TokenUsage class

One token-accounting shape embedded by every result and metrics message, replacing the input/output/total/throughput quadruple that was copied inline across LLM, VLM, and RAG results. Names follow the OpenAI Responses API; the timing fields follow llama.cpp's timings object, which names the phase it measures.

Available extensions

GeneratedMessageGenericExtensions

Constructors

TokenUsage({int? inputTokens, int? outputTokens, int? totalTokens, double? decodeTokensPerSecond, Int64? prefillMs, Int64? ttftMs, Int64? timeToFirstContentTokenMs, double? contentTokensPerSecond, bool? batchBuffered, bool? countsEstimated}): factory
TokenUsage.fromBuffer(List<int> data, [ExtensionRegistry registry = $pb.ExtensionRegistry.EMPTY]): factory
TokenUsage.fromJson(String json, [ExtensionRegistry registry = $pb.ExtensionRegistry.EMPTY]): factory

Properties

batchBuffered ↔ bool: True when the backend buffered the whole generation and flushed deltas at once, so the decode window is an artifact of the flush. Platforms must not re-derive this heuristic.
getter/setter pair
contentTokensPerSecond ↔ double: Content-only throughput over first-content-delta → last delta. Excludes reasoning tokens the accelerator also decoded. 0 when content count or window is unavailable.
getter/setter pair
countsEstimated ↔ bool: True when input_tokens / output_tokens were estimated (e.g. chars/4) rather than reported by the engine. Absence of the flag (false) means the counts are engine-measured.
getter/setter pair
decodeTokensPerSecond ↔ double: Decode-phase throughput only: output_tokens / decode_ms. Excludes prefill. cf. llama.cpp timings.predicted_per_second.
getter/setter pair
hashCode → int: Calculates a hash code based on the contents of the protobuf.
no setterinherited
info_ → BuilderInfo: no setter
inputTokens ↔ int: getter/setter pair
isFrozen → bool: Returns true if this message is marked read-only. Otherwise false.
no setterinherited
outputTokens ↔ int: getter/setter pair
prefillMs ↔ Int64: Prefill (prompt eval) wall time. cf. llama.cpp timings.prompt_ms, Ollama prompt_eval_duration. 0 when the backend does not report it.
getter/setter pair
runtimeType → Type: A representation of the runtime type of the object.
no setterinherited
timeToFirstContentTokenMs ↔ Int64: Request start to the first CONTENT delta — what the user actually waits for when the model reasons first. 0 when no content token was ever delivered. Distinct from ttft_ms; do not alias the two.
getter/setter pair
totalTokens ↔ int: getter/setter pair
ttftMs ↔ Int64: Request start to first output token of any kind (reasoning or content). The canonical spelling for every result type: LLMGenerationResult and VLMResult report TTFT here and nowhere else. SDKEvent's own telemetry fields (GenerationEvent.time_to_first_token_ms, first_token_latency_ms) keep their separate event-stream spelling.
getter/setter pair
unknownFields → UnknownFieldSet: no setterinherited

Methods

addExtension(Extension extension, Object? value) → void: Adds an extension field value to a repeated field.
inherited
check() → void: Throws a StateError if the message has required fields without a value.
inherited
clear() → void: Clears all data that was set in this message.
inherited
clearBatchBuffered() → void
clearContentTokensPerSecond() → void
clearCountsEstimated() → void
clearDecodeTokensPerSecond() → void
clearExtension(Extension extension) → void: Clears an extension field and also removes the extension.
inherited
clearField(int tagNumber) → void: Clears the contents of a given field.
inherited
clearInputTokens() → void
clearOutputTokens() → void
clearPrefillMs() → void
clearTimeToFirstContentTokenMs() → void
clearTotalTokens() → void
clearTtftMs() → void
clone() → TokenUsage: Creates a deep copy of the fields in this message.
copyWith(void updates(TokenUsage)) → TokenUsage: Apply updates to a copy of this message.
createEmptyInstance() → TokenUsage: Creates an empty instance of the same message type as this.
deepCopy() → T: Available on T, provided by the GeneratedMessageGenericExtensions extension
Returns a writable deep copy of this message.
extensionsAreInitialized() → bool: inherited
freeze() → GeneratedMessage: Make this message read-only.
inherited
getDefaultForField(int tagNumber) → dynamic: Returns the default value for the given field.
inherited
getExtension(Extension extension) → dynamic: Returns the value of extension.
inherited
getField(int tagNumber) → dynamic: Returns the value of the field associated with tagNumber, or the default value if it is not set.
inherited
getFieldOrNull(int tagNumber) → dynamic: Returns the value of a field, ignoring any defaults.
inherited
getTagNumber(String fieldName) → int?: inherited
hasBatchBuffered() → bool
hasContentTokensPerSecond() → bool
hasCountsEstimated() → bool
hasDecodeTokensPerSecond() → bool
hasExtension(Extension extension) → bool: Returns true if a value of extension is present.
inherited
hasField(int tagNumber) → bool: Whether this message has a field associated with tagNumber.
inherited
hasInputTokens() → bool
hasOutputTokens() → bool
hasPrefillMs() → bool
hasRequiredFields() → bool: Whether the message has required fields.
inherited
hasTimeToFirstContentTokenMs() → bool
hasTotalTokens() → bool
hasTtftMs() → bool
isInitialized() → bool: Whether all required fields in the message and embedded messages are set.
inherited
mergeFromBuffer(List<int> input, [ExtensionRegistry extensionRegistry = ExtensionRegistry.EMPTY]) → void: Merges serialized protocol buffer data into this message.
inherited
mergeFromCodedBufferReader(CodedBufferReader input, [ExtensionRegistry extensionRegistry = ExtensionRegistry.EMPTY]) → void: Same as mergeFromBuffer, but takes a CodedBufferReader input.
inherited
mergeFromJson(String data, [ExtensionRegistry extensionRegistry = ExtensionRegistry.EMPTY]) → void: Merges field values from data, a JSON object, encoded as described by GeneratedMessage.writeToJson.
inherited
mergeFromJsonMap(Map<String, dynamic> json, [ExtensionRegistry extensionRegistry = ExtensionRegistry.EMPTY]) → void: Merges field values from a JSON object represented as a Dart map.
inherited
mergeFromMessage(GeneratedMessage other) → void: Merges the contents of the other into this message.
inherited
mergeFromProto3Json(Object? json, {TypeRegistry typeRegistry = const TypeRegistry.empty(), bool ignoreUnknownFields = false, bool supportNamesWithUnderscores = true, bool permissiveEnums = false}) → void: Merges field values from json, a JSON object using proto3 encoding.
inherited
mergeUnknownFields(UnknownFieldSet unknownFieldSet) → void: inherited
noSuchMethod(Invocation invocation) → dynamic: Invoked when a nonexistent method or property is accessed.
inherited
rebuild(void updates(T)) → T: Available on T, provided by the GeneratedMessageGenericExtensions extension
Apply updates to a copy of this message.
setExtension(Extension extension, Object value) → void: Sets the value of a non-repeated extension field to value.
inherited
setField(int tagNumber, Object value) → void: Sets the value of a field by its tagNumber.
inherited
toBuilder() → GeneratedMessage: Creates a writable, shallow copy of this message.
inherited
toDebugString() → String: Returns a String representation of this message.
inherited
toProto3Json({TypeRegistry typeRegistry = const TypeRegistry.empty()}) → Object?: Returns an Object representing Proto3 JSON serialization of this.
inherited
toString() → String: Returns a String representation of this message.
inherited
writeToBuffer() → Uint8List: Serialize the message as the protobuf binary format.
inherited
writeToCodedBufferWriter(CodedBufferWriter output) → void: Same as writeToBuffer, but serializes to the given CodedBufferWriter.
inherited
writeToJson() → String: Returns a JSON string that encodes this message.
inherited
writeToJsonMap() → Map<String, dynamic>: Returns the JSON encoding of this message as a Dart Map.
inherited

Operators

operator ==(Object other) → bool: The equality operator.
inherited

Static Methods

create() → TokenUsage
getDefault() → TokenUsage