feat: support variation param for model and prompt

learning_assistant/api.py

@@ -103,7 +103,7 @@ def get_block_content(request, user_id, course_id, unit_usage_key):
     return cache_data['content_length'], cache_data['content_items']
 
 
-def render_prompt_template(request, user_id, course_run_id, unit_usage_key, course_id):
+def render_prompt_template(request, user_id, course_run_id, unit_usage_key, course_id, template_string):
     """
     Return a rendered prompt template, specified by the LEARNING_ASSISTANT_PROMPT_TEMPLATE setting.
     """
@@ -117,7 +117,6 @@ def render_prompt_template(request, user_id, course_run_id, unit_usage_key, cour
     skill_names = course_data['skill_names']
     title = course_data['title']
 
-    template_string = getattr(settings, 'LEARNING_ASSISTANT_PROMPT_TEMPLATE', '')
     template = Environment(loader=BaseLoader).from_string(template_string)
     data = template.render(unit_content=unit_content, skill_names=skill_names, title=title)
     return data

learning_assistant/constants.py

@@ -14,3 +14,12 @@
     "html": "TEXT",
     "video": "VIDEO",
 }
+
+
+class GptModels:
+    GPT_3_5_TURBO = 'gpt-3.5-turbo'
+    GPT_4o = 'gpt-4o'
+
+
+class ResponseVariations:
+    GPT4_UPDATED_PROMPT = 'updated_prompt'

learning_assistant/utils.py

@@ -52,18 +52,19 @@ def get_reduced_message_list(prompt_template, message_list):
     return [system_message] + new_message_list
 
 
-def create_request_body(prompt_template, message_list):
+def create_request_body(prompt_template, message_list, gpt_model):
     """
     Form request body to be passed to the chat endpoint.
     """
     response_body = {
-        'message_list': get_reduced_message_list(prompt_template, message_list)
+        'message_list': get_reduced_message_list(prompt_template, message_list),
+        'model': gpt_model,
     }
 
     return response_body
 
 
-def get_chat_response(prompt_template, message_list):
+def get_chat_response(prompt_template, message_list, gpt_model):
     """
     Pass message list to chat endpoint, as defined by the CHAT_COMPLETION_API setting.
     """
@@ -74,7 +75,7 @@ def get_chat_response(prompt_template, message_list):
         connect_timeout = getattr(settings, 'CHAT_COMPLETION_API_CONNECT_TIMEOUT', 1)
         read_timeout = getattr(settings, 'CHAT_COMPLETION_API_READ_TIMEOUT', 15)
 
-        body = create_request_body(prompt_template, message_list)
+        body = create_request_body(prompt_template, message_list, gpt_model)
 
         try:
             response = requests.post(

learning_assistant/views.py

@@ -3,6 +3,7 @@
 """
 import logging
 
+from django.conf import settings
 from edx_rest_framework_extensions.auth.jwt.authentication import JwtAuthentication
 from opaque_keys import InvalidKeyError
 from opaque_keys.edx.keys import CourseKey
@@ -20,6 +21,7 @@
     pass
 
 from learning_assistant.api import get_course_id, learning_assistant_enabled, render_prompt_template
+from learning_assistant.constants import GptModels, ResponseVariations
 from learning_assistant.serializers import MessageSerializer
 from learning_assistant.utils import get_chat_response, user_role_is_staff
 
@@ -73,6 +75,7 @@ def post(self, request, course_run_id):
             )
 
         unit_id = request.query_params.get('unit_id')
+        response_variation = request.query_params.get('response_variation')
 
         message_list = request.data
         serializer = MessageSerializer(data=message_list, many=True)
@@ -95,9 +98,17 @@ def post(self, request, course_run_id):
 
         course_id = get_course_id(course_run_id)
 
-        prompt_template = render_prompt_template(request, request.user.id, course_run_id, unit_id, course_id)
+        if response_variation == ResponseVariations.GPT4_UPDATED_PROMPT:
+            gpt_model = GptModels.GPT_4o
+            template_string = getattr(settings, 'LEARNING_ASSISTANT_EXPERIMENTAL_PROMPT_TEMPLATE', '')
+        else:
+            gpt_model = GptModels.GPT_3_5_TURBO
+            template_string = getattr(settings, 'LEARNING_ASSISTANT_PROMPT_TEMPLATE', '')
 
-        status_code, message = get_chat_response(prompt_template, message_list)
+        prompt_template = render_prompt_template(
+            request, request.user.id, course_run_id, unit_id, course_id, template_string
+        )
+        status_code, message = get_chat_response(prompt_template, message_list, gpt_model)
 
         return Response(status=status_code, data=message)
 

tests/test_api.py

@@ -202,8 +202,11 @@ def test_render_prompt_template(
         course_run_id = self.course_run_id
         unit_usage_key = 'block-v1:edX+A+B+type@vertical+block@verticalD'
         course_id = 'edx+test'
+        template_string = getattr(settings, 'LEARNING_ASSISTANT_PROMPT_TEMPLATE', '')
 
-        prompt_text = render_prompt_template(request, user_id, course_run_id, unit_usage_key, course_id)
+        prompt_text = render_prompt_template(
+            request, user_id, course_run_id, unit_usage_key, course_id, template_string
+        )
 
         if unit_content and flag_enabled:
             self.assertIn(unit_content, prompt_text)

tests/test_utils.py

@@ -27,7 +27,7 @@ def setUp(self):
         self.course_id = 'edx+test'
 
     def get_response(self):
-        return get_chat_response(self.prompt_template, self.message_list)
+        return get_chat_response(self.prompt_template, self.message_list, 'gpt-version-test')
 
     @override_settings(CHAT_COMPLETION_API=None)
     def test_no_endpoint_setting(self):
@@ -89,7 +89,8 @@ def test_post_request_structure(self, mock_requests):
         headers = {'Content-Type': 'application/json', 'x-api-key': settings.CHAT_COMPLETION_API_KEY}
 
         response_body = {
-            'message_list': [{'role': 'system', 'content': self.prompt_template}] + self.message_list
+            'message_list': [{'role': 'system', 'content': self.prompt_template}] + self.message_list,
+            'model': 'gpt-version-test',
         }
 
         self.get_response()

tests/test_views.py

@@ -14,6 +14,8 @@
 from django.test.client import Client
 from django.urls import reverse
 
+from learning_assistant.constants import GptModels, ResponseVariations
+
 User = get_user_model()
 
 
@@ -155,13 +157,15 @@ def test_invalid_messages(self, mock_role, mock_waffle, mock_render):
     @patch('learning_assistant.views.get_user_role')
     @patch('learning_assistant.views.CourseEnrollment.get_enrollment')
     @patch('learning_assistant.views.CourseMode')
-    def test_chat_response(self, mock_mode, mock_enrollment, mock_role, mock_waffle, mock_chat_response, mock_render):
+    def test_chat_response_default(
+        self, mock_mode, mock_enrollment, mock_role, mock_waffle, mock_chat_response, mock_render
+    ):
         mock_waffle.return_value = True
         mock_role.return_value = 'student'
         mock_mode.VERIFIED_MODES = ['verified']
         mock_enrollment.return_value = MagicMock(mode='verified')
         mock_chat_response.return_value = (200, {'role': 'assistant', 'content': 'Something else'})
-        mock_render.return_value = 'This is a template'
+        mock_render.return_value = 'This is the default template'
         test_unit_id = 'test-unit-id'
 
         test_data = [
@@ -179,6 +183,54 @@ def test_chat_response(self, mock_mode, mock_enrollment, mock_role, mock_waffle,
         render_args = mock_render.call_args.args
         self.assertIn(test_unit_id, render_args)
 
+        mock_chat_response.assert_called_with(
+            'This is the default template',
+            test_data,
+            GptModels.GPT_3_5_TURBO
+        )
+
+    @patch('learning_assistant.views.render_prompt_template')
+    @patch('learning_assistant.views.get_chat_response')
+    @patch('learning_assistant.views.learning_assistant_enabled')
+    @patch('learning_assistant.views.get_user_role')
+    @patch('learning_assistant.views.CourseEnrollment.get_enrollment')
+    @patch('learning_assistant.views.CourseMode')
+    def test_chat_response_variation(
+        self, mock_mode, mock_enrollment, mock_role, mock_waffle, mock_chat_response, mock_render
+    ):
+        mock_waffle.return_value = True
+        mock_role.return_value = 'student'
+        mock_mode.VERIFIED_MODES = ['verified']
+        mock_enrollment.return_value = MagicMock(mode='verified')
+        mock_chat_response.return_value = (200, {'role': 'assistant', 'content': 'Something else'})
+        mock_render.return_value = 'This is a template for GPT-4o variation'
+        test_unit_id = 'test-unit-id'
+        test_response_variation = ResponseVariations.GPT4_UPDATED_PROMPT
+
+        test_data = [
+            {'role': 'user', 'content': 'What is 2+2?'},
+            {'role': 'assistant', 'content': 'It is 4'}
+        ]
+
+        response = self.client.post(
+            reverse(
+                'chat',
+                kwargs={'course_run_id': self.course_id}
+            )+f'?unit_id={test_unit_id}&response_variation={test_response_variation}',
+            data=json.dumps(test_data),
+            content_type='application/json',
+        )
+        self.assertEqual(response.status_code, 200)
+
+        render_args = mock_render.call_args.args
+        self.assertIn(test_unit_id, render_args)
+
+        mock_chat_response.assert_called_with(
+            'This is a template for GPT-4o variation',
+            test_data,
+            GptModels.GPT_4o
+        )
+
 
 @ddt.ddt
 class LearningAssistantEnabledViewTests(LoggedInTestCase):