From 662185f3f17b53c014192a1d5094b65e1d00ec3a Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Sven=20=C4=86urkovi=C4=87?= <sven.curkovic@gmail.com>
Date: Wed, 29 Jan 2025 09:42:53 +0100
Subject: [PATCH 1/2] feat: add prediction_log.proto

---
 proto/tensorflow_serving/apis/logging.proto   | 17 +++++++
 .../apis/prediction_log.proto                 | 49 +++++++++++++++++++
 update.sh                                     |  1 +
 3 files changed, 67 insertions(+)
 create mode 100644 proto/tensorflow_serving/apis/logging.proto
 create mode 100644 proto/tensorflow_serving/apis/prediction_log.proto

diff --git a/proto/tensorflow_serving/apis/logging.proto b/proto/tensorflow_serving/apis/logging.proto
new file mode 100644
index 0000000..6298bb4
--- /dev/null
+++ b/proto/tensorflow_serving/apis/logging.proto
@@ -0,0 +1,17 @@
+syntax = "proto3";
+
+package tensorflow.serving;
+
+import "tensorflow_serving/apis/model.proto";
+import "tensorflow_serving/config/logging_config.proto";
+
+option cc_enable_arenas = true;
+
+// Metadata logged along with the request logs.
+message LogMetadata {
+  ModelSpec model_spec = 1;
+  SamplingConfig sampling_config = 2;
+  // List of tags used to load the relevant MetaGraphDef from SavedModel.
+  repeated string saved_model_tags = 3;
+  // TODO(b/33279154): Add more metadata as mentioned in the bug.
+}
diff --git a/proto/tensorflow_serving/apis/prediction_log.proto b/proto/tensorflow_serving/apis/prediction_log.proto
new file mode 100644
index 0000000..c75b68a
--- /dev/null
+++ b/proto/tensorflow_serving/apis/prediction_log.proto
@@ -0,0 +1,49 @@
+syntax = "proto3";
+
+package tensorflow.serving;
+
+import "tensorflow_serving/apis/classification.proto";
+import "tensorflow_serving/apis/inference.proto";
+import "tensorflow_serving/apis/logging.proto";
+import "tensorflow_serving/apis/predict.proto";
+import "tensorflow_serving/apis/regression.proto";
+import "tensorflow_serving/apis/session_service.proto";
+
+option cc_enable_arenas = true;
+
+message ClassifyLog {
+  ClassificationRequest request = 1;
+  ClassificationResponse response = 2;
+}
+
+message RegressLog {
+  RegressionRequest request = 1;
+  RegressionResponse response = 2;
+}
+
+message PredictLog {
+  PredictRequest request = 1;
+  PredictResponse response = 2;
+}
+
+message MultiInferenceLog {
+  MultiInferenceRequest request = 1;
+  MultiInferenceResponse response = 2;
+}
+
+message SessionRunLog {
+  SessionRunRequest request = 1;
+  SessionRunResponse response = 2;
+}
+
+// Logged model inference request.
+message PredictionLog {
+  LogMetadata log_metadata = 1;
+  oneof log_type {
+    ClassifyLog classify_log = 2;
+    RegressLog regress_log = 3;
+    PredictLog predict_log = 6;
+    MultiInferenceLog multi_inference_log = 4;
+    SessionRunLog session_run_log = 5;
+  }
+}
diff --git a/update.sh b/update.sh
index 3939581..1e9f094 100755
--- a/update.sh
+++ b/update.sh
@@ -43,3 +43,4 @@ function importProto {( set -e
 for i in ${repo}/${apis}/*_service.proto; do
   importProto ${i#${repo}/}
 done
+importProto ${apis}/prediction_log.proto

From 3890e2432d2f076e1e2ae62fc03e4d211073eb2e Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Sven=20=C4=86urkovi=C4=87?= <sven.curkovic@gmail.com>
Date: Wed, 29 Jan 2025 09:48:34 +0100
Subject: [PATCH 2/2] chore: update to 2.18.0

---
 VERSION                                       |   2 +-
 proto/tensorflow/core/example/example.proto   |   9 +-
 .../core/framework/attr_value.proto           |   2 +-
 .../tensorflow/core/framework/full_type.proto | 166 ++++++++++--
 proto/tensorflow/core/framework/graph.proto   |   4 +
 .../core/framework/graph_debug_info.proto     |  61 +++++
 .../tensorflow/core/framework/node_def.proto  |   7 +
 .../core/framework/resource_handle.proto      |   2 +
 proto/tensorflow/core/framework/tensor.proto  |   5 +
 proto/tensorflow/core/framework/types.proto   |  29 +-
 proto/tensorflow/core/protobuf/cluster.proto  |   3 +
 proto/tensorflow/core/protobuf/config.proto   | 256 ++++++++++++------
 .../core/protobuf/error_codes.proto           | 155 +----------
 .../tensorflow/core/protobuf/meta_graph.proto |  74 +----
 .../core/protobuf/rewriter_config.proto       |  30 +-
 .../core/protobuf/rpc_options.proto           |   7 +
 .../core/protobuf/saved_object_graph.proto    |  42 ++-
 proto/tensorflow/core/protobuf/struct.proto   |   9 +-
 .../protobuf/trackable_object_graph.proto     |  28 +-
 .../apis/classification.proto                 |   4 +-
 proto/tensorflow_serving/apis/logging.proto   |   5 +-
 proto/tensorflow_serving/apis/model.proto     |   9 +-
 .../apis/model_management.proto               |   9 +
 proto/tensorflow_serving/apis/predict.proto   | 145 +++++++++-
 .../apis/prediction_log.proto                 |   6 +
 .../tensorflow_serving/apis/regression.proto  |   4 +-
 .../file_system_storage_path_source.proto     |  11 +-
 .../config/logging_config.proto               |  17 +-
 28 files changed, 744 insertions(+), 357 deletions(-)
 create mode 100644 proto/tensorflow/core/framework/graph_debug_info.proto
 create mode 100644 proto/tensorflow/core/protobuf/rpc_options.proto

diff --git a/VERSION b/VERSION
index e70b452..cf86907 100644
--- a/VERSION
+++ b/VERSION
@@ -1 +1 @@
-2.6.0
+2.18.0
diff --git a/proto/tensorflow/core/example/example.proto b/proto/tensorflow/core/example/example.proto
index 0b49514..59e5428 100644
--- a/proto/tensorflow/core/example/example.proto
+++ b/proto/tensorflow/core/example/example.proto
@@ -12,7 +12,6 @@ option java_multiple_files = true;
 option java_package = "org.tensorflow.example";
 option go_package = "github.com/tensorflow/tensorflow/tensorflow/go/core/example";
 
-// LINT.IfChange
 // An Example is a mostly-normalized data format for storing data for
 // training and inference.  It contains a key-value store (features); where
 // each key (string) maps to a Feature message (which is oneof packed BytesList,
@@ -22,10 +21,8 @@ option go_package = "github.com/tensorflow/tensorflow/tensorflow/go/core/example
 // read and write this format.  That is, the Example is mostly *not* a
 // self-describing format.  In TensorFlow, Examples are read in row-major
 // format, so any configuration that describes data with rank-2 or above
-// should keep this in mind.  For example, to store an M x N matrix of Bytes,
-// the BytesList must contain M*N bytes, with M rows of N contiguous values
-// each.  That is, the BytesList value must store the matrix as:
-//     .... row 0 .... .... row 1 .... // ...........  // ... row M-1 ....
+// should keep this in mind. If you flatten a matrix into a FloatList it should
+// be stored as [ row 0 ... row 1 ... row M-1 ]
 //
 // An Example for a movie recommendation application:
 //   features {
@@ -302,5 +299,3 @@ message SequenceExample {
   Features context = 1;
   FeatureLists feature_lists = 2;
 }
-// LINT.ThenChange(
-//     https://www.tensorflow.org/code/tensorflow/python/training/training.py)
diff --git a/proto/tensorflow/core/framework/attr_value.proto b/proto/tensorflow/core/framework/attr_value.proto
index 3ca031b..06024cc 100644
--- a/proto/tensorflow/core/framework/attr_value.proto
+++ b/proto/tensorflow/core/framework/attr_value.proto
@@ -27,7 +27,7 @@ message AttrValue {
     repeated TensorProto tensor = 8;             // "list(tensor)"
     repeated NameAttrList func = 9;              // "list(attr)"
   }
-  // LINT.ThenChange(https://www.tensorflow.org/code/tensorflow/c/c_api.cc)
+  // LINT.ThenChange(//tensorflow/c/c_api.cc)
 
   oneof value {
     bytes s = 2;                 // "string"
diff --git a/proto/tensorflow/core/framework/full_type.proto b/proto/tensorflow/core/framework/full_type.proto
index 34e432a..0022abf 100644
--- a/proto/tensorflow/core/framework/full_type.proto
+++ b/proto/tensorflow/core/framework/full_type.proto
@@ -8,6 +8,7 @@ option java_multiple_files = true;
 option java_package = "org.tensorflow.framework";
 option go_package = "github.com/tensorflow/tensorflow/tensorflow/go/core/framework";
 
+// LINT.IfChange
 // Experimental. Represents the complete type information of a TensorFlow value.
 enum FullTypeId {
   // The default represents an uninitialized values.
@@ -25,7 +26,7 @@ enum FullTypeId {
   //   TFT_TENSOR[TFT_VAR["T"]], TFT_TENSOR[TFT_VAR["T"]] are two tensors of
   //     identical element types.
   //   TFT_TENSOR[TFT_VAR["P"]], TFT_TENSOR[TFT_VAR["Q"]] are two tensors of
-  //     potentially different element types.
+  //     independent element types.
   //
   TFT_VAR = 1;
 
@@ -45,14 +46,53 @@ enum FullTypeId {
   //
   TFT_PRODUCT = 3;
 
+  // Represents a named field, with the name stored in the attribute.
+  //
+  // Parametrization:
+  //   TFT_NAMED[<type>]{<name>}
+  //   * <type> is the type of the field
+  //   * <name> is the field name, as string (thpugh can theoretically be an int
+  //     as well)
+  //
+  // Example:
+  //   TFT_RECORD[
+  //     TFT_NAMED[TFT_TENSOR[TFT_INT32]]{'foo'},
+  //     TFT_NAMED[TFT_TENSOR[TFT_FLOAT32]]{'bar'},
+  //   ]
+  //     is a structure with two fields, an int tensor "foo" and a float tensor
+  //     "bar".
+  TFT_NAMED = 4;
+
+  // Template definition. Expands the variables by repeating a template as
+  // arguments of container.
+  //
+  // Parametrization:
+  //   TFT_FOR_EACH[<container_type>, <template>, <expansions>]
+  //   * <container_type> is the type of the container that the template will be
+  //     expanded into
+  //   * <template> is any type definition that potentially contains type
+  //     variables
+  //   * <expansions> is a TFT_VAR and may include more types in the future
+  //
+  // Example:
+  //   TFT_FOR_EACH[
+  //         TFT_PRODUCT,
+  //         TFT_TENSOR[TFT_VAR["t"]],
+  //         TFT_VAR["t"]
+  //     ]
+  //     will substitute a T = TFT_INT32 to TFT_PRODUCT[TFT_TENSOR[TFT_INT32]]
+  //     and a T = (TFT_INT32, TFT_INT64) to
+  //     TFT_PRODUCT[TFT_TENSOR[TFT_INT32], TFT_TENSOR[TFT_INT64]].
+  TFT_FOR_EACH = 20;
+
   // Callable types describe functions and ops.
   //
   // Parametrization:
   //   TFT_CALLABLE[<arg type>, <return type>]
-  //   * <arg_type> is the type of the arguments; TFT_PRODUCT represents
+  //   * <arg type> is the type of the arguments; TFT_PRODUCT represents
   //   multiple
   //     arguments.
-  //   * <return_type> is the return type; TFT_PRODUCT represents multiple
+  //   * <return type> is the return type; TFT_PRODUCT represents multiple
   //     return values (that means that callables returning multiple things
   //     don't necessarily return a single tuple).
   //
@@ -73,9 +113,9 @@ enum FullTypeId {
   //
   // Parametrization:
   //   TFT_TENSOR[<element type>, <shape type>]
-  //   * <element_type> is currently limited to one of the element types
+  //   * <element type> is currently limited to one of the element types
   //     defined below.
-  //   * <shape_type> is not yet defined, and may only be TFT_UNKNOWN for now.
+  //   * <shape type> is not yet defined, and may only be TFT_UNKNOWN for now.
   //
   // A TFT_SHAPE type will be defined in the future.
   //
@@ -97,7 +137,7 @@ enum FullTypeId {
   //
   // Parametrization:
   //   TFT_ARRAY[<element type>]
-  //   * <element_type> may be any concrete type.
+  //   * <element type> may be any concrete type.
   //
   // Examples:
   //   TFT_ARRAY[TFT_TENSOR[TFT_INT32]] is a TensorArray holding int32 Tensors
@@ -115,35 +155,47 @@ enum FullTypeId {
   //
   // Parametrization:
   //   TFT_OPTIONAL[<element type>]
-  //   * <element_type> may be any concrete type.
+  //   * <element type> may be any concrete type.
   //
   // Examples:
   //   TFT_OPTIONAL[TFT_TENSOR[TFT_INT32]] is an Optional holding an int32
   //     Tensor of any shape.
   TFT_OPTIONAL = 1002;
 
-  // Datasets created by tf.data ops and APIs. Datasets have generator/iterable
-  // semantics, that is, one can construct an iterator from them. Like
-  // Array, they are considered to return elements that can be described
-  // by a single type. Unlike Array, they do not support random access or
-  // mutation, and can potentially produce an infinite number of elements.
-  // A datasets can produce logical structures (e.g. multiple elements). This
-  // is expressed using TFT_PRODUCT.
+  // Literal types describe compile-time constant values.
+  // Literal types may also participate in dependent types.
+  //
+  // Parametrization:
+  //   TFT_LITERAL[<value type>]{<value>}
+  //   * <value type> may be any concrete type compatible that can hold <value>
+  //   * <value> is the type's attribute, and holds the actual literal value
   //
+  // Examples:
+  //   TFT_LITERAL[TFT_INT32]{1} is the compile-time constant 1.
+  TFT_LITERAL = 1003;
+
+  // Encoding types describe a value of a certain type, encoded as a different
+  // type.
   //
-  // Parametrization: TFT_ARRAY[<element type>].
-  // <element_type> may be a concrete type or a type symbol. It represents the
-  //   data type of the elements produced by the dataset.
+  // Parametrization:
+  //   TFT_ENCODED[<encoded type>, <encoding type>]
+  //   * <encoded type> may be any type
+  //   * <encoding type> may be any type
   //
   // Examples:
-  //   TFT_DATSET[TFT_TENSOR[TFT_INT32]] is a Dataset producing single int32
-  //     Tensors of unknown shape.
-  //   TFT_DATSET[TFT_PRODUCT[TFT_TENSOR[TFT_INT32], TFT_TENSOR[TFT_FLOAT32]] is
-  //   a
-  //     Dataset producing pairs of Tensors, one integer and one float.
-  // Note: The high ID number is to prepare for the eventuality that Datasets
-  // will be supported by user types in the future.
-  TFT_DATASET = 10102;
+  //   TFT_ENCODING[TFT_INT32, TFT_STRING] is an integer encoded as string.
+  TFT_ENCODED = 1004;
+
+  // The type of "shape tensors" where the runtime value is the shape of
+  // some tensor(s), i.e. the output of tf.shape.
+  // Shape tensors have special, host-only placement, in contrast to
+  // TFT_TENSOR[TFT_INT32] which is the type of a normal numeric tensor
+  // with no special placement.
+  //
+  // Examples:
+  //   TFT_SHAPE_TENSOR[TFT_INT32] is the most common
+  //   TFT_SHAPE_TENSOR[TFT_INT64] is also allowed
+  TFT_SHAPE_TENSOR = 1005;
 
   // Type attributes. These always appear in the parametrization of a type,
   // never alone. For example, there is no such thing as a "bool" TensorFlow
@@ -172,6 +224,65 @@ enum FullTypeId {
   TFT_COMPLEX128 = 213;
   // The string element type.
   TFT_STRING = 214;
+
+  // Other types that we don't know yet whether they will become part of the
+  // core type system or be consisdered third-party (and consequently moved to
+  // user-defined type mechanisms). Presently, they are effectively in the core
+  // type system, because key compilation passes like Placer account for their
+  // existence.
+
+  // Datasets created by tf.data ops and APIs. Datasets have generator/iterable
+  // semantics, that is, one can construct an iterator from them. Like
+  // Array, they are considered to return elements that can be described
+  // by a single type. Unlike Array, they do not support random access or
+  // mutation, and can potentially produce an infinite number of elements.
+  // A datasets can produce logical structures (e.g. multiple elements). This
+  // is expressed using TFT_PRODUCT.
+  //
+  //
+  // Parametrization: TFT_DATASET[<element type>].
+  //   * <element type> may be a concrete type or a type symbol. It represents
+  //     the data type of the elements produced by the dataset.
+  //
+  // Examples:
+  //   TFT_DATSET[TFT_TENSOR[TFT_INT32]] is a Dataset producing single int32
+  //     Tensors of unknown shape.
+  //   TFT_DATSET[TFT_PRODUCT[TFT_TENSOR[TFT_INT32], TFT_TENSOR[TFT_FLOAT32]] is
+  //     a Dataset producing pairs of Tensors, one integer and one float.
+  // Note: The high ID number is to prepare for the eventuality that Datasets
+  // will be supported by user types in the future.
+  TFT_DATASET = 10102;
+
+  // A ragged tensor created by tf.ragged ops and APIs.
+  //
+  // Parametrization: TFT_RAGGED[<element_type>].
+  TFT_RAGGED = 10103;
+
+  // Iterators created by tf.data ops and APIs. Very similar to Datasets, except
+  // they are mutable.
+  //
+  //
+  // Parametrization: TFT_ITERATOR[<element type>].
+  //   * <element type> may be a concrete type or a type symbol. It represents
+  //     the data type of the elements produced by the dataset.
+  TFT_ITERATOR = 10104;
+
+  // A mutex lock tensor, produced by tf.raw_ops.MutexLock.
+  // Unlike strict execution models, where ownership of a lock is denoted by
+  // "running after the lock has been acquired", in non-strict mode, lock
+  // ownership is in the true sense: "the op argument representing the lock is
+  // available".
+  // Mutex locks are the dynamic counterpart of control dependencies.
+  // TODO(mdan): Properly document this thing.
+  //
+  // Parametrization: TFT_MUTEX_LOCK[].
+  TFT_MUTEX_LOCK = 10202;
+
+  // The equivalent of a Tensor with DT_VARIANT dtype, kept here to simplify
+  // translation. This type should not normally appear after type inference.
+  // Note that LEGACY_VARIANT != ANY: TENSOR[INT32] is a subtype of ANY, but is
+  // not a subtype of LEGACY_VARIANT.
+  TFT_LEGACY_VARIANT = 10203;
 }
 
 // Highly experimental and very likely to change.
@@ -186,11 +297,14 @@ message FullTypeDef {
 
   repeated FullTypeDef args = 2;
 
-  // Literal values of this type object, if the the type admits one.
+  // Literal values of this type object, if the type admits one.
   // For example, a type variable admits a string attribute - its name.
   // Shape-related types may admit int attributes - their static shape values.
   // Fields for more data types to be added as needed.
   oneof attr {
     string s = 3;
+    int64 i = 4;
+    // TODO(mdan): list/tensor, map? Need to reconcile with TFT_RECORD, etc.
   }
 }
+// LINT.ThenChange(../ir/types/attributes.td)
diff --git a/proto/tensorflow/core/framework/graph.proto b/proto/tensorflow/core/framework/graph.proto
index 73d6453..ef16dec 100644
--- a/proto/tensorflow/core/framework/graph.proto
+++ b/proto/tensorflow/core/framework/graph.proto
@@ -3,6 +3,7 @@ syntax = "proto3";
 package tensorflow;
 
 import "tensorflow/core/framework/function.proto";
+import "tensorflow/core/framework/graph_debug_info.proto";
 import "tensorflow/core/framework/node_def.proto";
 import "tensorflow/core/framework/versions.proto";
 
@@ -53,4 +54,7 @@ message GraphDef {
   //     consumer does not start until all return values of the callee
   //     function are ready.
   FunctionDefLibrary library = 2;
+
+  // Stack traces for the nodes in this graph.
+  GraphDebugInfo debug_info = 5;
 }
diff --git a/proto/tensorflow/core/framework/graph_debug_info.proto b/proto/tensorflow/core/framework/graph_debug_info.proto
new file mode 100644
index 0000000..af90b5f
--- /dev/null
+++ b/proto/tensorflow/core/framework/graph_debug_info.proto
@@ -0,0 +1,61 @@
+syntax = "proto2";
+
+package tensorflow;
+
+option cc_enable_arenas = true;
+option java_outer_classname = "GraphDebugInfoProtos";
+option java_multiple_files = true;
+option java_package = "org.tensorflow.framework";
+option go_package = "github.com/tensorflow/tensorflow/tensorflow/go/core/protobuf";
+
+message GraphDebugInfo {
+  // This represents a file/line location in the source code.
+  message FileLineCol {
+    // File name index, which can be used to retrieve the file name string from
+    // `files`. The value should be between 0 and (len(files)-1)
+    optional int32 file_index = 1;
+
+    // Line number in the file.
+    optional int32 line = 2;
+
+    // Col number in the file line.
+    optional int32 col = 3;
+
+    // Name of function contains the file line.
+    optional string func = 4;
+
+    // Source code contained in this file line.
+    optional string code = 5;
+  }
+
+  // This represents a stack trace which is a ordered list of `FileLineCol`.
+  message StackTrace {
+    repeated FileLineCol file_line_cols = 1;  // Deprecated.
+    repeated fixed64 frame_id = 2 [packed = true];
+  }
+
+  // This stores all the source code file names and can be indexed by the
+  // `file_index`.
+  repeated string files = 1;
+
+  // Stack traces and frames are uniqueified during construction. These maps
+  // index from the unique id for a frame/trace to the value.
+  map<fixed64, FileLineCol> frames_by_id = 4;
+  map<fixed64, StackTrace> traces_by_id = 6;
+
+  map<string, StackTrace> traces = 2;  // Deprecated.
+
+  // This maps a node name to a trace id contained in `traces_by_id`.
+  //
+  // The map key is a mangling of the containing function and op name with
+  // syntax:
+  //   op.name '@' func_name
+  // For ops in the top-level graph, the func_name is the empty string and hence
+  // the `@` may be ommitted.
+  // Note that op names are restricted to a small number of characters which
+  // exclude '@', making it impossible to collide keys of this form. Function
+  // names accept a much wider set of characters.
+  // It would be preferable to avoid mangling and use a tuple key of (op.name,
+  // func_name), but this is not supported with protocol buffers.
+  map<string, fixed64> name_to_trace_id = 5;
+}
diff --git a/proto/tensorflow/core/framework/node_def.proto b/proto/tensorflow/core/framework/node_def.proto
index a1df938..daf4c78 100644
--- a/proto/tensorflow/core/framework/node_def.proto
+++ b/proto/tensorflow/core/framework/node_def.proto
@@ -3,6 +3,7 @@ syntax = "proto3";
 package tensorflow;
 
 import "tensorflow/core/framework/attr_value.proto";
+import "tensorflow/core/framework/full_type.proto";
 
 option cc_enable_arenas = true;
 option java_outer_classname = "NodeProto";
@@ -85,4 +86,10 @@ message NodeDef {
 
   // This stores debug information associated with the node.
   ExperimentalDebugInfo experimental_debug_info = 6;
+
+  // The complete type of this node. Experimental and subject to change.
+  // Currently, the field only contains the return types of the node. That will
+  // extend in the future to contain the entire signature of the node, as a
+  // function type.
+  FullTypeDef experimental_type = 7;
 }
diff --git a/proto/tensorflow/core/framework/resource_handle.proto b/proto/tensorflow/core/framework/resource_handle.proto
index 3f84121..2b6276f 100644
--- a/proto/tensorflow/core/framework/resource_handle.proto
+++ b/proto/tensorflow/core/framework/resource_handle.proto
@@ -34,7 +34,9 @@ message ResourceHandleProto {
 
   // Protocol buffer representing a pair of (data type, tensor shape).
   message DtypeAndShape {
+    // Data type of the tensor.
     DataType dtype = 1;
+    // Shape of the tensor.
     TensorShapeProto shape = 2;
   }
 
diff --git a/proto/tensorflow/core/framework/tensor.proto b/proto/tensorflow/core/framework/tensor.proto
index 626af7b..b092b7e 100644
--- a/proto/tensorflow/core/framework/tensor.proto
+++ b/proto/tensorflow/core/framework/tensor.proto
@@ -14,6 +14,7 @@ option go_package = "github.com/tensorflow/tensorflow/tensorflow/go/core/framewo
 
 // Protocol buffer representing a tensor.
 message TensorProto {
+  // Data type of the tensor.
   DataType dtype = 1;
 
   // Shape of the tensor.  TODO(touts): sort out the 0-rank issues.
@@ -83,6 +84,10 @@ message TensorProto {
 
   // DT_UINT64
   repeated uint64 uint64_val = 17 [packed = true];
+
+  // DT_FLOAT8_*, use variable-sized set of bytes
+  // (i.e. the equivalent of repeated uint8, if such a thing existed).
+  bytes float8_val = 18;
 }
 
 // Protocol buffer representing the serialization format of DT_VARIANT tensors.
diff --git a/proto/tensorflow/core/framework/types.proto b/proto/tensorflow/core/framework/types.proto
index 5a77eda..30294ba 100644
--- a/proto/tensorflow/core/framework/types.proto
+++ b/proto/tensorflow/core/framework/types.proto
@@ -30,7 +30,7 @@ enum DataType {
   DT_QINT8 = 11;     // Quantized int8
   DT_QUINT8 = 12;    // Quantized uint8
   DT_QINT32 = 13;    // Quantized int32
-  DT_BFLOAT16 = 14;  // Float32 truncated to 16 bits.  Only for cast ops.
+  DT_BFLOAT16 = 14;  // Float32 truncated to 16 bits.
   DT_QINT16 = 15;    // Quantized int16
   DT_QUINT16 = 16;   // Quantized uint16
   DT_UINT16 = 17;
@@ -40,9 +40,19 @@ enum DataType {
   DT_VARIANT = 21;  // Arbitrary C++ data types
   DT_UINT32 = 22;
   DT_UINT64 = 23;
+  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_INT4 = 29;
+  DT_UINT4 = 30;
 
-  // Do not use!  These are only for parameters.  Every enum above
-  // should have a corresponding value below (verified by types_test).
+  // Do not use!  These are only for TF1's obsolete reference Variables.
+  // Every enum above should have a corresponding value below (verified by
+  // types_test).
   DT_FLOAT_REF = 101;
   DT_DOUBLE_REF = 102;
   DT_INT32_REF = 103;
@@ -66,6 +76,14 @@ enum DataType {
   DT_VARIANT_REF = 121;
   DT_UINT32_REF = 122;
   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_INT4_REF = 129;
+  DT_UINT4_REF = 130;
 }
 // LINT.ThenChange(
 //    https://www.tensorflow.org/code/tensorflow/c/tf_datatype.h,
@@ -75,3 +93,8 @@ enum DataType {
 //    https://www.tensorflow.org/code/tensorflow/core/framework/types.cc,
 //    https://www.tensorflow.org/code/tensorflow/python/framework/dtypes.py,
 //    https://www.tensorflow.org/code/tensorflow/python/framework/function.py)
+
+// Represents a serialized tf.dtypes.Dtype
+message SerializedDType {
+  DataType datatype = 1;
+}
diff --git a/proto/tensorflow/core/protobuf/cluster.proto b/proto/tensorflow/core/protobuf/cluster.proto
index a1d4749..cc86353 100644
--- a/proto/tensorflow/core/protobuf/cluster.proto
+++ b/proto/tensorflow/core/protobuf/cluster.proto
@@ -74,6 +74,9 @@ message JobDef {
   // If the `name` field contains "worker", and the `tasks` map contains a
   // mapping from 7 to "example.org:2222", then the device prefix
   // "/job:worker/task:7" will be assigned to "example.org:2222".
+  //
+  // If a job has multiple replicas, host-ports will be comma-delimited, with
+  // one entry for each replica.
   map<int32, string> tasks = 2;
 }
 
diff --git a/proto/tensorflow/core/protobuf/config.proto b/proto/tensorflow/core/protobuf/config.proto
index 7fd6ce7..83e9852 100644
--- a/proto/tensorflow/core/protobuf/config.proto
+++ b/proto/tensorflow/core/protobuf/config.proto
@@ -8,6 +8,8 @@ import "tensorflow/core/framework/step_stats.proto";
 import "tensorflow/core/protobuf/cluster.proto";
 import "tensorflow/core/protobuf/debug.proto";
 import "tensorflow/core/protobuf/rewriter_config.proto";
+import "tensorflow/core/protobuf/rpc_options.proto";
+import "tsl/protobuf/coordination_config.proto";
 
 option cc_enable_arenas = true;
 option java_outer_classname = "ConfigProtos";
@@ -16,9 +18,9 @@ option java_package = "org.tensorflow.framework";
 option go_package = "github.com/tensorflow/tensorflow/tensorflow/go/core/protobuf";
 
 message GPUOptions {
-  // Fraction of the available GPU memory to allocate for each process.
+  // Fraction of the total GPU memory to allocate for each process.
   // 1 means to allocate all of the GPU memory, 0.5 means the process
-  // allocates up to ~50% of the available GPU memory.
+  // allocates up to ~50% of the total GPU memory.
   //
   // GPU memory is pre-allocated unless the allow_growth option is enabled.
   //
@@ -104,8 +106,8 @@ message GPUOptions {
       // Per "virtual" device memory limit, in MB. The number of elements in
       // the list is the number of virtual devices to create on the
       // corresponding visible GPU (see "virtual_devices" below).
-      // If empty, it will create single virtual device taking all available
-      // memory from the device.
+      // If empty and `num_virtual_devices_per_gpu` is not set, it will create
+      // single virtual device taking all available memory from the device.
       //
       // For the concept of "visible" and "virtual" GPU, see the comments for
       // "visible_device_list" above for more information.
@@ -122,6 +124,12 @@ message GPUOptions {
       // created with the default. If this field has values set, then the size
       // of this must match with the above memory_limit_mb.
       repeated int32 priority = 2;
+
+      // Virtual Device ordinal number determines the device ID of the device.
+      // A Virtual device with a lower ordinal number always receives the a
+      // smaller device id. The phyiscal device id and location in the
+      // virtual device list is used to break ties.
+      repeated int32 device_ordinal = 3;
     }
 
     // The multi virtual device settings. If empty (not set), it will create
@@ -131,15 +139,30 @@ message GPUOptions {
     // "visible_device_list" filtering if it is set), and the string represented
     // device names (e.g. /device:GPU:<id>) will refer to the virtual
     // devices and have the <id> field assigned sequentially starting from 0,
-    // according to the order they appear in this list and the "memory_limit"
-    // list inside each element. For example,
+    // according to the order of the virtual devices determined by
+    // device_ordinal and the location in the virtual device list.
+    //
+    // For example,
     //   visible_device_list = "1,0"
     //   virtual_devices { memory_limit: 1GB memory_limit: 2GB }
-    //   virtual_devices {}
-    // will create three virtual devices as:
+    //   virtual_devices { memory_limit: 3GB memory_limit: 4GB }
+    // will create 4 virtual devices as:
     //   /device:GPU:0 -> visible GPU 1 with 1GB memory
     //   /device:GPU:1 -> visible GPU 1 with 2GB memory
-    //   /device:GPU:2 -> visible GPU 0 with all available memory
+    //   /device:GPU:2 -> visible GPU 0 with 3GB memory
+    //   /device:GPU:3 -> visible GPU 0 with 4GB memory
+    //
+    // but
+    //   visible_device_list = "1,0"
+    //   virtual_devices { memory_limit: 1GB memory_limit: 2GB
+    //                     device_ordinal: 10 device_ordinal: 20}
+    //   virtual_devices { memory_limit: 3GB memory_limit: 4GB
+    //                     device_ordinal: 10 device_ordinal: 20}
+    // will create 4 virtual devices as:
+    //   /device:GPU:0 -> visible GPU 1 with 1GB memory  (ordinal 10)
+    //   /device:GPU:1 -> visible GPU 0 with 3GB memory  (ordinal 10)
+    //   /device:GPU:2 -> visible GPU 1 with 2GB memory  (ordinal 20)
+    //   /device:GPU:3 -> visible GPU 0 with 4GB memory  (ordinal 20)
     //
     // NOTE:
     // 1. It's invalid to set both this and "per_process_gpu_memory_fraction"
@@ -149,6 +172,12 @@ message GPUOptions {
     //    result in undefined behavior.
     repeated VirtualDevices virtual_devices = 1;
 
+    // The number of virtual devices to create on each visible GPU. The
+    // available memory will be split equally among all virtual devices. If the
+    // field `memory_limit_mb` in `VirtualDevices` is not empty, this field will
+    // be ignored.
+    int32 num_virtual_devices_per_gpu = 15;
+
     // If true, uses CUDA unified memory for memory allocations. If
     // per_process_gpu_memory_fraction option is greater than 1.0, then unified
     // memory is used regardless of the value for this field. See comments for
@@ -211,6 +240,68 @@ message GPUOptions {
 
     // When true, use CUDA cudaMallocAsync API instead of TF gpu allocator.
     bool use_cuda_malloc_async = 11;
+
+    // By default, BFCAllocator may sleep when it runs out of memory, in the
+    // hopes that another thread will free up memory in the meantime.  Setting
+    // this to true disables the sleep; instead we'll OOM immediately.
+    bool disallow_retry_on_allocation_failure = 12;
+
+    // Memory limit for "GPU host allocator", aka pinned memory allocator.  This
+    // can also be set via the envvar TF_GPU_HOST_MEM_LIMIT_IN_MB.
+    float gpu_host_mem_limit_in_mb = 13;
+
+    // If true, then the host allocator allocates its max memory all upfront and
+    // never grows.  This can be useful for latency-sensitive systems, because
+    // growing the GPU host memory pool can be expensive.
+    //
+    // You probably only want to use this in combination with
+    // gpu_host_mem_limit_in_mb, because the default GPU host memory limit is
+    // quite high.
+    bool gpu_host_mem_disallow_growth = 14;
+
+    // Memory limit for gpu system. This can also be set by
+    // TF_DEVICE_MIN_SYS_MEMORY_IN_MB, which takes precedence over
+    // gpu_system_memory_size_in_mb. With this, user can configure the gpu
+    // system memory size for better resource estimation of multi-tenancy(one
+    // gpu with multiple model) use case.
+    int32 gpu_system_memory_size_in_mb = 16;
+
+    // If true, save information needed for created a PjRt GPU client for
+    // creating a client with remote devices.
+    bool populate_pjrt_gpu_client_creation_info = 17;
+
+    // node_id for use when creating a PjRt GPU client with remote devices,
+    // which enumerates jobs*tasks from a ServerDef.
+    int32 node_id = 18;
+
+    // Whether to merge data transfer streams into the compute stream in the
+    // same stream group. Stream merging helps reduce the overhead caused by
+    // stream synchronization, especially when data transfers are frequent. For
+    // example, setting "merge_host_to_device_stream = true" will make the
+    // compute stream responsible for both computation and host to device memory
+    // copy.
+    message StreamMergeOptions {
+      // If true, the compute stream will be used for host_to_device copy as
+      // well. It's no longer necessary to record an event before the copy to
+      // let the copy stream wait for the compute stream to finish. There is
+      // also no need to wait for the copy to complete before executing the
+      // callback function.
+      bool merge_host_to_device_stream = 1;
+
+      // If true, the compute stream will be used for device_to_host copy as
+      // well. It's no longer necessary to record an event before the copy to
+      // let the copy stream wait for the compute stream to finish.
+      bool merge_device_to_host_stream = 2;
+
+      // If true, the compute stream will be used for device_to_device copy as
+      // well. It's no longer necessary to record an event before the copy to
+      // let the copy stream wait for the compute stream of the sending device
+      // to finish. There is also no need to wait for the compute stream of the
+      // receiving device to finish if the copy is within the same device.
+      bool merge_device_to_device_stream = 3;
+    }
+
+    StreamMergeOptions stream_merge_options = 19;
   }
 
   // Everything inside experimental is subject to change and is not subject
@@ -270,6 +361,11 @@ message OptimizerOptions {
     ON_2 = 2;
   }
   GlobalJitLevel global_jit_level = 5;
+
+  // CPU code will be autoclustered only if global_jit_level >= ON_1 and either:
+  //  - this flag is true, or
+  //  - TF_XLA_FLAGS contains --tf_xla_cpu_global_jit=true.
+  bool cpu_global_jit = 7;
 }
 
 message GraphOptions {
@@ -344,41 +440,6 @@ message ThreadPoolOptionProto {
   string global_name = 2;
 }
 
-message RPCOptions {
-  // If true, always use RPC to contact the session target.
-  //
-  // If false (the default option), TensorFlow may use an optimized
-  // transport for client-master communication that avoids the RPC
-  // stack. This option is primarily for used testing the RPC stack.
-  bool use_rpc_for_inprocess_master = 1;
-
-  // The compression algorithm to be used. One of "deflate", "gzip".
-  string compression_algorithm = 2;
-
-  // If compression_algorithm is set, the compression level to be used.
-  // From 0 (no compression), up to 3.
-  int32 compression_level = 3;
-
-  // Setting cache_rpc_response to true will enable sender side caching of
-  // response for RecvTensorAsync and RecvBufAsync to allow receiver to retry
-  // requests . This is only necessary when the network fabric is experiencing a
-  // significant error rate.  Without it we'll fail a step on an network error,
-  // while with it we'll be able to complete long steps (like complex
-  // initializations) in the face of some network errors during RecvTensor.
-  bool cache_rpc_response = 4;
-
-  // Disables TCP connection sharing when opening a new RPC channel.
-  bool disable_session_connection_sharing = 5;
-
-  // Setting num_channels_per_target > 0 allows uses of multiple channels to
-  // communicate to the same target. This can be used to improve the aggregate
-  // throughput on high speed links (e.g 100G) where single connection is not
-  // sufficient to maximize link utilization. Note that a single RPC only goes
-  // on a single channel, this only helps in situations where there are multiple
-  // transfers to the same target overlapping in time.
-  int32 num_channels_per_target = 6;
-}
-
 // Metadata about the session.
 //
 // This can be used by the runtime and the Ops for debugging, monitoring, etc.
@@ -475,6 +536,9 @@ message ConfigProto {
   // Options that apply to all GPUs.
   GPUOptions gpu_options = 6;
 
+  // Options that apply to pluggable devices.
+  GPUOptions pluggable_device_options = 18;
+
   // Whether soft placement is allowed. If allow_soft_placement is true,
   // an op will be placed on CPU if
   //   1. there's no GPU implementation for the OP
@@ -584,7 +648,8 @@ message ConfigProto {
     // If set, this can be used by the runtime and the Ops for debugging,
     // monitoring, etc.
     //
-    // NOTE: This is currently used and propagated only by the direct session.
+    // NOTE: This is currently used and propagated only by the direct session
+    // and EagerContext.
     SessionMetadata session_metadata = 11;
 
     // If true, the session may treat the graph as being static for optimization
@@ -595,13 +660,10 @@ message ConfigProto {
     // Session::Extend() may not be supported.
     bool optimize_for_static_graph = 12;
 
-    // This field will eventually be deprecated and replaced by
-    // mlir_bridge_rollout (b/166038521).
-    //
-    // Whether to enable the MLIR-based TF->XLA bridge.
+    // Whether to enable the MLIR-based TF->XLA bridge. This is only used if set
+    // to true. Default value or false is ignored. Use mlir_bridge_rollout for
+    // finer control.
     //
-    // This is a replacement to the existing bridge, and not ready for
-    // production usage yet.
     // If this option is set to true when a session is created, MLIR is used to
     // perform the set of graph transformations to put the graph in a form that
     // can be executed with delegation of some computations to an accelerator.
@@ -620,22 +682,10 @@ message ConfigProto {
       MLIR_BRIDGE_ROLLOUT_ENABLED = 1;
       // Disabling the MLIR bridge disables it for all graphs in this session.
       MLIR_BRIDGE_ROLLOUT_DISABLED = 2;
-      // Enable the MLIR bridge on a per graph basis based on an analysis of
-      // the features used in the graph. If the features used by the graph are
-      // supported by the MLIR bridge, the MLIR bridge will be used to run the
-      // graph.
-      MLIR_BRIDGE_ROLLOUT_SAFE_MODE_ENABLED = 3;
-      // Enable the MLIR bridge in a fallback mode on a per graph basis based
-      // on an analysis of the features used in the graph.
-      // Running the MLIR bridge in the fallback mode means that it is
-      // executed and it commits all the changes to the TF graph in case
-      // of success. And it does not in case of failures and let the old bridge
-      // to process the TF graph.
-      MLIR_BRIDGE_ROLLOUT_SAFE_MODE_FALLBACK_ENABLED = 4;
+      reserved 3, 4;
+      reserved "MLIR_BRIDGE_ROLLOUT_SAFE_MODE_ENABLED",
+          "MLIR_BRIDGE_ROLLOUT_SAFE_MODE_FALLBACK_ENABLED";
     }
-    // This field is underdevelopment, for now use enable_mlir_bridge
-    // (b/166038521).
-    //
     // Whether to enable the MLIR-based TF->XLA bridge.
     MlirBridgeRollout mlir_bridge_rollout = 17;
 
@@ -663,22 +713,73 @@ message ConfigProto {
     // Whether runtime execution uses TFRT.
     bool use_tfrt = 18;
 
-    // Distributed coordination service to be enabled if set.
-    // Currently only effective in multi-client setup.
-    string coordination_service = 19;
+    // If true, use Pathways with TFRT API for multi host support.
+    bool enable_multi_host = 27;
+
+    // If true, use ifrt as the backend for TFRT. This is only used when
+    // `use_tfrt` is true.
+    bool tfrt_use_ifrt = 32;
+
+    // Port for the Pathways server. Ignored if enable_multi_host=false.
+    int32 backend_server_port = 28;
+
+    // If true, TFRT will use TPU specific compiler passes and perform TPU
+    // specific initialization.
+    bool target_tpu = 29;
+
+    // If true, TFRT will use GPU specific compiler passes and perform GPU
+    // specific initialization.
+    bool target_gpu = 30;
+
+    // The threshold to merge small streams in TFRT. The stream with cost
+    // smaller than the threshold will be merged. Setting it to value 1
+    // disables all merges.
+    int32 stream_merge_threshold = 31;
+
+    // The field "coordination_service was previously specified as a string;
+    // this has been replaced with a message below.
+    reserved 19;
+
+    // We removed the flag fetch_remote_devices_in_multi_client. Marking the tag
+    // number as reserved.
+    reserved 20;
+
+    // Whether functional control flow op lowering should be disabled. This is
+    // useful when executing within a portable runtime where control flow op
+    // kernels may not be loaded due to selective registration.
+    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.
+    bool xla_prefer_single_graph_cluster = 22;
 
-    // Whether the remote devices in the cluster should be fetched during setup
-    // of multi-client cluster. If enabled, the workers will run an extra device
-    // information exchange step during startup and the workers' EagerContexts
-    // will become aware of remote devices in the cluster as well.
-    bool fetch_remote_devices_in_multi_client = 20;
+    // Distributed coordination service configurations.
+    CoordinationServiceConfig coordination_config = 23;
 
-    // Next: 21
+    // If true, the session will treat the graph as being non-static for
+    // optimization purposes.
+    //
+    // If this option is set to true when a session is created, the full
+    // GraphDef will be retained to enable calls to Session::Extend().
+    // Calling Extend() without setting this flag will result in errors.
+    //
+    // This option is meant to replace `optimize_for_static_graph` and it
+    // aims to negate its value.
+    bool disable_optimize_for_static_graph = 24;
+
+    // Whether eager remote execution will stream all the function calls or
+    // allow them to happen in parallel. When true, streaming execution is
+    // disabled, and parallel execution is allowed.
+    bool disable_eager_executor_streaming_enqueue = 26;
+
+    reserved 25;
+
+    // Next: 33
   }
 
   Experimental experimental = 16;
 
-  // Next: 18
+  // Next: 19
 }
 
 // Options for a single Run() call.
@@ -777,6 +878,9 @@ message RunMetadata {
   // level idea of what the built graph looks like (since the various graph
   // optimization passes might change the structure of the graph significantly).
   repeated FunctionGraphs function_graphs = 4;
+
+  // Metadata about the session.
+  SessionMetadata session_metadata = 5;
 }
 
 // Defines a connection between two tensors in a `GraphDef`.
diff --git a/proto/tensorflow/core/protobuf/error_codes.proto b/proto/tensorflow/core/protobuf/error_codes.proto
index 7058807..0e1b8d4 100644
--- a/proto/tensorflow/core/protobuf/error_codes.proto
+++ b/proto/tensorflow/core/protobuf/error_codes.proto
@@ -1,152 +1,11 @@
 syntax = "proto3";
 
-package tensorflow.error;
+// Add a dummy package name. Having no package, like
+// core/lib/core/error_codes.proto, or having tensorflow.error, like
+// tsl/protobuf/error_codes.proto, results in name collision errors in generated
+// code for some users that use JS through J2CL.
+package tensorflow.error.dummy;
 
-option cc_enable_arenas = true;
-option java_outer_classname = "ErrorCodesProtos";
-option java_multiple_files = true;
-option java_package = "org.tensorflow.framework";
-option go_package = "github.com/tensorflow/tensorflow/tensorflow/go/core/protobuf";
-
-// The canonical error codes for TensorFlow APIs.
-//
-// Warnings:
-//
-// -   Do not change any numeric assignments.
-// -   Changes to this list should only be made if there is a compelling
-//     need that can't be satisfied in another way.  Such changes
-//     must be approved by at least two OWNERS.
-// -   These error codes must match gRPC and protobuf error codes (except for
-//     DO_NOT_USE_RESERVED_FOR_FUTURE_EXPANSION_USE_DEFAULT_IN_SWITCH_INSTEAD_).
-//
-// Sometimes multiple error codes may apply.  Services should return
-// the most specific error code that applies.  For example, prefer
-// OUT_OF_RANGE over FAILED_PRECONDITION if both codes apply.
-// Similarly prefer NOT_FOUND or ALREADY_EXISTS over FAILED_PRECONDITION.
-enum Code {
-  // Not an error; returned on success
-  OK = 0;
-
-  // The operation was cancelled (typically by the caller).
-  CANCELLED = 1;
-
-  // Unknown error.  An example of where this error may be returned is
-  // if a Status value received from another address space belongs to
-  // an error-space that is not known in this address space.  Also
-  // errors raised by APIs that do not return enough error information
-  // may be converted to this error.
-  UNKNOWN = 2;
-
-  // Client specified an invalid argument.  Note that this differs
-  // from FAILED_PRECONDITION.  INVALID_ARGUMENT indicates arguments
-  // that are problematic regardless of the state of the system
-  // (e.g., a malformed file name).
-  INVALID_ARGUMENT = 3;
-
-  // Deadline expired before operation could complete.  For operations
-  // that change the state of the system, this error may be returned
-  // even if the operation has completed successfully.  For example, a
-  // successful response from a server could have been delayed long
-  // enough for the deadline to expire.
-  DEADLINE_EXCEEDED = 4;
-
-  // Some requested entity (e.g., file or directory) was not found.
-  // For privacy reasons, this code *may* be returned when the client
-  // does not have the access right to the entity.
-  NOT_FOUND = 5;
-
-  // Some entity that we attempted to create (e.g., file or directory)
-  // already exists.
-  ALREADY_EXISTS = 6;
-
-  // The caller does not have permission to execute the specified
-  // operation.  PERMISSION_DENIED must not be used for rejections
-  // caused by exhausting some resource (use RESOURCE_EXHAUSTED
-  // instead for those errors).  PERMISSION_DENIED must not be
-  // used if the caller can not be identified (use UNAUTHENTICATED
-  // instead for those errors).
-  PERMISSION_DENIED = 7;
-
-  // The request does not have valid authentication credentials for the
-  // operation.
-  UNAUTHENTICATED = 16;
+import public "tsl/protobuf/error_codes.proto";
 
-  // Some resource has been exhausted, perhaps a per-user quota, or
-  // perhaps the entire file system is out of space.
-  RESOURCE_EXHAUSTED = 8;
-
-  // Operation was rejected because the system is not in a state
-  // required for the operation's execution.  For example, directory
-  // to be deleted may be non-empty, an rmdir operation is applied to
-  // a non-directory, etc.
-  //
-  // A litmus test that may help a service implementor in deciding
-  // between FAILED_PRECONDITION, ABORTED, and UNAVAILABLE:
-  //  (a) Use UNAVAILABLE if the client can retry just the failing call.
-  //  (b) Use ABORTED if the client should retry at a higher-level
-  //      (e.g., restarting a read-modify-write sequence).
-  //  (c) Use FAILED_PRECONDITION if the client should not retry until
-  //      the system state has been explicitly fixed.  E.g., if an "rmdir"
-  //      fails because the directory is non-empty, FAILED_PRECONDITION
-  //      should be returned since the client should not retry unless
-  //      they have first fixed up the directory by deleting files from it.
-  //  (d) Use FAILED_PRECONDITION if the client performs conditional
-  //      REST Get/Update/Delete on a resource and the resource on the
-  //      server does not match the condition. E.g., conflicting
-  //      read-modify-write on the same resource.
-  FAILED_PRECONDITION = 9;
-
-  // The operation was aborted, typically due to a concurrency issue
-  // like sequencer check failures, transaction aborts, etc.
-  //
-  // See litmus test above for deciding between FAILED_PRECONDITION,
-  // ABORTED, and UNAVAILABLE.
-  ABORTED = 10;
-
-  // Operation tried to iterate past the valid input range.  E.g., seeking or
-  // reading past end of file.
-  //
-  // Unlike INVALID_ARGUMENT, this error indicates a problem that may
-  // be fixed if the system state changes. For example, a 32-bit file
-  // system will generate INVALID_ARGUMENT if asked to read at an
-  // offset that is not in the range [0,2^32-1], but it will generate
-  // OUT_OF_RANGE if asked to read from an offset past the current
-  // file size.
-  //
-  // There is a fair bit of overlap between FAILED_PRECONDITION and
-  // OUT_OF_RANGE.  We recommend using OUT_OF_RANGE (the more specific
-  // error) when it applies so that callers who are iterating through
-  // a space can easily look for an OUT_OF_RANGE error to detect when
-  // they are done.
-  OUT_OF_RANGE = 11;
-
-  // Operation is not implemented or not supported/enabled in this service.
-  UNIMPLEMENTED = 12;
-
-  // Internal errors.  Means some invariant expected by the underlying
-  // system has been broken.  If you see one of these errors,
-  // something is very broken.
-  INTERNAL = 13;
-
-  // The service is currently unavailable.  This is a most likely a
-  // transient condition and may be corrected by retrying with
-  // a backoff.
-  //
-  // See litmus test above for deciding between FAILED_PRECONDITION,
-  // ABORTED, and UNAVAILABLE.
-  UNAVAILABLE = 14;
-
-  // Unrecoverable data loss or corruption.
-  DATA_LOSS = 15;
-
-  // An extra enum entry to prevent people from writing code that
-  // fails to compile when a new code is added.
-  //
-  // Nobody should ever reference this enumeration entry. In particular,
-  // if you write C++ code that switches on this enumeration, add a default:
-  // case instead of a case that mentions this enumeration entry.
-  //
-  // Nobody should rely on the value (currently 20) listed here.  It
-  // may change in the future.
-  DO_NOT_USE_RESERVED_FOR_FUTURE_EXPANSION_USE_DEFAULT_IN_SWITCH_INSTEAD_ = 20;
-}
+option go_package = "github.com/tensorflow/tensorflow/tensorflow/go/core/protobuf";
diff --git a/proto/tensorflow/core/protobuf/meta_graph.proto b/proto/tensorflow/core/protobuf/meta_graph.proto
index 2554bee..6df9969 100644
--- a/proto/tensorflow/core/protobuf/meta_graph.proto
+++ b/proto/tensorflow/core/protobuf/meta_graph.proto
@@ -5,6 +5,7 @@ package tensorflow;
 import "google/protobuf/any.proto";
 import "tensorflow/core/framework/graph.proto";
 import "tensorflow/core/framework/op_def.proto";
+import "tensorflow/core/framework/tensor.proto";
 import "tensorflow/core/framework/tensor_shape.proto";
 import "tensorflow/core/framework/types.proto";
 import "tensorflow/core/protobuf/saved_object_graph.proto";
@@ -17,9 +18,6 @@ option java_multiple_files = true;
 option java_package = "org.tensorflow.framework";
 option go_package = "github.com/tensorflow/tensorflow/tensorflow/go/core/protobuf";
 
-// NOTE: This protocol buffer is evolving, and will go through revisions in the
-// coming months.
-//
 // Protocol buffer containing the following which are necessary to restart
 // training, run inference. It can be used to serialize/de-serialize memory
 // objects necessary for running computation in a graph when crossing the
@@ -259,75 +257,21 @@ message TensorInfo {
 
 // SignatureDef defines the signature of a computation supported by a TensorFlow
 // graph.
-//
-// For example, a model with two loss computations, sharing a single input,
-// might have the following signature_def map, in a MetaGraphDef message.
-//
-// Note that across the two SignatureDefs "loss_A" and "loss_B", the input key,
-// output key, and method_name are identical, and will be used by system(s) that
-// implement or rely upon this particular loss method. The output tensor names
-// differ, demonstrating how different outputs can exist for the same method.
-//
-// signature_def {
-//   key: "loss_A"
-//   value {
-//     inputs {
-//       key: "input"
-//       value {
-//         name: "input:0"
-//         dtype: DT_STRING
-//         tensor_shape: ...
-//       }
-//     }
-//     outputs {
-//       key: "loss_output"
-//       value {
-//         name: "loss_output_A:0"
-//         dtype: DT_FLOAT
-//         tensor_shape: ...
-//       }
-//     }
-//     method_name: "some/package/compute_loss"
-//   }
-//   ...
-// }
-// signature_def {
-//   key: "loss_B"
-//   value {
-//     inputs {
-//       key: "input"
-//       value {
-//         name: "input:0"
-//         dtype: DT_STRING
-//         tensor_shape: ...
-//       }
-//     }
-//     outputs {
-//       key: "loss_output"
-//       value {
-//         name: "loss_output_B:0"
-//         dtype: DT_FLOAT
-//         tensor_shape: ...
-//       }
-//     }
-//     method_name: "some/package/compute_loss"
-//   }
-//   ...
-// }
 message SignatureDef {
   // Named input parameters.
   map<string, TensorInfo> inputs = 1;
   // Named output parameters.
   map<string, TensorInfo> outputs = 2;
-  // Extensible method_name information enabling third-party users to mark a
-  // SignatureDef as supporting a particular method. This enables producers and
-  // consumers of SignatureDefs, e.g. a model definition library and a serving
-  // library to have a clear hand-off regarding the semantics of a computation.
+  // Deprecated: TensorFlow 2 always sets this to a fixed value;
+  // open-source TF Serving stopped checking by default since release 2.4.
   //
-  // Note that multiple SignatureDefs in a single MetaGraphDef may have the same
-  // method_name. This is commonly used to support multi-headed computation,
-  // where a single graph computation may return multiple results.
+  // In TensorFlow 1, the method_name enabled users to mark a SignatureDef as
+  // supporting a particular method. Multiple SignatureDefs in a single
+  // MetaGraphDef could have the same method_name (e.g., to support multi-headed
+  // computation).
   string method_name = 3;
+  // Named input to corresponding default values if any.
+  map<string, TensorProto> defaults = 4;
 }
 
 // An asset file def for a single file or a set of sharded files with the same
diff --git a/proto/tensorflow/core/protobuf/rewriter_config.proto b/proto/tensorflow/core/protobuf/rewriter_config.proto
index a6bf729..0a2021e 100644
--- a/proto/tensorflow/core/protobuf/rewriter_config.proto
+++ b/proto/tensorflow/core/protobuf/rewriter_config.proto
@@ -37,6 +37,13 @@ message RewriterConfig {
     // may break. For example, assume the shape of a placeholder matches its
     // actual feed.
     AGGRESSIVE = 3;
+    // Run MLIR pass if there's one implemented in TFG, do nothing otherwise.
+    // I.e., if there's no corresponding TFG pass, it's an OFF. This is supposed
+    // to be mapped with `ON` and there's no `AGGRESSIVE` in MLIR pass now.
+    EXPERIMENTAL_MLIR = 4;
+    // Run both MLIR and Grappler passes consecutively and MLIR pass will come
+    // first.
+    EXPERIMENTAL_BOTH = 5;
   }
 
   // Enum for layout conversion between NCHW and NHWC on CPU. Default is OFF.
@@ -95,19 +102,36 @@ message RewriterConfig {
   // Enable the swap of kernel implementations based on the device placement
   // (default is ON).
   Toggle implementation_selector = 22;
-  // Optimize data types for CUDA (default is OFF).
-  // This will try to use float16 on GPU which is faster.
+  // Optimize data types for CUDA/oneDNN (default is OFF).
+  // This will try to use float16 on GPU/CPU which is faster.
   // Note that this can change the numerical stability of the graph and may
   // require the use of loss scaling to maintain model convergence.
   Toggle auto_mixed_precision = 23;
-  // Optimize data types for MKL (default is OFF).
+  // Optimize data types for oneDNN (default is OFF).
   // This will try to use bfloat16 on CPUs, which is faster.
   // Note that this can change the numerical stability of the graph.
+  // Note: this is deprecated.
+  // It is replaced by auto_mixed_precision_onednn_bfloat16
   Toggle auto_mixed_precision_mkl = 25;
+  // Optimize data types for oneDNN (default is OFF).
+  // This will try to use bfloat16 on CPUs, which is faster.
+  // Note that this can change the numerical stability of the graph.
+  // Note: this is equivalent to the deprecated option auto_mixed_precision_mkl
+  Toggle auto_mixed_precision_onednn_bfloat16 = 31;
+  // Emulate a model using data type float16 on CPU (default is OFF).
+  // This will try to emulate the float16 inputs and outputs of an operator
+  // on CPU to have better correlation with float16 on GPU; however the
+  // computation in the operator is based on float32.
+  // Note that this can change the numerical stability of the graph.
+  Toggle auto_mixed_precision_cpu = 29;
   // Disable the entire meta optimizer (off by default).
   bool disable_meta_optimizer = 19;
+  // Disable the TFG optimizer (off by default).
+  bool disable_tfg_optimizer = 32;
   // Optimizers registered by plugin (default is ON)
   Toggle use_plugin_optimizers = 28;
+  // Conditional code motion (default is ON).
+  Toggle experimental_conditional_code_motion = 30;
 
   // Controls how many times we run the optimizers in meta optimizer (default
   // is once).
diff --git a/proto/tensorflow/core/protobuf/rpc_options.proto b/proto/tensorflow/core/protobuf/rpc_options.proto
new file mode 100644
index 0000000..c8716af
--- /dev/null
+++ b/proto/tensorflow/core/protobuf/rpc_options.proto
@@ -0,0 +1,7 @@
+syntax = "proto3";
+
+package tensorflow.dummy;
+
+import public "tsl/protobuf/rpc_options.proto";
+
+option go_package = "github.com/tensorflow/tensorflow/tensorflow/go/core/protobuf";
diff --git a/proto/tensorflow/core/protobuf/saved_object_graph.proto b/proto/tensorflow/core/protobuf/saved_object_graph.proto
index 8e23091..3a00db3 100644
--- a/proto/tensorflow/core/protobuf/saved_object_graph.proto
+++ b/proto/tensorflow/core/protobuf/saved_object_graph.proto
@@ -2,6 +2,7 @@ syntax = "proto3";
 
 package tensorflow;
 
+import "google/protobuf/any.proto";
 import "tensorflow/core/framework/tensor_shape.proto";
 import "tensorflow/core/framework/types.proto";
 import "tensorflow/core/framework/variable.proto";
@@ -37,9 +38,16 @@ message SavedObject {
   // Objects which this object depends on: named edges in the dependency
   // graph.
   //
-  // Note: currently only valid if kind == "user_object" or "resource".
+  // Note: All kinds of SavedObject may have children, except
+  // "constant" and "captured_tensor".
   repeated TrackableObjectGraph.TrackableObject.ObjectReference children = 1;
 
+  // Ordered list of dependencies that must be loaded before this object.
+  // SavedModel loads with the bottom-up approach, by first creating all objects
+  // (in the order defined by the dependencies), then connecting the edges.
+  repeated TrackableObjectGraph.TrackableObject.ObjectReference dependencies =
+      15;
+
   // Removed when forking SavedObject from TrackableObjectGraph.
   reserved "attributes";
   reserved 2;
@@ -63,7 +71,37 @@ message SavedObject {
     CapturedTensor captured_tensor = 12;
   }
 
+  // Stores the functions used to save and restore this object. At most one of
+  // `saveable_objects` or `registered_saver` is defined for each SavedObject.
+  // See the comment below for the difference between SaveableObject and
+  // registered savers.
   map<string, SaveableObject> saveable_objects = 11;
+
+  // The fields below are filled when the user serializes a registered Trackable
+  // class or an object with a registered saver function.
+  //
+  // Registered classes may save additional metadata and supersede the
+  // default loading process where nodes are recreated from the proto.
+  // If the registered class cannot be found, then the object will load as one
+  // one of the default trackable objects: Autotrackable (a class similar to
+  // tf.Module), tf.function, or tf.Variable.
+  //
+  // Unlike SaveableObjects, which store the functions for saving and restoring
+  // from tensors, registered savers allow Trackables to write checkpoint shards
+  // directly (e.g. for performance or coordination reasons).
+  // *All registered savers must be available when loading the SavedModel.*
+
+  // The name of the registered class of the form "{package}.{class_name}".
+  // This field is used to search for the registered class at loading time.
+  string registered_name = 13;
+  // The user-generated proto storing metadata for this object, to be passed to
+  // the registered classes's _deserialize_from_proto method when this object is
+  // loaded from the SavedModel.
+  google.protobuf.Any serialized_user_proto = 14;
+
+  // String name of the registered saver. At most one of `saveable_objects` or
+  // `registered_saver` is defined for each SavedObject.
+  string registered_saver = 16;
 }
 
 // A SavedUserObject is an object (in the object-oriented language of the
@@ -167,7 +205,6 @@ message SavedVariable {
   repeated SavedVariable experimental_distributed_variable_components = 8;
 }
 
-
 // Represents `FunctionSpec` used in `Function`. This represents a
 // function that has been wrapped as a TensorFlow `Function`.
 message FunctionSpec {
@@ -208,6 +245,7 @@ message SavedResource {
 
 message SaveableObject {
   // Node ids of concrete functions for saving and loading from a checkpoint.
+  // These functions save and restore directly from tensors.
   int32 save_function = 2;
   int32 restore_function = 3;
 }
diff --git a/proto/tensorflow/core/protobuf/struct.proto b/proto/tensorflow/core/protobuf/struct.proto
index 1bc8cdc..06d8be3 100644
--- a/proto/tensorflow/core/protobuf/struct.proto
+++ b/proto/tensorflow/core/protobuf/struct.proto
@@ -72,6 +72,10 @@ message StructuredValue {
     DictValue dict_value = 53;
     // Represents Python's namedtuple.
     NamedTupleValue named_tuple_value = 54;
+    // Represents a value for tf.Tensor.
+    tensorflow.TensorProto tensor_value = 55;
+    // Represents a value for np.ndarray.
+    tensorflow.TensorProto numpy_value = 56;
   }
 }
 
@@ -149,9 +153,12 @@ message TypeSpecProto {
   //  * If type_spec_class == REGISTERED_TYPE_SPEC, the TypeSpec class is
   //    the one registered under this name. For types registered outside
   //    core TensorFlow by an add-on library, that library must be loaded
-  //    before this value can be deserialized by StructureCoder.
+  //    before this value can be deserialized by nested_structure_coder.
   //  * If type_spec_class specifies a particular TypeSpec class, this field is
   //    redundant with the type_spec_class enum, and is only used for error
   //    reporting in older binaries that do not know the tupe_spec_class enum.
   string type_spec_class_name = 3;
+
+  // The number of flat tensor components required by this TypeSpec.
+  int32 num_flat_components = 4;
 }
diff --git a/proto/tensorflow/core/protobuf/trackable_object_graph.proto b/proto/tensorflow/core/protobuf/trackable_object_graph.proto
index 9371af8..7e1e190 100644
--- a/proto/tensorflow/core/protobuf/trackable_object_graph.proto
+++ b/proto/tensorflow/core/protobuf/trackable_object_graph.proto
@@ -2,6 +2,8 @@ syntax = "proto3";
 
 package tensorflow;
 
+import "google/protobuf/wrappers.proto";
+
 option cc_enable_arenas = true;
 option go_package = "github.com/tensorflow/tensorflow/tensorflow/go/core/protobuf";
 
@@ -31,10 +33,10 @@ message TrackableObjectGraph {
       string full_name = 2;
       // The generated name of the Tensor in the checkpoint.
       string checkpoint_key = 3;
-      // Whether checkpoints should be considered as matching even without this
-      // value restored. Used for non-critical values which don't affect the
-      // TensorFlow graph, such as layer configurations.
-      bool optional_restore = 4;
+      // Deprecated bool field for optional restore. This field has never been
+      // set to True.
+      reserved "optional_restore";
+      reserved 4;
     }
 
     message SlotVariableReference {
@@ -54,7 +56,25 @@ message TrackableObjectGraph {
     repeated SerializedTensor attributes = 2;
     // Slot variables owned by this object.
     repeated SlotVariableReference slot_variables = 3;
+
+    // The registered saver used to save this object. If this saver is not
+    // present when loading the checkpoint, then loading will fail.
+    RegisteredSaver registered_saver = 4;
+
+    // Whether this object has checkpoint values or descendants with checkpoint
+    // values. This is computed at save time to avoid traversing the entire
+    // object graph proto when restoring (which also has to traverse the live
+    // object graph).
+    google.protobuf.BoolValue has_checkpoint_values = 5;
   }
 
   repeated TrackableObject nodes = 1;
 }
+
+message RegisteredSaver {
+  // The name of the registered saver/restore function.
+  string name = 1;
+
+  // Unique auto-generated name of the object.
+  string object_name = 2;
+}
diff --git a/proto/tensorflow_serving/apis/classification.proto b/proto/tensorflow_serving/apis/classification.proto
index 16a90b1..4de0b7d 100644
--- a/proto/tensorflow_serving/apis/classification.proto
+++ b/proto/tensorflow_serving/apis/classification.proto
@@ -1,11 +1,11 @@
 syntax = "proto3";
 
-option cc_enable_arenas = true;
+package tensorflow.serving;
 
 import "tensorflow_serving/apis/input.proto";
 import "tensorflow_serving/apis/model.proto";
 
-package tensorflow.serving;
+option cc_enable_arenas = true;
 
 // A single class.
 message Class {
diff --git a/proto/tensorflow_serving/apis/logging.proto b/proto/tensorflow_serving/apis/logging.proto
index 6298bb4..0f9e876 100644
--- a/proto/tensorflow_serving/apis/logging.proto
+++ b/proto/tensorflow_serving/apis/logging.proto
@@ -13,5 +13,8 @@ message LogMetadata {
   SamplingConfig sampling_config = 2;
   // List of tags used to load the relevant MetaGraphDef from SavedModel.
   repeated string saved_model_tags = 3;
-  // TODO(b/33279154): Add more metadata as mentioned in the bug.
+  int64 timestamp_secs = 4;        // Seconds since epoch.
+  string dc = 5;                   // Datacenter where the request was logged.
+  string request_origin = 6;       // Request origin identifier.
+  string request_criticality = 7;  // Request QoS.
 }
diff --git a/proto/tensorflow_serving/apis/model.proto b/proto/tensorflow_serving/apis/model.proto
index 56493f6..afd9590 100644
--- a/proto/tensorflow_serving/apis/model.proto
+++ b/proto/tensorflow_serving/apis/model.proto
@@ -1,10 +1,11 @@
 syntax = "proto3";
 
 package tensorflow.serving;
-option cc_enable_arenas = true;
 
 import "google/protobuf/wrappers.proto";
 
+option cc_enable_arenas = true;
+
 // Metadata for an inference request such as the model name and version.
 message ModelSpec {
   // Required servable name.
@@ -12,8 +13,10 @@ message ModelSpec {
 
   // Optional choice of which version of the model to use.
   //
-  // Recommended to be left unset in the common case. Should be specified only
-  // when there is a strong version consistency requirement.
+  // Expected to be left unset in the common case. Should be specified when
+  // there is a strong version consistency requirement (e.g. when the model
+  // signature changes across versions and requests need to be
+  // version-specific).
   //
   // When left unspecified, the system will serve the best available version.
   // This is typically the latest version, though during version transitions,
diff --git a/proto/tensorflow_serving/apis/model_management.proto b/proto/tensorflow_serving/apis/model_management.proto
index 6bbf00f..2eba09b 100644
--- a/proto/tensorflow_serving/apis/model_management.proto
+++ b/proto/tensorflow_serving/apis/model_management.proto
@@ -9,8 +9,17 @@ option cc_enable_arenas = true;
 
 message ReloadConfigRequest {
   ModelServerConfig config = 1;
+  repeated string metric_names = 2;
 }
 
 message ReloadConfigResponse {
   StatusProto status = 1;
+  repeated Metric metric = 2;
+}
+
+message Metric {
+  string name = 1;
+  oneof value_increase {
+    int64 int64_value_increase = 2;
+  }
 }
diff --git a/proto/tensorflow_serving/apis/predict.proto b/proto/tensorflow_serving/apis/predict.proto
index 0e9ebf3..e243b7a 100644
--- a/proto/tensorflow_serving/apis/predict.proto
+++ b/proto/tensorflow_serving/apis/predict.proto
@@ -1,11 +1,12 @@
 syntax = "proto3";
 
 package tensorflow.serving;
-option cc_enable_arenas = true;
 
 import "tensorflow/core/framework/tensor.proto";
 import "tensorflow_serving/apis/model.proto";
 
+option cc_enable_arenas = true;
+
 // PredictRequest specifies which TensorFlow model to run, as well as
 // how inputs are mapped to tensors and how outputs are filtered before
 // returning to user.
@@ -28,6 +29,148 @@ message PredictRequest {
   // exception that when none is specified, all tensors specified in the
   // named signature will be run/fetched and returned.
   repeated string output_filter = 3;
+
+  // Reserved field 4.
+  reserved 4;
+
+  // Options for streaming requests to control how multiple requests/responses
+  // are handled within a single stream.
+  PredictStreamedOptions predict_streamed_options = 5;
+
+  // Client identifier to group requests belonging to a specific entity.
+  // Example entities can be product ids, service names, user ids etc.
+  // Servers can use this to optimize placement, caching and colocation.
+  // TODO(b/329897437): Migrate to client_id in RequestOptions.
+  optional bytes client_id = 6;
+
+  // Options for PredictRequest.
+  message RequestOptions {
+    // Client identifier to group requests belonging to a specific entity.
+    // Example entities can be product ids, service names, user ids etc.
+    // Servers can use this to optimize placement, caching and colocation.
+    optional bytes client_id = 1;
+
+    // Deterministic mode for the request. When specified, model servers will
+    // reduce numeric instability based on different mode selections.
+    enum DeterministicMode {
+      DETERMINISTIC_MODE_UNSPECIFIED = 0;
+
+      // Only supported in disaggregated serving. When set, the request will be
+      // pinned to a fixed decoder slot index that's deterministic across
+      // processes.
+      FIXED_DECODER_SLOT = 1;
+    }
+
+    optional DeterministicMode deterministic_mode = 2;
+  }
+
+  optional RequestOptions request_options = 7;
+}
+
+// Options only used for streaming requests that control how inputs/ouputs are
+// handled in the stream.
+message PredictStreamedOptions {
+  // Request state used to handle splitting of requests.  NONE is the
+  // default when the stream request is not split and used for a single-turn,
+  // single request.
+  //
+  // SPLIT is used when multiple streamed requests are used to generate a
+  // logical request. END_SPLIT should be called for the last split of the
+  // multi-turn request to start the processing of the current turn. NONE can
+  // not be interspersed with SPLIT and END_SPLIT messages.
+  // If another request is sent on the same stream after END_SPLIT, it can be
+  // either SPLIT or END_SPLIT to start accumulating input or trigger the next
+  // model turn respectively.
+  //
+  // Some examples with a mix of request states and the logical request.
+  //
+  // Example 1:
+  //   NONE
+  //
+  // Single turn, single request.
+  //
+  // Example 2 :
+  //   END_SPLIT
+  //
+  // Will be treated as a single logical input request for a single turn,
+  // similar to Example 1.
+  //
+  // Example 3:
+  //   SPLIT
+  //   SPLIT
+  //   END_SPLIT
+  //
+  // Will be treated as a single logical input request for a single turn,
+  // similar to Example 1.
+  //
+  // Example 4:
+  //   END_SPLIT
+  //   END_SPLIT
+  //
+  // Will be treated as two logical turn requests (1. END_SPLIT 2. END_SPLIT)
+  //
+  // Example 5:
+  //   SPLIT
+  //   END_SPLIT
+  //   SPLIT
+  //   SPLIT
+  //   END_SPLIT
+  //
+  // Will be treated as two logical turn requests (1. SPLIT, END_SPLIT 2. SPLIT,
+  // SPLIT, END_SPLIT)
+  //
+  // Incorrect Example 1:
+  //   NONE
+  //   END_SPLIT
+  //
+  // Invalid because NONE and END_SPLIT are interspersed.
+  //
+  // Incorrect Example 2:
+  //   SPLIT
+  //   SPLIT
+  //
+  // Invalid because END_SPLIT is never called.
+  //
+  // Incorrect Example 3:
+  //   SPLIT
+  //   NONE
+  //   SPLIT
+  //   END_SPLIT
+  //
+  // Invalid because NONE is interspersed with SPLIT/END_SPLIT.
+
+  enum RequestState {
+    NONE = 0;
+    SPLIT = 1;
+    END_SPLIT = 2;
+  }
+
+  // Request state used to handle segmentation of requests.
+  RequestState request_state = 1;
+
+  // Input tensors split dimensions.
+  // Defines the dimension used to split input tensors specified
+  // in PredictRequest.inputs. The dimension will be used
+  // for concatenation of multiple SPLIT requests.
+  //
+  // For input tensor in PredictRequest.inputs that are not contained in this
+  // map, the tensors from the first SPLIT request will be used.
+  //
+  // For example, with an original input tensor of [[1, 2, 3, 4], [5, 6, 7, 8]].
+  //
+  // For a split dimension of 0 and two requests (SPLIT and END_SPLIT), the
+  // input tensors for request 1 should be [1, 2, 3, 4] and request 2 should be
+  // be [5, 6, 7, 8].
+  //
+  // For a split dimension of 1 and two requests (SPLIT and END_SPLIT), the
+  // input tensors for request 1 should be [[1, 2], [5, 6]] and request 2 should
+  // be [[3, 4], [7, 8]].
+  map<string, int32> split_dimensions = 2;
+
+  // If true, there will be a single PredictResponse output.
+  // If false, output can be split into 1 or more PredictResponses.
+  // Value of this field should be the same for all requests in the stream.
+  bool return_single_response = 3;
 }
 
 // Response for PredictRequest on successful run.
diff --git a/proto/tensorflow_serving/apis/prediction_log.proto b/proto/tensorflow_serving/apis/prediction_log.proto
index c75b68a..9deebc7 100644
--- a/proto/tensorflow_serving/apis/prediction_log.proto
+++ b/proto/tensorflow_serving/apis/prediction_log.proto
@@ -26,6 +26,11 @@ message PredictLog {
   PredictResponse response = 2;
 }
 
+message PredictStreamedLog {
+  repeated PredictRequest request = 1;
+  repeated PredictResponse response = 2;
+}
+
 message MultiInferenceLog {
   MultiInferenceRequest request = 1;
   MultiInferenceResponse response = 2;
@@ -43,6 +48,7 @@ message PredictionLog {
     ClassifyLog classify_log = 2;
     RegressLog regress_log = 3;
     PredictLog predict_log = 6;
+    PredictStreamedLog predict_streamed_log = 7;
     MultiInferenceLog multi_inference_log = 4;
     SessionRunLog session_run_log = 5;
   }
diff --git a/proto/tensorflow_serving/apis/regression.proto b/proto/tensorflow_serving/apis/regression.proto
index e0ac300..5fb79ba 100644
--- a/proto/tensorflow_serving/apis/regression.proto
+++ b/proto/tensorflow_serving/apis/regression.proto
@@ -1,11 +1,11 @@
 syntax = "proto3";
 
-option cc_enable_arenas = true;
+package tensorflow.serving;
 
 import "tensorflow_serving/apis/input.proto";
 import "tensorflow_serving/apis/model.proto";
 
-package tensorflow.serving;
+option cc_enable_arenas = true;
 
 // Regression result for a single item (tensorflow.Example).
 message Regression {
diff --git a/proto/tensorflow_serving/config/file_system_storage_path_source.proto b/proto/tensorflow_serving/config/file_system_storage_path_source.proto
index add7aa2..6924fe4 100644
--- a/proto/tensorflow_serving/config/file_system_storage_path_source.proto
+++ b/proto/tensorflow_serving/config/file_system_storage_path_source.proto
@@ -16,8 +16,7 @@ message FileSystemStoragePathSourceConfig {
     }
 
     // Serve all versions found on disk.
-    message All {
-    }
+    message All {}
 
     // Serve a specific version (or set of versions).
     //
@@ -56,12 +55,6 @@ message FileSystemStoragePathSourceConfig {
   // The servables to monitor for new versions, and aspire.
   repeated ServableToMonitor servables = 5;
 
-  // A single servable name/base_path pair to monitor.
-  // DEPRECATED: Use 'servables' instead.
-  // TODO(b/30898016): Stop using these fields, and ultimately remove them here.
-  string servable_name = 1 [deprecated = true];
-  string base_path = 2 [deprecated = true];
-
   // How long to wait between file-system polling to look for children of
   // 'base_path', in seconds.
   //
@@ -85,4 +78,6 @@ message FileSystemStoragePathSourceConfig {
   // contain at least one version under the base path. In addition, if a polling
   // loop find the base path empty, it will not unload existing servables.
   bool servable_versions_always_present = 6;
+
+  reserved 1, 2;
 }
diff --git a/proto/tensorflow_serving/config/logging_config.proto b/proto/tensorflow_serving/config/logging_config.proto
index 8ae4d4c..31fbe94 100644
--- a/proto/tensorflow_serving/config/logging_config.proto
+++ b/proto/tensorflow_serving/config/logging_config.proto
@@ -1,14 +1,25 @@
 syntax = "proto3";
 
 package tensorflow.serving;
-option cc_enable_arenas = true;
 
 import "tensorflow_serving/config/log_collector_config.proto";
 
+option cc_enable_arenas = true;
+
 message SamplingConfig {
-  // Requests will be logged uniformly at random with this probability. Valid
-  // range: [0, 1.0].
+  // Requests will be logged uniformly at random with this probability.
+  // Valid range: [0, 1.0].
   double sampling_rate = 1;
+
+  // Attributes of requests that can be optionally sampled.
+  // Note: Enabling more attributes will increase logging storage requirements.
+  enum Attributes {
+    ATTR_DEFAULT = 0x0;
+    ATTR_REQUEST_ORIGIN = 0x1;
+    ATTR_REQUEST_CRITICALITY = 0x2;
+  }
+  // Bitwise OR of above attributes
+  int32 attributes = 2;
 }
 
 // Configuration for logging query/responses.