DSPy: The framework for programming—not prompting—language models
Python
fix(clients): honor caching=False per-call in dspy.Embedder (#9708)
Sitananda Prasadb0baa1d165f9
diff --git a/dspy/clients/embedding.py b/dspy/clients/embedding.pyindex 6016a372a..4ba65b34a 100644--- a/dspy/clients/embedding.py+++ b/dspy/clients/embedding.py@@ -93,7 +93,7 @@ class Embedder:raise ValueError("All inputs must be strings.")batch_size = batch_size or self.batch_size- caching = caching or self.caching+ caching = caching if caching is not None else self.cachingmerged_kwargs = self.default_kwargs.copy()merged_kwargs.update(kwargs)diff --git a/tests/clients/test_embedding.py b/tests/clients/test_embedding.pyindex ec83a944a..fae0ea42c 100644--- a/tests/clients/test_embedding.py+++ b/tests/clients/test_embedding.py@@ -151,3 +151,49 @@ async def test_async_embedding():assert len(result) == len(inputs)np.testing.assert_allclose(result, mock_embeddings)+++def test_call_caching_false_overrides_instance_true(cache):+ model = "text-embedding-ada-002"+ inputs = ["hello"]+ with patch("litellm.embedding") as mock_litellm:+ mock_litellm.return_value = MockEmbeddingResponse([[0.1, 0.2, 0.3]])+ embedding = Embedder(model, caching=True)+ embedding(inputs)+ embedding(inputs, caching=False)+ assert mock_litellm.call_count == 2+++def test_call_caching_true_overrides_instance_false(cache):+ model = "text-embedding-ada-002"+ inputs = ["hello"]+ with patch("litellm.embedding") as mock_litellm:+ mock_litellm.return_value = MockEmbeddingResponse([[0.1, 0.2, 0.3]])+ embedding = Embedder(model, caching=False)+ embedding(inputs, caching=True)+ embedding(inputs, caching=True)+ assert mock_litellm.call_count == 1+++async def test_acall_caching_false_overrides_instance_true(cache):+ model = "text-embedding-ada-002"+ inputs = ["hello"]+ with patch("litellm.aembedding") as mock_litellm:+ mock_litellm.return_value = MockEmbeddingResponse([[0.1, 0.2, 0.3]])+ embedding = Embedder(model, caching=True)+ await embedding.acall(inputs)+ await embedding.acall(inputs, caching=False)+ assert mock_litellm.call_count == 2+++async def test_acall_caching_true_overrides_instance_false(cache):+ model = "text-embedding-ada-002"+ inputs = ["hello"]+ with patch("litellm.aembedding") as mock_litellm:+ mock_litellm.return_value = MockEmbeddingResponse([[0.1, 0.2, 0.3]])+ embedding = Embedder(model, caching=False)+ await embedding.acall(inputs, caching=True)+ await embedding.acall(inputs, caching=True)+ assert mock_litellm.call_count == 1