feat: Reduce input stream for PromptTemplate, LLM, ChatModel, Retriev…

…er and Tool (#388)

docs/modules/memory/memory.md

@@ -42,7 +42,7 @@ You may want to use this class directly if you are managing memory outside of a
 ```dart
 final history = ChatMessageHistory();
 
-history.addUserChatMessage('hi!');
+history.addHumanChatMessage('hi!');
 history.addAIChatMessage('whats up?');
 
 print(await history.getChatMessages());
@@ -60,7 +60,7 @@ We can first extract it as a string.
 ```dart
 final memory = ConversationBufferMemory();
 
-memory.chatHistory.addUserChatMessage('hi!');
+memory.chatHistory.addHumanChatMessage('hi!');
 memory.chatHistory.addAIChatMessage('whats up
 
 print(await memory.loadMemoryVariables());
@@ -72,7 +72,7 @@ We can also get the history as a list of messages:
 ```dart
 final memory = ConversationBufferMemory(returnMessages: true);
 
-memory.chatHistory.addUserChatMessage('hi!');
+memory.chatHistory.addHumanChatMessage('hi!');
 memory.chatHistory.addAIChatMessage('whats up?');
 
 print(await memory.loadMemoryVariables());

docs/modules/model_io/models/llms/how_to/fake_llm.md

@@ -4,14 +4,14 @@ We expose some fake LLM classes that can be used for testing. This allows you
 to mock out calls to the LLM and simulate what would happen if the LLM 
 responded in a certain way.
 
-## FakeListLLM
+## FakeLLM
 
 You can configure a list of responses that the LLM will return in order.
 
 Example:
 ```dart
 test('Test LLMChain call', () async {
-  final model = FakeListLLM(responses: ['Hello world!']);
+  final model = FakeLLM(responses: ['Hello world!']);
   final prompt = PromptTemplate.fromTemplate('Print {foo}');
   final chain = LLMChain(prompt: prompt, llm: model);
   final res = await chain.call({'foo': 'Hello world!'});

examples/browser_summarizer/pubspec.lock

@@ -382,6 +382,14 @@ packages:
       url: "https://pub.dev"
     source: hosted
     version: "0.2.0"
+  rxdart:
+    dependency: transitive
+    description:
+      name: rxdart
+      sha256: "0c7c0cedd93788d996e33041ffecda924cc54389199cde4e6a34b440f50044cb"
+      url: "https://pub.dev"
+    source: hosted
+    version: "0.27.7"
   shared_preferences:
     dependency: "direct main"
     description:

examples/docs_examples/pubspec.lock

@@ -356,6 +356,14 @@ packages:
       url: "https://pub.dev"
     source: hosted
     version: "0.2.0"
+  rxdart:
+    dependency: transitive
+    description:
+      name: rxdart
+      sha256: "0c7c0cedd93788d996e33041ffecda924cc54389199cde4e6a34b440f50044cb"
+      url: "https://pub.dev"
+    source: hosted
+    version: "0.27.7"
   source_span:
     dependency: transitive
     description:

examples/hello_world_backend/pubspec.lock

@@ -165,6 +165,14 @@ packages:
       url: "https://pub.dev"
     source: hosted
     version: "1.9.0"
+  rxdart:
+    dependency: transitive
+    description:
+      name: rxdart
+      sha256: "0c7c0cedd93788d996e33041ffecda924cc54389199cde4e6a34b440f50044cb"
+      url: "https://pub.dev"
+    source: hosted
+    version: "0.27.7"
   shelf:
     dependency: "direct main"
     description:

examples/hello_world_cli/pubspec.lock

@@ -157,6 +157,14 @@ packages:
       url: "https://pub.dev"
     source: hosted
     version: "1.9.0"
+  rxdart:
+    dependency: transitive
+    description:
+      name: rxdart
+      sha256: "0c7c0cedd93788d996e33041ffecda924cc54389199cde4e6a34b440f50044cb"
+      url: "https://pub.dev"
+    source: hosted
+    version: "0.27.7"
   source_span:
     dependency: transitive
     description:

examples/hello_world_flutter/pubspec.lock

@@ -210,6 +210,14 @@ packages:
       url: "https://pub.dev"
     source: hosted
     version: "6.1.1"
+  rxdart:
+    dependency: transitive
+    description:
+      name: rxdart
+      sha256: "0c7c0cedd93788d996e33041ffecda924cc54389199cde4e6a34b440f50044cb"
+      url: "https://pub.dev"
+    source: hosted
+    version: "0.27.7"
   sky_engine:
     dependency: transitive
     description: flutter

packages/langchain/example/langchain_example.dart

@@ -5,7 +5,7 @@ void main() async {
   final promptTemplate = PromptTemplate.fromTemplate(
     'tell me a joke about {subject}',
   );
-  final llm = FakeListLLM(
+  final llm = FakeLLM(
     responses: ['Why did the AI go on a diet? Because it had too many bytes!'],
   );
   final chain = promptTemplate.pipe(llm).pipe(const StringOutputParser());

packages/langchain/test/chains/base_test.dart

@@ -94,15 +94,15 @@ void main() {
 
   group('Runnable tests', () {
     test('Chain as Runnable', () async {
-      final model = FakeListLLM(responses: ['Hello world!']);
+      final model = FakeLLM(responses: ['Hello world!']);
       final prompt = PromptTemplate.fromTemplate('Print {foo}');
       final run = LLMChain(prompt: prompt, llm: model);
       final res = await run.invoke({'foo': 'Hello world!'});
       expect(res[LLMChain.defaultOutputKey], 'Hello world!');
     });
 
     test('Streaming Chain', () async {
-      final model = FakeListLLM(responses: ['Hello world!']);
+      final model = FakeLLM(responses: ['Hello world!']);
       final prompt = PromptTemplate.fromTemplate('Print {foo}');
       final run = LLMChain(prompt: prompt, llm: model);
       final stream = run.stream({'foo': 'Hello world!'});

packages/langchain/test/chains/combine_documents/map_reduce_test.dart

@@ -37,7 +37,7 @@ void main() {
     }
 
     test('Test MapReduceDocumentsChain with LLM', () async {
-      final model = FakeListLLM(
+      final model = FakeLLM(
         responses: [
           // Summarize this content: Hello 1!
           '1',

packages/langchain/test/chains/combine_documents/reduce_test.dart

@@ -4,7 +4,7 @@ import 'package:test/test.dart';
 void main() {
   group('ReduceDocumentsChain tests', () {
     test('Test reduce', () async {
-      final llm = FakeListLLM(
+      final llm = FakeLLM(
         responses: [
           // Summarize this content: Hello 1!\n\nHello 2!\n\nHello 3!\n\nHello 4!
           'Hello 1234!',
@@ -32,7 +32,7 @@ void main() {
     });
 
     test('Test reduce and collapse', () async {
-      final llm = FakeListLLM(
+      final llm = FakeLLM(
         responses: [
           // Collapse this content: Hello 1!\n\nHello 2!\n\nHello 3!
           'Hello 123!',

packages/langchain_core/lib/src/chat_models/base.dart

@@ -2,6 +2,7 @@ import 'package:meta/meta.dart';
 
 import '../language_models/language_models.dart';
 import '../prompts/types.dart';
+import '../utils/reduce.dart';
 import 'types.dart';
 
 /// {@template base_chat_model}
@@ -15,22 +16,15 @@ abstract class BaseChatModel<Options extends ChatModelOptions>
     required super.defaultOptions,
   });
 
-  /// Runs the chat model on the given prompt value.
-  ///
-  /// - [input] The prompt value to pass into the model.
-  /// - [options] Generation options to pass into the Chat Model.
-  ///
-  /// Example:
-  /// ```dart
-  /// final result = await chat.invoke(
-  ///   PromptValue.chat([ChatMessage.humanText('say hi!')]),
-  /// );
-  /// ```
   @override
-  Future<ChatResult> invoke(
-    final PromptValue input, {
+  Stream<ChatResult> streamFromInputStream(
+    final Stream<PromptValue> inputStream, {
     final Options? options,
-  });
+  }) async* {
+    final input = await inputStream.toList();
+    final reduced = reduce<PromptValue>(input);
+    yield* stream(reduced, options: options);
+  }
 
   /// Runs the chat model on the given messages and returns a chat message.
   ///

packages/langchain_core/lib/src/chat_models/types.dart

@@ -31,7 +31,7 @@ class ChatResult extends LanguageModelResult<AIChatMessage> {
   String get outputAsString => output.content;
 
   @override
-  LanguageModelResult<AIChatMessage> concat(
+  ChatResult concat(
     final LanguageModelResult<AIChatMessage> other,
   ) {
     return ChatResult(

packages/langchain_core/lib/src/language_models/base.dart

@@ -25,16 +25,6 @@ abstract class BaseLanguageModel<
   /// Return type of language model.
   String get modelType;
 
-  /// Runs the Language Model on the given prompt value.
-  ///
-  /// - [input] The prompt value to pass into the model.
-  /// - [options] Generation options to pass into the model.
-  @override
-  Future<Output> invoke(
-    final PromptValue input, {
-    final Options? options,
-  });
-
   /// Tokenizes the given prompt using the encoding used by the language
   /// model.
   ///

packages/langchain_core/lib/src/llms/base.dart

@@ -16,23 +16,6 @@ abstract class BaseLLM<Options extends LLMOptions>
     required super.defaultOptions,
   });
 
-  /// Runs the LLM on the given prompt value.
-  ///
-  /// - [input] The prompt value to pass into the model.
-  /// - [options] Generation options to pass into the LLM.
-  ///
-  /// Example:
-  /// ```dart
-  /// final result = await openai.invoke(
-  ///   PromptValue.string('Tell me a joke.'),
-  /// );
-  /// ```
-  @override
-  Future<LLMResult> invoke(
-    final PromptValue input, {
-    final Options? options,
-  });
-
   /// Runs the LLM on the given String prompt and returns a String with the
   /// generated text.
   ///

packages/langchain_core/lib/src/llms/fake.dart

@@ -7,9 +7,9 @@ import 'types.dart';
 /// Fake LLM for testing.
 /// You can pass in a list of responses to return in order when called.
 /// {@endtemplate}
-class FakeListLLM extends SimpleLLM {
+class FakeLLM extends SimpleLLM {
   /// {@macro fake_list_llm}
-  FakeListLLM({
+  FakeLLM({
     required this.responses,
   }) : super(defaultOptions: const LLMOptions());
 
@@ -29,6 +29,24 @@ class FakeListLLM extends SimpleLLM {
     return Future<String>.value(responses[_i++ % responses.length]);
   }
 
+  @override
+  Stream<LLMResult> stream(
+    final PromptValue input, {
+    final LLMOptions? options,
+  }) {
+    final res = responses[_i++ % responses.length].split('');
+    return Stream.fromIterable(res).map(
+      (final item) => LLMResult(
+        id: 'fake-echo',
+        output: item,
+        finishReason: FinishReason.unspecified,
+        metadata: const {},
+        usage: const LanguageModelUsage(),
+        streaming: true,
+      ),
+    );
+  }
+
   @override
   Future<List<int>> tokenize(
     final PromptValue promptValue, {

packages/langchain_core/lib/src/llms/types.dart

@@ -31,7 +31,7 @@ class LLMResult extends LanguageModelResult<String> {
   String get outputAsString => output;
 
   @override
-  LanguageModelResult<String> concat(
+  LLMResult concat(
     final LanguageModelResult<String> other,
   ) {
     return LLMResult(

packages/langchain_core/lib/src/prompts/base_prompt.dart

@@ -3,6 +3,7 @@ import 'package:meta/meta.dart';
 
 import '../langchain/types.dart';
 import '../runnables/runnable.dart';
+import '../utils/reduce.dart';
 import 'template.dart';
 import 'types.dart';
 
@@ -68,46 +69,15 @@ abstract base class BasePromptTemplate
     return Future.value(formatPrompt(input));
   }
 
-  @override
-  Stream<PromptValue> stream(
-    final InputValues input, {
-    final BaseLangChainOptions? options,
-  }) {
-    return streamFromInputStream(
-      Stream.value(input).asBroadcastStream(),
-      options: options,
-    );
-  }
-
   @override
   Stream<PromptValue> streamFromInputStream(
     final Stream<InputValues> inputStream, {
     final BaseLangChainOptions? options,
-  }) {
-    final userKeys = inputVariables.difference(
-      partialVariables?.keys.toSet() ?? {},
-    );
-    final userInput = <String, dynamic>{};
-    return inputStream
-        .asyncMap((final InputValues inputValues) {
-          for (final input in inputValues.entries) {
-            final key = input.key;
-            final value = input.value;
-            if (value is String) {
-              userInput[key] = (userInput[key] as String? ?? '') + value;
-            } else {
-              userInput[key] = value;
-            }
-          }
-          final hasAllUserValues = userKeys.every(userInput.containsKey);
-          if (hasAllUserValues) {
-            return formatPrompt(userInput);
-          } else {
-            return null;
-          }
-        })
-        .where((final res) => res != null)
-        .cast();
+  }) async* {
+    final List<InputValues> input = await inputStream.toList();
+    final InputValues reduced =
+        input.isEmpty ? const {} : reduce<InputValues>(input);
+    yield* stream(reduced, options: options);
   }
 
   /// Format the prompt given the input values and return a formatted string.