DSPy: The framework for programming—not prompting—language models
Python
Stabilize real LM tests with deterministic temperature (#9218)
Omar Khattab9d579f22d063
diff --git a/tests/primitives/test_base_module.py b/tests/primitives/test_base_module.pyindex a8960cfda..fb538f906 100644--- a/tests/primitives/test_base_module.py+++ b/tests/primitives/test_base_module.py@@ -247,7 +247,7 @@ def test_load_with_version_mismatch(tmp_path):@pytest.mark.llm_calldef test_single_module_call_with_usage_tracker(lm_for_test):- dspy.configure(lm=dspy.LM(lm_for_test, cache=False), track_usage=True)+ dspy.configure(lm=dspy.LM(lm_for_test, cache=False, temperature=0.0), track_usage=True)predict = dspy.ChainOfThought("question -> answer")output = predict(question="What is the capital of France?")@@ -259,7 +259,7 @@ def test_single_module_call_with_usage_tracker(lm_for_test):assert lm_usage[lm_for_test]["total_tokens"] > 0# Test no usage being tracked when cache is enabled- dspy.configure(lm=dspy.LM(lm_for_test, cache=True), track_usage=True)+ dspy.configure(lm=dspy.LM(lm_for_test, cache=True, temperature=0.0), track_usage=True)for _ in range(2):output = predict(question="What is the capital of France?")@@ -268,7 +268,7 @@ def test_single_module_call_with_usage_tracker(lm_for_test):@pytest.mark.llm_calldef test_multi_module_call_with_usage_tracker(lm_for_test):- dspy.configure(lm=dspy.LM(lm_for_test, cache=False), track_usage=True)+ dspy.configure(lm=dspy.LM(lm_for_test, cache=False, temperature=0.0), track_usage=True)class MyProgram(dspy.Module):def __init__(self):diff --git a/tests/utils/test_usage_tracker.py b/tests/utils/test_usage_tracker.pyindex f3ae61273..af303ba12 100644--- a/tests/utils/test_usage_tracker.py+++ b/tests/utils/test_usage_tracker.py@@ -138,7 +138,7 @@ def test_track_usage_with_multiple_models():def test_track_usage_context_manager(lm_for_test):- lm = dspy.LM(lm_for_test, cache=False)+ lm = dspy.LM(lm_for_test, cache=False, temperature=0.0)dspy.configure(lm=lm)predict = dspy.ChainOfThought("question -> answer")