Skip sections that should be filtered from quizzes

Author: elasticsounds

adt_press/llm/section_quiz.py

@@ -1,8 +1,9 @@
 # mypy: ignore-errors
-import instructor
 from banks import Prompt
-from litellm import acompletion
 from pydantic import BaseModel, ValidationInfo, field_validator
+import structlog
+
+from adt_press.llm import get_instructor_client
 
 from adt_press.models.config import QuizPromptConfig
 from adt_press.models.quiz import SectionQuiz
@@ -11,6 +12,7 @@
 from adt_press.utils.file import cached_read_text_file
 from adt_press.utils.languages import Language
 
+log = structlog.get_logger(__name__)
 
 class Quiz(BaseModel):
     question: str
@@ -60,7 +62,7 @@ class QuizResponse(BaseModel):
 
 async def generate_quiz(
     config: QuizPromptConfig, language: Language, sections: list[PageSection], text_groups_by_id: dict[str, TextGroup]
-) -> SectionQuiz:
+) -> SectionQuiz | None:
     context = dict(
         sections=sections,
         text_groups=text_groups_by_id,
@@ -69,14 +71,19 @@ async def generate_quiz(
     )
 
     prompt = Prompt(cached_read_text_file(config.template_path))
-    client = instructor.from_litellm(acompletion)
-    response: QuizResponse = await client.chat.completions.create(
-        model=config.model,
-        response_model=QuizResponse,
-        messages=[m.model_dump(exclude_none=True) for m in prompt.chat_messages(context)],
-        max_retries=config.max_retries,
-        context={},
-    )
+    client = get_instructor_client()
+    try:
+        response: QuizResponse = await client.chat.completions.create(
+            model=config.model,
+            response_model=QuizResponse,
+            messages=[m.model_dump(exclude_none=True) for m in prompt.chat_messages(context)],
+            max_retries=config.max_retries,
+            timeout=config.timeout,
+            context={},
+        )
+    except Exception as exc:
+        log.warning("quiz_generation_failed", error=str(exc))
+        return None
 
     after_section = sections[-1]
 

adt_press/nodes/config_nodes.py

@@ -383,6 +383,10 @@ def pruned_section_types_config(config: DictConfig) -> list[SectionTypeName]:
     return list[SectionTypeName](config.get("section_filters", {}).get("pruned_section_types", []))
 
 
+def quiz_count_section_types_config(config: DictConfig) -> list[SectionTypeName]:
+    return list[SectionTypeName](config.get("section_filters", {}).get("quiz_count_section_types", []))
+
+
 @cache(behavior="recompute")
 def activity_strategy_config(config: DictConfig) -> str:
     """Get the activity generation strategy from config."""

adt_press/nodes/section_nodes.py

@@ -93,10 +93,13 @@ def quizzes_by_section_id__llm(
     filtered_sections_by_page_id: dict[PageID, PageSections],
     pdf_text_groups_by_id: dict[TextGroupID, TextGroup],
     quiz_prompt_config: QuizPromptConfig,
+    quiz_count_section_types_config: list[SectionTypeName],
 ) -> dict[SectionID, SectionQuiz]:
     if quiz_prompt_config.sections_per_quiz < 1:
         raise ValueError("sections_per_quiz must be at least 1, use quiz_strategy 'none' to disable quizzes")
 
+    quiz_count_section_types = set(quiz_count_section_types_config)
+
     async def get_quizzes():
         tasks = []
         sections = []
@@ -107,6 +110,9 @@ async def get_quizzes():
                 if section.is_pruned:
                     continue
 
+                if quiz_count_section_types and section.section_type.name not in quiz_count_section_types:
+                    continue
+
                 count += 1
                 sections.append(section)
 
@@ -118,7 +124,7 @@ async def get_quizzes():
         return await gather_with_limit(tasks, quiz_prompt_config.rate_limit)
 
     results = run_async_task(get_quizzes)
-    return {quiz.section_id: quiz for quiz in results}
+    return {quiz.section_id: quiz for quiz in results if quiz}
 
 
 @config.when(explanation_strategy="llm")

config/config.yaml

@@ -297,6 +297,13 @@ section_filters:
     - back_cover
     - credits
     - inside_cover
+  # Only these section types count toward quiz grouping. Empty list means all non-pruned sections count.
+  quiz_count_section_types:
+    - boxed_text
+    - text_only
+    - text_and_single_image
+    - text_and_images
+    - images_only
 
 prompts:
   metadata_extraction:

prompts/section_quiz.jinja2

@@ -16,6 +16,7 @@ For the quiz format:
   - The explanation inside each option must describe that specific option
   - The option at `answer_index` must have the congratulatory/correct explanation, and the others must provide gentle corrections
   - Each explanation option should start with an emoji: ❌ for incorrect answers and ✅ for the correct answer
+- IMPORTANT: The `quiz` object MUST include the full `options` array with 3 objects. Do not omit any fields.
 
 Return your response as JSON with this exact structure:
 {