feat: Support Anyscale in ChatOpenAI and OpenAIEmbeddings wrappers (#305

)

README.md

@@ -0,0 +1 @@
+./packages/langchain/README.md

docs/_sidebar.md

@@ -53,6 +53,7 @@
           - [Mistral AI](/modules/model_io/models/chat_models/integrations/mistralai.md)
           - [OpenRouter](/modules/model_io/models/chat_models/integrations/open_router.md)
           - [Together AI](/modules/model_io/models/chat_models/integrations/together_ai.md)
+          - [Anyscale](/modules/model_io/models/chat_models/integrations/anyscale.md)
           - [Prem App](/modules/model_io/models/chat_models/integrations/prem.md)
     - [Output parsers](/modules/model_io/output_parsers/output_parsers.md)
       - [String output parser](/modules/model_io/output_parsers/string.md)
@@ -75,6 +76,7 @@
         - [Ollama](/modules/retrieval/text_embedding/integrations/ollama.md)
         - [Mistral AI](/modules/retrieval/text_embedding/integrations/mistralai.md)
         - [Together AI](/modules/retrieval/text_embedding/integrations/together_ai.md)
+        - [Anyscale](/modules/retrieval/text_embedding/integrations/anyscale.md)
         - [Prem App](/modules/retrieval/text_embedding/integrations/prem.md)
     - [Vector stores](/modules/retrieval/vector_stores/vector_stores.md)
       - Integrations

docs/modules/model_io/models/chat_models/integrations/anyscale.md

@@ -0,0 +1,84 @@
+# Anyscale
+
+[Anyscale](https://www.anyscale.com/) offers a unified OpenAI-compatible API for a broad range of [models](https://docs.endpoints.anyscale.com/guides/models/#chat-models) running serverless or on your own dedicated instances.
+
+It also allows to fine-tune models on your own data or train new models from scratch.
+
+You can consume Anyscale API using the `ChatOpenAI` wrapper in the same way you would use the OpenAI API.
+
+The only difference is that you need to change the base URL to `https://api.endpoints.anyscale.com/v1`:
+
+```dart
+final chatModel = ChatOpenAI(
+  apiKey: togetherAiApiKey,
+  baseUrl: 'https://api.endpoints.anyscale.com/v1',
+  defaultOptions: const ChatOpenAIOptions(
+    model: 'meta-llama/Llama-2-70b-chat-hf',
+  ),
+);
+```
+
+## Invoke
+
+```dart
+final anyscaleApiKey = Platform.environment['ANYSCALE_API_KEY'];
+
+final promptTemplate = ChatPromptTemplate.fromTemplates(const [
+  (
+  ChatMessageType.system,
+  'You are a helpful assistant that translates {input_language} to {output_language}.',
+  ),
+  (ChatMessageType.human, '{text}'),
+]);
+
+final chatModel = ChatOpenAI(
+  apiKey: anyscaleApiKey,
+  baseUrl: 'https://api.endpoints.anyscale.com/v1',
+  defaultOptions: const ChatOpenAIOptions(
+    model: 'meta-llama/Llama-2-70b-chat-hf',
+  ),
+);
+
+final chain = promptTemplate | chatModel | const StringOutputParser();
+
+final res = await chain.invoke({
+  'input_language': 'English',
+  'output_language': 'French',
+  'text': 'I love programming.',
+});
+print(res);
+// -> "I love programming" se traduit en français sous la forme "J'aime passionnément la programmation"
+```
+
+## Stream
+
+```dart
+final anyscaleApiKey = Platform.environment['ANYSCALE_API_KEY'];
+
+final promptTemplate = ChatPromptTemplate.fromTemplates(const [
+  (
+  ChatMessageType.system,
+  'You are a helpful assistant that replies only with numbers '
+      'in order without any spaces or commas',
+  ),
+  (ChatMessageType.human, 'List the numbers from 1 to {max_num}'),
+]);
+
+final chatModel = ChatOpenAI(
+  apiKey: anyscaleApiKey,
+  baseUrl: 'https://api.endpoints.anyscale.com/v1',
+  defaultOptions: const ChatOpenAIOptions(
+    model: 'mistralai/Mixtral-8x7B-Instruct-v0.1',
+  ),
+);
+
+final chain = promptTemplate.pipe(chatModel).pipe(const StringOutputParser());
+
+final stream = chain.stream({'max_num': '9'});
+await stream.forEach(print);
+// 1
+// 2
+// 3
+// ...
+// 9
+```

docs/modules/retrieval/text_embedding/integrations/anyscale.md

@@ -0,0 +1,30 @@
+# Anyscale Embeddings
+
+[Anyscale](https://www.anyscale.com/) offers several [embedding models](https://docs.endpoints.anyscale.com/guides/models/#embedding-models) through its OpenAI compatible API.
+
+You can consume Anyscale API using the `OpenAIEmbeddings` wrapper in the same way you would use the OpenAI API.
+
+The only difference is that you need to change the base URL to `https://api.endpoints.anyscale.com/v1`:
+
+```dart
+final anyscaleApiKey = Platform.environment['ANYSCALE_API_KEY'];
+final embeddings = OpenAIEmbeddings(
+  apiKey: anyscaleApiKey,
+  baseUrl: 'https://api.endpoints.anyscale.com/v1',
+  model: 'thenlper/gte-large',
+);
+
+// Embedding a document
+const doc = Document(pageContent: 'This is a test document.');
+final res1 = await embeddings.embedDocuments([doc]);
+print(res1);
+// [[-0.0011281073093414307, -0.013280618004500866, 0.02164546772837639, ...]]
+
+// Embedding a retrieval query
+const text = 'This is a test query.';
+final res2 = await embeddings.embedQuery(text);
+print(res2);
+// [-0.027850965037941933, 0.00269310618750751, 0.008118202909827232, ...]
+
+embeddings.close();
+```