feat: integrate system prompt

CHANGELOG.rst

@@ -11,9 +11,10 @@ Change Log
 
 .. There should always be an "Unreleased" section for changes pending release.
 
-Unreleased
-**********
+2.0.0 - 2024-01-03
+******************
 * Add content cache
+* Integrate system prompt setting
 
 1.5.0 - 2023-10-18
 ******************

learning_assistant/__init__.py

@@ -2,6 +2,6 @@
 Plugin for a learning assistant backend, intended for use within edx-platform.
 """
 
-__version__ = '1.5.0'
+__version__ = '2.0.0'
 
 default_app_config = 'learning_assistant.apps.LearningAssistantConfig'  # pylint: disable=invalid-name

learning_assistant/api.py

@@ -1,9 +1,12 @@
 """
 Library for the learning_assistant app.
 """
+import logging
+
 from django.conf import settings
 from django.core.cache import cache
 from edx_django_utils.cache import get_cache_key
+from jinja2 import BaseLoader, Environment
 
 from learning_assistant.constants import ACCEPTED_CATEGORY_TYPES, CATEGORY_TYPE_MAP
 from learning_assistant.models import CoursePrompt
@@ -16,6 +19,8 @@
 )
 from learning_assistant.text_utils import html_to_text
 
+log = logging.getLogger(__name__)
+
 
 def get_deserialized_prompt_content_by_course_id(course_id):
     """
@@ -112,3 +117,17 @@ def get_block_content(request, user_id, course_id, unit_usage_key):
         cache.set(cache_key, cache_data, getattr(settings, 'LEARNING_ASSISTANT_CACHE_TIMEOUT', 360))
 
     return cache_data['content_length'], cache_data['content_items']
+
+
+def render_prompt_template(request, user_id, course_id, unit_usage_key):
+    """
+    Return a rendered prompt template, specified by the LEARNING_ASSISTANT_PROMPT_TEMPLATE setting.
+    """
+    unit_content = ''
+    if unit_usage_key:
+        _, unit_content = get_block_content(request, user_id, course_id, unit_usage_key)
+
+    template_string = getattr(settings, 'LEARNING_ASSISTANT_PROMPT_TEMPLATE', '')
+    template = Environment(loader=BaseLoader).from_string(template_string)
+    data = template.render(unit_content=unit_content)
+    return data

learning_assistant/platform_imports.py

@@ -22,7 +22,7 @@ def get_text_transcript(video_block):
 def get_single_block(request, user_id, course_id, usage_key_string, course=None):
     """Load a single xblock."""
     # pylint: disable=import-error, import-outside-toplevel
-    from lms.djangoapps.courseware.block_renderer import load_single_xblock
+    from lms.djangoapps.courseware.block_render import load_single_xblock
     return load_single_xblock(request, user_id, course_id, usage_key_string, course)
 
 
@@ -45,3 +45,15 @@ def block_get_children(block):
     # pylint: disable=import-error, import-outside-toplevel
     from openedx.core.lib.graph_traversals import get_children
     return get_children(block)
+
+
+def get_cache_course_run_data(course_run_id, fields):
+    """
+    Return course run related data given a course run id.
+
+    This function makes use of the discovery course run cache, which is necessary because
+    only the discovery service stores the relation between courseruns and courses.
+    """
+    # pylint: disable=import-error, import-outside-toplevel
+    from openedx.core.djangoapps.catalog.utils import get_course_run_data
+    return get_course_run_data(course_run_id, fields)

learning_assistant/utils.py

@@ -1,6 +1,7 @@
 """
 Utils file for learning-assistant.
 """
+import copy
 import json
 import logging
 
@@ -9,6 +10,8 @@
 from requests.exceptions import ConnectTimeout
 from rest_framework import status as http_status
 
+from learning_assistant.platform_imports import get_cache_course_run_data
+
 log = logging.getLogger(__name__)
 
 
@@ -22,21 +25,31 @@ def _estimated_message_tokens(message):
     return int((len(message) - message.count(' ')) / chars_per_token) + json_padding
 
 
-def get_reduced_message_list(system_list, message_list):
+def get_reduced_message_list(prompt_template, message_list):
     """
     If messages are larger than allotted token amount, return a smaller list of messages.
     """
-    total_system_tokens = sum(_estimated_message_tokens(system_message['content']) for system_message in system_list)
+    # the total number of system tokens is a sum of estimated tokens that includes the prompt template, the
+    # course title, and the course skills. It is necessary to include estimations for the course title and
+    # course skills, as the chat endpoint the prompt is being passed to is responsible for filling in the values
+    # for both of those variables.
+    total_system_tokens = (
+        _estimated_message_tokens(prompt_template)
+        + _estimated_message_tokens('.' * 40)  # average number of characters per course name is 40
+        + _estimated_message_tokens('.' * 116)  # average number of characters for skill names is 116
+    )
 
     max_tokens = getattr(settings, 'CHAT_COMPLETION_MAX_TOKENS', 16385)
     response_tokens = getattr(settings, 'CHAT_COMPLETION_RESPONSE_TOKENS', 1000)
     remaining_tokens = max_tokens - response_tokens - total_system_tokens
 
     new_message_list = []
+    # use copy of list, as it is modified as part of the reduction
+    message_list_copy = copy.deepcopy(message_list)
     total_message_tokens = 0
 
-    while total_message_tokens < remaining_tokens and len(message_list) != 0:
-        new_message = message_list.pop()
+    while total_message_tokens < remaining_tokens and len(message_list_copy) != 0:
+        new_message = message_list_copy.pop()
         total_message_tokens += _estimated_message_tokens(new_message['content'])
         if total_message_tokens >= remaining_tokens:
             break
@@ -47,7 +60,34 @@ def get_reduced_message_list(system_list, message_list):
     return new_message_list
 
 
-def get_chat_response(system_list, message_list):
+def get_course_id(course_run_id):
+    """
+    Given a course run id (str), return the associated course key.
+    """
+    course_data = get_cache_course_run_data(course_run_id, ['course'])
+    course_key = course_data['course']
+    return course_key
+
+
+def create_request_body(prompt_template, message_list, courserun_id):
+    """
+    Form request body to be passed to the chat endpoint.
+    """
+    response_body = {
+        'context': {
+            'content': prompt_template,
+            'render': {
+                'doc_id': get_course_id(courserun_id),
+                'fields': ['skillNames', 'title']
+            }
+        },
+        'message_list': get_reduced_message_list(prompt_template, message_list)
+    }
+
+    return response_body
+
+
+def get_chat_response(prompt_template, message_list, courserun_id):
     """
     Pass message list to chat endpoint, as defined by the CHAT_COMPLETION_API setting.
     """
@@ -58,8 +98,7 @@ def get_chat_response(system_list, message_list):
         connect_timeout = getattr(settings, 'CHAT_COMPLETION_API_CONNECT_TIMEOUT', 1)
         read_timeout = getattr(settings, 'CHAT_COMPLETION_API_READ_TIMEOUT', 15)
 
-        reduced_messages = get_reduced_message_list(system_list, message_list)
-        body = {'message_list': system_list + reduced_messages}
+        body = create_request_body(prompt_template, message_list, courserun_id)
 
         try:
             response = requests.post(

learning_assistant/views.py

@@ -20,7 +20,7 @@
     # If the waffle flag is false, the endpoint will force an early return.
     learning_assistant_is_active = False
 
-from learning_assistant.api import get_setup_messages
+from learning_assistant.api import render_prompt_template
 from learning_assistant.serializers import MessageSerializer
 from learning_assistant.utils import get_chat_response
 
@@ -46,16 +46,16 @@ def post(self, request, course_id):
             ]
         }
         """
-        course_key = CourseKey.from_string(course_id)
-        if not learning_assistant_is_active(course_key):
+        courserun_key = CourseKey.from_string(course_id)
+        if not learning_assistant_is_active(courserun_key):
             return Response(
                 status=http_status.HTTP_403_FORBIDDEN,
                 data={'detail': 'Learning assistant not enabled for course.'}
             )
 
         # If user does not have an enrollment record, or is not staff, they should not have access
-        user_role = get_user_role(request.user, course_key)
-        enrollment_object = CourseEnrollment.get_enrollment(request.user, course_key)
+        user_role = get_user_role(request.user, courserun_key)
+        enrollment_object = CourseEnrollment.get_enrollment(request.user, courserun_key)
         enrollment_mode = enrollment_object.mode if enrollment_object else None
         if (
             (enrollment_mode not in CourseMode.ALL_MODES)
@@ -66,12 +66,7 @@ def post(self, request, course_id):
                 data={'detail': 'Must be staff or have valid enrollment.'}
             )
 
-        prompt_messages = get_setup_messages(course_id)
-        if not prompt_messages:
-            return Response(
-                status=http_status.HTTP_404_NOT_FOUND,
-                data={'detail': 'Learning assistant not enabled for course.'}
-            )
+        unit_id = request.query_params.get('unit_id')
 
         message_list = request.data
         serializer = MessageSerializer(data=message_list, many=True)
@@ -84,16 +79,16 @@ def post(self, request, course_id):
                 data={'detail': 'Invalid data', 'errors': serializer.errors}
             )
 
-        # append system message to beginning of message list
-        message_setup = prompt_messages
-
         log.info(
             'Attempting to retrieve chat response for user_id=%(user_id)s in course_id=%(course_id)s',
             {
                 'user_id': request.user.id,
                 'course_id': course_id
             }
         )
-        status_code, message = get_chat_response(message_setup, message_list)
+
+        prompt_template = render_prompt_template(request, request.user.id, course_id, unit_id)
+
+        status_code, message = get_chat_response(prompt_template, message_list, course_id)
 
         return Response(status=status_code, data=message)

requirements/base.in

@@ -7,3 +7,4 @@ djangorestframework
 edx-drf-extensions
 edx-rest-api-client
 edx-opaque-keys
+jinja2

requirements/base.txt

@@ -58,11 +58,15 @@ edx-rest-api-client==5.6.1
     # via -r requirements/base.in
 idna==3.6
     # via requests
+jinja2==3.1.2
+    # via -r requirements/base.in
+markupsafe==2.1.3
+    # via jinja2
 newrelic==9.3.0
     # via edx-django-utils
 pbr==6.0.0
     # via stevedore
-psutil==5.9.6
+psutil==5.9.7
     # via edx-django-utils
 pycparser==2.21
     # via cffi
@@ -96,8 +100,6 @@ stevedore==5.1.0
     #   edx-django-utils
     #   edx-opaque-keys
 typing-extensions==4.9.0
-    # via
-    #   asgiref
-    #   edx-opaque-keys
+    # via edx-opaque-keys
 urllib3==2.1.0
     # via requests

requirements/ci.txt

@@ -16,7 +16,7 @@ codecov==2.1.13
     # via -r requirements/ci.in
 colorama==0.4.6
     # via tox
-coverage==7.3.2
+coverage==7.4.0
     # via codecov
 distlib==0.3.8
     # via virtualenv
@@ -40,10 +40,6 @@ pyproject-api==1.6.1
     # via tox
 requests==2.31.0
     # via codecov
-tomli==2.0.1
-    # via
-    #   pyproject-api
-    #   tox
 tox==4.11.4
     # via -r requirements/ci.in
 urllib3==2.1.0