Ollama fix handle incomplete JSON chunks in stream

lib/langchain/llm/ollama.rb

@@ -287,8 +287,16 @@ def auth_headers
     end
 
     def json_responses_chunk_handler(&block)
+      incomplete_chunk_line = nil
       proc do |chunk, _size|
         chunk.split("\n").each do |chunk_line|
+          if incomplete_chunk_line
+            chunk_line = incomplete_chunk_line + chunk_line
+            incomplete_chunk_line = nil
+          end
+
+          next incomplete_chunk_line = chunk_line unless chunk_line.end_with?("}")
+
           parsed_chunk = JSON.parse(chunk_line)
           block.call(parsed_chunk)
         end

spec/lib/langchain/llm/ollama_spec.rb

@@ -87,6 +87,11 @@
       expect(response.completion).to eq("Complicated.")
     end
 
+    it "handles multiline json", :vcr do
+      expect { response }.not_to raise_error
+      expect(response.completion).to eq("Complicated.")
+    end
+
     it "does not use streamed responses", vcr: {cassette_name: "Langchain_LLM_Ollama_complete_returns_a_completion"} do
       expect(client).to receive(:post).with("api/generate", hash_including(stream: false)).and_call_original
       response
@@ -102,6 +107,11 @@
         expect(response.total_tokens).to eq(36)
       end
 
+      it "handles multiline json", :vcr do
+        expect { response }.not_to raise_error
+        expect(response.completion).to eq("Complicated.")
+      end
+
       it "uses streamed responses", vcr: {cassette_name: "Langchain_LLM_Ollama_complete_when_passing_a_block_returns_a_completion"} do
         expect(client).to receive(:post).with("api/generate", hash_including(stream: true)).and_call_original
         response
@@ -125,6 +135,11 @@
       expect(response.chat_completion).to include("I'm just a language model")
     end
 
+    it "handles multiline json", :vcr do
+      expect { response }.not_to raise_error
+      expect(response.chat_completion).to include("I'm just a language model")
+    end
+
     it "does not use streamed responses", vcr: {cassette_name: "Langchain_LLM_Ollama_chat_returns_a_chat_completion"} do
       expect(client).to receive(:post).with("api/chat", hash_including(stream: false)).and_call_original
       response
@@ -139,6 +154,11 @@
         expect(response.chat_completion).to include("I'm just a language model")
       end
 
+      it "handles multiline json", :vcr do
+        expect { response }.not_to raise_error
+        expect(response.chat_completion).to include("I'm just a language model")
+      end
+
       it "uses streamed responses", vcr: {cassette_name: "Langchain_LLM_Ollama_chat_when_passing_a_block_returns_a_chat_completion"} do
         expect(client).to receive(:post).with("api/chat", hash_including(stream: true)).and_call_original
         response