Ability to pass a system prompt for chat completion

lib/langchain/vectorsearch/base.rb

@@ -168,6 +168,27 @@ def generate_hyde_prompt(question:)
       prompt_template.format(question: question)
     end
 
+    # Ask a question and return the answer
+    # @param question [String] The question to ask
+    # @param context [Array] The context generated by RAG
+    # @param system_prompt [String] Content of the prompt to send as "system"
+    # @yield [String] Stream responses back one String at a time
+    # @return [String] The answer to the question
+    def generate_messages_and_chat(question: , context: , system_prompt: nil, &block)
+      context = context.join("\n---\n")
+
+      prompt = generate_rag_prompt(question: question, context: context)
+
+      messages = [
+        system_prompt ? {role: 'system', content: system_prompt} : nil,
+        {role: "user", content: prompt}
+      ].compact
+      response = llm.chat(messages: messages, &block)
+
+      response.context = context
+      response
+    end
+
     # Retrieval Augmented Generation (RAG)
     #
     # @param question [String] User's question

lib/langchain/vectorsearch/chroma.rb

@@ -123,24 +123,17 @@ def similarity_search_by_vector(
     # Ask a question and return the answer
     # @param question [String] The question to ask
     # @param k [Integer] The number of results to have in context
+    # @param system_prompt [String] Content of the prompt to send as "system"
     # @yield [String] Stream responses back one String at a time
     # @return [String] The answer to the question
-    def ask(question:, k: 4, &block)
+    def ask(question:, k: 4, system_prompt: nil, &block)
       search_results = similarity_search(query: question, k: k)
 
       context = search_results.map do |result|
         result.document
       end
 
-      context = context.join("\n---\n")
-
-      prompt = generate_rag_prompt(question: question, context: context)
-
-      messages = [{role: "user", content: prompt}]
-      response = llm.chat(messages: messages, &block)
-
-      response.context = context
-      response
+      generate_messages_and_chat(question: question, context: context, system_prompt: system_prompt, &block)
     end
 
     private

lib/langchain/vectorsearch/elasticsearch.rb

@@ -141,22 +141,17 @@ def default_query(query_vector)
     # Ask a question and return the answer
     # @param question [String] The question to ask
     # @param k [Integer] The number of results to have in context
+    # @param system_prompt [String] Content of the prompt to send as "system"
     # @yield [String] Stream responses back one String at a time
     # @return [String] The answer to the question
-    def ask(question:, k: 4, &block)
+    def ask(question:, k: 4, system_prompt: nil, &block)
       search_results = similarity_search(query: question, k: k)
 
       context = search_results.map do |result|
         result[:input]
-      end.join("\n---\n")
-
-      prompt = generate_rag_prompt(question: question, context: context)
-
-      messages = [{role: "user", content: prompt}]
-      response = llm.chat(messages: messages, &block)
+      end
 
-      response.context = context
-      response
+      generate_messages_and_chat(question: question, context: context, system_prompt: system_prompt, &block)
     end
 
     # Search for similar texts

lib/langchain/vectorsearch/epsilla.rb

@@ -125,23 +125,17 @@ def similarity_search_by_vector(embedding:, k: 4)
     # Ask a question and return the answer
     # @param question [String] The question to ask
     # @param k [Integer] The number of results to have in context
+    # @param system_prompt [String] Content of the prompt to send as "system"
     # @yield [String] Stream responses back one String at a time
     # @return [String] The answer to the question
-    def ask(question:, k: 4, &block)
+    def ask(question:, k: 4, system_prompt: nil, &block)
       search_results = similarity_search(query: question, k: k)
 
       context = search_results.map do |result|
         result.to_s
       end
-      context = context.join("\n---\n")
 
-      prompt = generate_rag_prompt(question: question, context: context)
-
-      messages = [{role: "user", content: prompt}]
-      response = llm.chat(messages: messages, &block)
-
-      response.context = context
-      response
+      generate_messages_and_chat(question: question, context: context, system_prompt: system_prompt, &block)
     end
   end
 end

lib/langchain/vectorsearch/milvus.rb

@@ -139,22 +139,15 @@ def similarity_search_by_vector(embedding:, k: 4)
     # Ask a question and return the answer
     # @param question [String] The question to ask
     # @param k [Integer] The number of results to have in context
+    # @param system_prompt [String] Content of the prompt to send as "system"
     # @yield [String] Stream responses back one String at a time
     # @return [String] The answer to the question
-    def ask(question:, k: 4, &block)
+    def ask(question:, k: 4, system_prompt: nil, &block)
       search_results = similarity_search(query: question, k: k)
 
-      content_data = search_results.dig("data").map { |result| result.dig("content") }
+      context = search_results.dig("data").map { |result| result.dig("content") }
 
-      context = content_data.join("\n---\n")
-
-      prompt = generate_rag_prompt(question: question, context: context)
-
-      messages = [{role: "user", content: prompt}]
-      response = llm.chat(messages: messages, &block)
-
-      response.context = context
-      response
+      generate_messages_and_chat(question: question, context: context, system_prompt: system_prompt, &block)
     end
   end
 end

lib/langchain/vectorsearch/pgvector.rb

@@ -144,23 +144,17 @@ def similarity_search_by_vector(embedding:, k: 4)
     # Ask a question and return the answer
     # @param question [String] The question to ask
     # @param k [Integer] The number of results to have in context
+    # @param system_prompt [String] Content of the prompt to send as "system"
     # @yield [String] Stream responses back one String at a time
     # @return [String] The answer to the question
-    def ask(question:, k: 4, &block)
+    def ask(question:, k: 4, system_prompt: nil, &block)
       search_results = similarity_search(query: question, k: k)
 
       context = search_results.map do |result|
         result.content.to_s
       end
-      context = context.join("\n---\n")
 
-      prompt = generate_rag_prompt(question: question, context: context)
-
-      messages = [{role: "user", content: prompt}]
-      response = llm.chat(messages: messages, &block)
-
-      response.context = context
-      response
+      generate_messages_and_chat(question: question, context: context, system_prompt: system_prompt, &block)
     end
   end
 end

lib/langchain/vectorsearch/pinecone.rb

@@ -168,24 +168,18 @@ def similarity_search_by_vector(embedding:, k: 4, namespace: "", filter: nil)
     # @param question [String] The question to ask
     # @param namespace [String] The namespace to search in
     # @param k [Integer] The number of results to have in context
+    # @param system_prompt [String] Content of the prompt to send as "system"
     # @param filter [String] The filter to use
     # @yield [String] Stream responses back one String at a time
     # @return [String] The answer to the question
-    def ask(question:, namespace: "", filter: nil, k: 4, &block)
+    def ask(question:, namespace: "", filter: nil, k: 4, system_prompt: nil, &block)
       search_results = similarity_search(query: question, namespace: namespace, filter: filter, k: k)
 
       context = search_results.map do |result|
         result.dig("metadata").to_s
       end
-      context = context.join("\n---\n")
 
-      prompt = generate_rag_prompt(question: question, context: context)
-
-      messages = [{role: "user", content: prompt}]
-      response = llm.chat(messages: messages, &block)
-
-      response.context = context
-      response
+      generate_messages_and_chat(question: question, context: context, system_prompt: system_prompt, &block)
     end
 
     # Pinecone index

lib/langchain/vectorsearch/qdrant.rb

@@ -136,23 +136,17 @@ def similarity_search_by_vector(
     # Ask a question and return the answer
     # @param question [String] The question to ask
     # @param k [Integer] The number of results to have in context
+    # @param system_prompt [String] Content of the prompt to send as "system"
     # @yield [String] Stream responses back one String at a time
     # @return [String] The answer to the question
-    def ask(question:, k: 4, &block)
+    def ask(question:, k: 4, system_prompt: nil, &block)
       search_results = similarity_search(query: question, k: k)
 
       context = search_results.map do |result|
         result.dig("payload").to_s
       end
-      context = context.join("\n---\n")
 
-      prompt = generate_rag_prompt(question: question, context: context)
-
-      messages = [{role: "user", content: prompt}]
-      response = llm.chat(messages: messages, &block)
-
-      response.context = context
-      response
+      generate_messages_and_chat(question: question, context: context, system_prompt: system_prompt, &block)
     end
   end
 end

lib/langchain/vectorsearch/weaviate.rb

@@ -142,23 +142,17 @@ def similarity_search_by_vector(embedding:, k: 4)
     # Ask a question and return the answer
     # @param question [String] The question to ask
     # @param k [Integer] The number of results to have in context
+    # @param system_prompt [String] Content of the prompt to send as "system"
     # @yield [String] Stream responses back one String at a time
     # @return [Hash] The answer
-    def ask(question:, k: 4, &block)
+    def ask(question:, k: 4, system_prompt: nil, &block)
       search_results = similarity_search(query: question, k: k)
 
       context = search_results.map do |result|
         result.dig("content").to_s
       end
-      context = context.join("\n---\n")
 
-      prompt = generate_rag_prompt(question: question, context: context)
-
-      messages = [{role: "user", content: prompt}]
-      response = llm.chat(messages: messages, &block)
-
-      response.context = context
-      response
+      generate_messages_and_chat(question: question, context: context, system_prompt: system_prompt, &block)
     end
 
     private

spec/langchain/vectorsearch/chroma_spec.rb

@@ -166,6 +166,18 @@
       end
     end
 
+    context 'with system prompt' do
+      let(:system_prompt) { 'You are a helpful assistant' }
+      before do
+        allow(subject.llm).to receive(:chat).with(messages: [{role: 'system', content: system_prompt}, *messages]).and_return(response)
+        expect(response).to receive(:context=).with(text)
+      end
+
+      it "asks a question and returns the answer" do
+        expect(subject.ask(question: question, system_prompt: system_prompt).completion).to eq(answer)
+      end
+    end
+
     context "with block" do
       let(:block) { proc { |chunk| puts "Received chunk: #{chunk}" } }
 

spec/langchain/vectorsearch/elasticsearch_spec.rb

@@ -274,6 +274,18 @@
       end
     end
 
+    context 'with system prompt' do
+      let(:system_prompt) { 'You are a helpful assistant' }
+      before do
+        allow(subject.llm).to receive(:chat).with(messages: [{role: 'system', content: system_prompt}, *messages]).and_return(response)
+        expect(response).to receive(:context=).with(text)
+      end
+
+      it "asks a question and returns the answer" do
+        expect(subject.ask(question: question, k: 4, system_prompt: system_prompt).completion).to eq(answer)
+      end
+    end
+
     context "with block" do
       let(:block) { proc { |chunk| puts "Received chunk: #{chunk}" } }
 

spec/langchain/vectorsearch/epsilla_spec.rb

@@ -236,6 +236,18 @@
         end
       end
 
+      context 'with system prompt' do
+        let(:system_prompt) { 'You are a helpful assistant' }
+        before do
+          allow(subject.llm).to receive(:chat).with(messages: [{role: 'system', content: system_prompt}, *messages]).and_return(response)
+          expect(response).to receive(:context=).with(text)
+        end
+
+        it "asks a question and returns the answer" do
+          expect(subject.ask(question: question, k: k, system_prompt: system_prompt).completion).to eq(answer)
+        end
+      end
+
       context "with block" do
         let(:block) { proc { |chunk| puts "Received chunk: #{chunk}" } }
 

spec/langchain/vectorsearch/milvus_spec.rb

@@ -141,6 +141,18 @@
       end
     end
 
+    context 'with system prompt' do
+      let(:system_prompt) { 'You are a helpful assistant' }
+      before do
+        allow(subject.llm).to receive(:chat).with(messages: [{role: 'system', content: system_prompt}, *messages]).and_return(response)
+        expect(response).to receive(:context=).with(text)
+      end
+
+      it "asks a question and returns the answer" do
+        expect(subject.ask(question: question, system_prompt: system_prompt).completion).to eq(answer)
+      end
+    end
+
     context "with block" do
       let(:block) { proc { |chunk| puts "Received chunk: #{chunk}" } }
 

spec/langchain/vectorsearch/pgvector_spec.rb

@@ -234,6 +234,17 @@
         end
       end
 
+      context 'with system prompt' do
+        before do
+          allow(subject.llm).to receive(:chat).with(messages: messages).and_return(response)
+          expect(response).to receive(:context=).with(text)
+        end
+
+        it "asks a question and returns the answer" do
+          expect(subject.ask(question: question, k: k, system_prompt: 'You are a helpful assistant').completion).to eq(answer)
+        end
+      end
+
       context "with block" do
         let(:block) { proc { |chunk| puts "Received chunk: #{chunk}" } }
 

spec/langchain/vectorsearch/pinecone_spec.rb

@@ -410,6 +410,22 @@
       end
     end
 
+    describe "with system prompt" do
+      let(:system_prompt) { 'You are a helpful assistant' }
+
+      before do
+        allow(subject).to receive(:similarity_search).with(
+          query: question, namespace: "", filter: nil, k: k
+        ).and_return(matches)
+        allow(subject.llm).to receive(:chat).with(messages: [{role: 'system', content: system_prompt}, *messages]).and_return(response)
+        expect(response).to receive(:context=).with(metadata.to_s)
+      end
+
+      it "asks a question" do
+        expect(subject.ask(question: question, system_prompt: system_prompt).completion).to eq(answer)
+      end
+    end
+
     describe "with block" do
       let(:block) { proc { |chunk| puts "Received chunk: #{chunk}" } }
 

spec/langchain/vectorsearch/qdrant_spec.rb

@@ -157,6 +157,18 @@
       end
     end
 
+    context 'with system prompt' do
+      let(:system_prompt) { 'You are a helpful assistant' }
+      before do
+        allow(subject.llm).to receive(:chat).with(messages: [{role: 'system', content: system_prompt}, *messages]).and_return(response)
+        expect(response).to receive(:context=).with(text)
+      end
+
+      it "asks a question and returns the answer" do
+        expect(subject.ask(question: question, k: k, system_prompt: system_prompt).completion).to eq(answer)
+      end
+    end
+
     context "with block" do
       let(:block) { proc { |chunk| puts "Received chunk: #{chunk}" } }