Move yield of metrics chunk after generation chunk (#216)

## Move yield of metrics chunk after generation chunk
- when using mistral and streaming is enabled,the final chunk includes a
stop_reason. There is nothing to say this final chunk doesn't also
include some generated text. The existing implementation would result in
that final chunk never getting sent back
- this update moves the yield of the metrics chunk after the generation
chunk
- also included a change to include invocation metrics for cohere models

Closes #215

Author: ihmaws

libs/aws/langchain_aws/llms/bedrock.py

@@ -369,7 +369,17 @@ def prepare_output_stream(
             ):
                 return
 
-            elif (
+            generation_chunk = _stream_response_to_generation_chunk(
+                chunk_obj,
+                provider=provider,
+                output_key=output_key,
+                messages_api=messages_api,
+                coerce_content_to_string=coerce_content_to_string,
+            )
+            if generation_chunk:
+                yield generation_chunk
+
+            if (
                 provider == "mistral"
                 and chunk_obj.get(output_key, [{}])[0].get("stop_reason", "") == "stop"
             ):
@@ -384,18 +394,6 @@ def prepare_output_stream(
                 yield _get_invocation_metrics_chunk(chunk_obj)
                 return
 
-            generation_chunk = _stream_response_to_generation_chunk(
-                chunk_obj,
-                provider=provider,
-                output_key=output_key,
-                messages_api=messages_api,
-                coerce_content_to_string=coerce_content_to_string,
-            )
-            if generation_chunk:
-                yield generation_chunk
-            else:
-                continue
-
     @classmethod
     async def aprepare_output_stream(
         cls,

libs/aws/tests/unit_tests/llms/test_bedrock.py

@@ -270,6 +270,11 @@ def test__human_assistant_format() -> None:
     {"chunk": {"bytes": b'{"text": " you"}'}},
 ]
 
+MOCK_STREAMING_RESPONSE_MISTRAL = [
+    {"chunk": {"bytes": b'{"outputs": [{"text": "Thank","stop_reason": null}]}'}},
+    {"chunk": {"bytes": b'{"outputs": [{"text": "you.","stop_reason": "stop"}]}'}},
+]
+
 
 async def async_gen_mock_streaming_response() -> AsyncGenerator[Dict, None]:
     for item in MOCK_STREAMING_RESPONSE:
@@ -331,6 +336,12 @@ def mistral_response():
     return response
 
 
+@pytest.fixture
+def mistral_streaming_response():
+    response = dict(body=MOCK_STREAMING_RESPONSE_MISTRAL)
+    return response
+
+
 @pytest.fixture
 def cohere_response():
     body = MagicMock()
@@ -412,6 +423,18 @@ def test_prepare_output_for_mistral(mistral_response):
     assert result["stop_reason"] is None
 
 
+def test_prepare_output_stream_for_mistral(mistral_streaming_response) -> None:
+    results = [
+        chunk.text
+        for chunk in LLMInputOutputAdapter.prepare_output_stream(
+            "mistral", mistral_streaming_response
+        )
+    ]
+
+    assert results[0] == "Thank"
+    assert results[1] == "you."
+
+
 def test_prepare_output_for_cohere(cohere_response):
     result = LLMInputOutputAdapter.prepare_output("cohere", cohere_response)
     assert result["text"] == "This is the Cohere output text."