feat(llms): Add streaming support to OpenAI (davidmigloz#196)

KennethKnudsen97 · Apr 22, 2024 · 3f3cf0f · 3f3cf0f
1 parent d4e5cb3
commit 3f3cf0f
Show file tree

Hide file tree

Showing 3 changed files with 70 additions and 9 deletions.
diff --git a/packages/langchain_openai/lib/src/llms/models/mappers.dart b/packages/langchain_openai/lib/src/llms/models/mappers.dart
@@ -2,7 +2,7 @@ import 'package:langchain/langchain.dart';
 import 'package:openai_dart/openai_dart.dart';
 
 extension CreateCompletionResponseMapper on CreateCompletionResponse {
-  LLMResult toLLMResult() {
+  LLMResult toLLMResult({final bool streaming = false}) {
     return LLMResult(
       generations: choices
           .map((final choice) => choice.toLLMGeneration())
@@ -13,6 +13,7 @@ extension CreateCompletionResponseMapper on CreateCompletionResponse {
         'created': created,
         'model': model,
       },
+      streaming: streaming,
     );
   }
 }

diff --git a/packages/langchain_openai/lib/src/llms/openai.dart b/packages/langchain_openai/lib/src/llms/openai.dart
@@ -257,6 +257,45 @@ class OpenAI extends BaseLLM<OpenAIOptions> {
     return completion.toLLMResult();
   }
 
+  @override
+  Stream<LLMResult> stream(
+    final PromptValue input, {
+    final OpenAIOptions? options,
+  }) {
+    return _client
+        .createCompletionStream(
+          request: CreateCompletionRequest(
+            model: CompletionModel.string(model),
+            prompt: CompletionPrompt.string(input.toString()),
+            bestOf: bestOf,
+            frequencyPenalty: frequencyPenalty,
+            logitBias: logitBias,
+            logprobs: logprobs,
+            maxTokens: maxTokens,
+            n: n,
+            presencePenalty: presencePenalty,
+            stop: options?.stop != null
+                ? CompletionStop.arrayString(options!.stop!)
+                : null,
+            suffix: suffix,
+            temperature: temperature,
+            topP: topP,
+            user: options?.user ?? user,
+          ),
+        )
+        .map((final completion) => completion.toLLMResult(streaming: true));
+  }
+
+  @override
+  Stream<LLMResult> streamFromInputStream(
+    final Stream<PromptValue> inputStream, {
+    final OpenAIOptions? options,
+  }) {
+    return inputStream.asyncExpand((final input) {
+      return stream(input, options: options);
+    });
+  }
+
   /// Tokenizes the given prompt using tiktoken with the encoding used by the
   /// [model]. If an encoding model is specified in [encoding] field, that
   /// encoding is used instead.

diff --git a/packages/langchain_openai/test/llms/openai_test.dart b/packages/langchain_openai/test/llms/openai_test.dart
@@ -67,36 +67,57 @@ void main() {
     });
 
     test('Test OpenAI wrapper with multiple completions', () async {
-      final chat = OpenAI(apiKey: openaiApiKey, n: 5, bestOf: 5);
-      final res = await chat.generate('Hello, how are you?');
+      final llm = OpenAI(apiKey: openaiApiKey, n: 5, bestOf: 5);
+      final res = await llm.generate('Hello, how are you?');
       expect(res.generations.length, 5);
       for (final generation in res.generations) {
         expect(generation.output, isNotEmpty);
       }
     });
 
     test('Test tokenize', () async {
-      final chat = OpenAI(apiKey: openaiApiKey);
+      final llm = OpenAI(apiKey: openaiApiKey);
       const text = 'Hello, how are you?';
 
-      final tokens = await chat.tokenize(PromptValue.string(text));
+      final tokens = await llm.tokenize(PromptValue.string(text));
       expect(tokens, [15496, 11, 703, 389, 345, 30]);
     });
 
     test('Test different encoding than the model', () async {
-      final chat = OpenAI(apiKey: openaiApiKey, encoding: 'cl100k_base');
+      final llm = OpenAI(apiKey: openaiApiKey, encoding: 'cl100k_base');
       const text = 'Hello, how are you?';
 
-      final tokens = await chat.tokenize(PromptValue.string(text));
+      final tokens = await llm.tokenize(PromptValue.string(text));
       expect(tokens, [9906, 11, 1268, 527, 499, 30]);
     });
 
     test('Test countTokens', () async {
-      final chat = OpenAI(apiKey: openaiApiKey);
+      final llm = OpenAI(apiKey: openaiApiKey);
       const text = 'Hello, how are you?';
 
-      final numTokens = await chat.countTokens(PromptValue.string(text));
+      final numTokens = await llm.countTokens(PromptValue.string(text));
       expect(numTokens, 6);
     });
+
+    test('Test streaming', () async {
+      final promptTemplate = PromptTemplate.fromTemplate(
+        'List the numbers from 1 to {max_num} in order without any spaces or commas',
+      );
+      final llm = OpenAI(apiKey: openaiApiKey);
+      const stringOutputParser = StringOutputParser<String>();
+
+      final chain = promptTemplate.pipe(llm).pipe(stringOutputParser);
+
+      final stream = chain.stream({'max_num': '9'});
+
+      String content = '';
+      int count = 0;
+      await for (final res in stream) {
+        content += res;
+        count++;
+      }
+      expect(count, greaterThan(1));
+      expect(content, contains('123456789'));
+    });
   });
 }