Resolve the issue of the AWS Bedrock chat model always returning nil (#801)

* Fix issue the AWS Bedrock chat model always `nil`

* Add `chat_model` arg for `Langchain::LLM::AwsBedrock#initialize`

* Fix the AwsBedrock constructor

* CHANGELOG entry

---------

Co-authored-by: Andrei Bondarev <[email protected]>
Co-authored-by: Andrei Bondarev <[email protected]>

Author: khaile

CHANGELOG.md

@@ -1,4 +1,5 @@
 ## [Unreleased]
+- [BREAKING] Modify `Langchain::LLM::AwsBedrock` constructor to pass model options via default_options: {...}
 - Added support for streaming with Anthropic
 - Bump anthropic gem
 - Default Langchain::LLM::Anthropic chat model is "claude-3-5-sonnet-20240620" now

lib/langchain/llm/aws_bedrock.rb

@@ -11,8 +11,9 @@ module Langchain::LLM
   #
   class AwsBedrock < Base
     DEFAULTS = {
+      chat_completion_model_name: "anthropic.claude-v2",
       completion_model_name: "anthropic.claude-v2",
-      embedding_model_name: "amazon.titan-embed-text-v1",
+      embeddings_model_name: "amazon.titan-embed-text-v1",
       max_tokens_to_sample: 300,
       temperature: 1,
       top_k: 250,
@@ -52,13 +53,11 @@ class AwsBedrock < Base
     SUPPORTED_CHAT_COMPLETION_PROVIDERS = %i[anthropic].freeze
     SUPPORTED_EMBEDDING_PROVIDERS = %i[amazon cohere].freeze
 
-    def initialize(completion_model: DEFAULTS[:completion_model_name], embedding_model: DEFAULTS[:embedding_model_name], aws_client_options: {}, default_options: {})
+    def initialize(aws_client_options: {}, default_options: {})
       depends_on "aws-sdk-bedrockruntime", req: "aws-sdk-bedrockruntime"
 
       @client = ::Aws::BedrockRuntime::Client.new(**aws_client_options)
       @defaults = DEFAULTS.merge(default_options)
-        .merge(completion_model_name: completion_model)
-        .merge(embedding_model_name: embedding_model)
 
       chat_parameters.update(
         model: {default: @defaults[:chat_completion_model_name]},
@@ -85,7 +84,7 @@ def embed(text:, **params)
       parameters = compose_embedding_parameters params.merge(text:)
 
       response = client.invoke_model({
-        model_id: @defaults[:embedding_model_name],
+        model_id: @defaults[:embeddings_model_name],
         body: parameters.to_json,
         content_type: "application/json",
         accept: "application/json"
@@ -180,7 +179,7 @@ def completion_provider
     end
 
     def embedding_provider
-      @defaults[:embedding_model_name].split(".").first.to_sym
+      @defaults[:embeddings_model_name].split(".").first.to_sym
     end
 
     def wrap_prompt(prompt)

spec/langchain/llm/aws_bedrock_spec.rb

@@ -24,12 +24,14 @@
         }.to_json
       end
 
+      let(:model_id) { "anthropic.claude-3-sonnet-20240229-v1:0" }
+
       before do
         response_object = double("response_object")
         allow(response_object).to receive(:body).and_return(StringIO.new(response))
         allow(subject.client).to receive(:invoke_model)
           .with(matching(
-            model_id: "anthropic.claude-3-sonnet-20240229-v1:0",
+            model_id:,
             body: {
               messages: [{role: "user", content: "What is the capital of France?"}],
               stop_sequences: ["stop"],
@@ -46,12 +48,25 @@
         expect(
           subject.chat(
             messages: [{role: "user", content: "What is the capital of France?"}],
-            model: "anthropic.claude-3-sonnet-20240229-v1:0",
+            model: model_id,
             stop_sequences: ["stop"]
           ).chat_completion
         ).to eq("The capital of France is Paris.")
       end
 
+      context "without default model" do
+        let(:model_id) { "anthropic.claude-v2" }
+
+        it "returns a completion" do
+          expect(
+            subject.chat(
+              messages: [{role: "user", content: "What is the capital of France?"}],
+              stop_sequences: ["stop"]
+            ).chat_completion
+          ).to eq("The capital of France is Paris.")
+        end
+      end
+
       context "with streaming" do
         let(:chunks) do
           [
@@ -201,7 +216,7 @@
     end
 
     context "with ai21 provider" do
-      let(:subject) { described_class.new(completion_model: "ai21.j2-ultra-v1") }
+      let(:subject) { described_class.new(default_options: {completion_model_name: "ai21.j2-ultra-v1"}) }
 
       let(:response) do
         StringIO.new("{\"completions\":[{\"data\":{\"text\":\"\\nWhat is the meaning of life? What is the meaning of life?\\nWhat is the meaning\"}}]}")
@@ -308,8 +323,11 @@
       context "with custom default_options" do
         let(:subject) {
           described_class.new(
-            completion_model: "ai21.j2-ultra-v1",
-            default_options: {max_tokens_to_sample: 100, temperature: 0.7}
+            default_options: {
+              completion_model_name: "ai21.j2-ultra-v1",
+              max_tokens_to_sample: 100,
+              temperature: 0.7
+            }
           )
         }
         let(:response_object) { double("response_object") }
@@ -363,7 +381,7 @@
     end
 
     context "with cohere provider" do
-      let(:subject) { described_class.new(completion_model: "cohere.command-text-v14") }
+      let(:subject) { described_class.new(default_options: {completion_model_name: "cohere.command-text-v14"}) }
 
       let(:response) do
         StringIO.new("{\"generations\":[{\"text\":\"\\nWhat is the meaning of life? What is the meaning of life?\\nWhat is the meaning\"}]}")
@@ -424,8 +442,11 @@
       context "with custom default_options" do
         let(:subject) {
           described_class.new(
-            completion_model: "cohere.command-text-v14",
-            default_options: {max_tokens_to_sample: 100, temperature: 0.7}
+            default_options: {
+              completion_model_name: "cohere.command-text-v14",
+              max_tokens_to_sample: 100,
+              temperature: 0.7
+            }
           )
         }
         let(:response_object) { double("response_object") }
@@ -456,7 +477,7 @@
     end
 
     context "with unsupported provider" do
-      let(:subject) { described_class.new(completion_model: "unsupported.provider") }
+      let(:subject) { described_class.new(default_options: {completion_model_name: "unsupported.provider"}) }
 
       it "raises an exception" do
         expect { subject.complete(prompt: "Hello World") }.to raise_error("Completion provider unsupported is not supported.")
@@ -492,7 +513,7 @@
     end
 
     context "with cohere provider" do
-      let(:subject) { described_class.new(embedding_model: "cohere.embed-multilingual-v3") }
+      let(:subject) { described_class.new(default_options: {embeddings_model_name: "cohere.embed-multilingual-v3"}) }
 
       let(:response) do
         StringIO.new("{\"embeddings\":[[0.1,0.2,0.3,0.4,0.5]]}")
@@ -522,7 +543,7 @@
     end
 
     context "with unsupported provider" do
-      let(:subject) { described_class.new(embedding_model: "unsupported.provider") }
+      let(:subject) { described_class.new(default_options: {embeddings_model_name: "unsupported.provider"}) }
 
       it "raises an exception" do
         expect { subject.embed(text: "Hello World") }.to raise_error("Completion provider unsupported is not supported.")