fix issue 743 (#753)

Co-authored-by: FelixFehse <[email protected]>

CHANGELOG.md

@@ -11,6 +11,7 @@
 ### Fixes
 - fix: `ExpectedSearchOutput` has only relevant fields and supports generic document-`ID` rather than just str
 - fix: `SearchEvaluationLogic` explicitly compares documents by ids
+- fix: In `RecusrsiveSummarize.do_run`, `num_generated_tokens` not uninitialized anymore. [See Issue 743.](https://github.com/Aleph-Alpha/intelligence-layer/issues/743).
 
 
 ## 0.9.0

src/intelligence_layer/use_cases/summarize/recursive_summarize.py

@@ -51,6 +51,7 @@ def do_run(
         num_partial_summaries = 0
         text_to_summarize = input.text
         summary = ""
+        num_generated_tokens = 0
         while True:
             summarize_output = self.long_context_summarize_task.run(
                 LongContextSummarizeInput(

tests/use_cases/summarize/test_recursive_summarize.py

@@ -68,6 +68,17 @@ def test_recursive_summarize_stops_when_num_partial_summaries_stays_same(
     assert output.generated_tokens > 50
 
 
+def test_recursive_summarize_stops_when_num_partial_summaries_stays_same_with_empty_text(
+    steerable_long_context_summarize: SteerableLongContextSummarize,
+) -> None:
+    max_tokens = 2048
+    input = RecursiveSummarizeInput(text="", max_tokens=max_tokens)
+    task = RecursiveSummarize(steerable_long_context_summarize)
+    output = task.run(input, NoOpTracer())
+
+    assert output.generated_tokens == 0
+
+
 def test_recursive_summarize_stops_after_one_chunk(
     recursive_counting_client: RecursiveCountingClient,
 ) -> None: