  bool cluster_register_with_barrier = 14;
+
   // Heartbeat timeout, if a task does not record heartbeat in this time
   // window, it will be considered disconnected.
   // Note: This is also used as a grace period to accept any heartbeats after

tensorboard/compat/proto/types.proto

@@ -43,10 +43,13 @@ enum DataType {
   DT_FLOAT8_E5M2 = 24;    // 5 exponent bits, 2 mantissa bits.
   DT_FLOAT8_E4M3FN = 25;  // 4 exponent bits, 3 mantissa bits, finite-only, with
                           // 2 NaNs (0bS1111111).
-  // TODO - b/299182407: Leaving room for remaining float8 types.
-  // DT_FLOAT8_E4M3FNUZ = 26;
-  // DT_FLOAT8_E4M3B11FNUZ = 27;
-  // DT_FLOAT8_E5M2FNUZ = 28;
+  DT_FLOAT8_E4M3FNUZ = 26;     // 4 exponent bits, 3 mantissa bits, finite-only,
+                               // with NaN.
+  DT_FLOAT8_E4M3B11FNUZ = 27;  // 4 exponent bits, 3 mantissa bits, 11 bits
+                               // bias, finite-only, with NaNs.
+  DT_FLOAT8_E5M2FNUZ = 28;     // 5 exponent bits, 2 mantissa bits, finite-only,
+                               // with NaN.
+
   DT_INT4 = 29;
   DT_UINT4 = 30;
 
@@ -78,10 +81,10 @@ enum DataType {
   DT_UINT64_REF = 123;
   DT_FLOAT8_E5M2_REF = 124;
   DT_FLOAT8_E4M3FN_REF = 125;
-  // TODO - b/299182407: Leaving room for remaining float8 types.
-  // DT_FLOAT8_E4M3FNUZ_REF = 126;
-  // DT_FLOAT8_E4M3B11FNUZ_REF = 127;
-  // DT_FLOAT8_E5M2FNUZ_REF = 128;
+
+  DT_FLOAT8_E4M3FNUZ_REF = 126;
+  DT_FLOAT8_E4M3B11FNUZ_REF = 127;
+  DT_FLOAT8_E5M2FNUZ_REF = 128;
   DT_INT4_REF = 129;
   DT_UINT4_REF = 130;
 }