memmachine-server 0.3.4__tar.gz → 0.3.5__tar.gz

{memmachine_server-0.3.4/src/memmachine_server.egg-info → memmachine_server-0.3.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: memmachine-server
-Version: 0.3.4
+Version: 0.3.5
 Summary: MemMachine Server - The complete MemMachine memory system server with episodic and profile memory
 Author-email: MemMachine <noreply@memmachine.ai>
 License-Expression: Apache-2.0

{memmachine_server-0.3.4 → memmachine_server-0.3.5}/server_tests/memmachine_server/common/embedder/test_openai_embedder.py

@@ -124,3 +124,48 @@ async def test_embed_fail_after_max_retries_on_internal_server_error(mock_sleep)
     assert mock_sleep.call_count == 2
     mock_sleep.assert_any_await(1)
     mock_sleep.assert_any_await(2)
+
+
+@pytest.mark.asyncio
+async def test_embed_oversized_input_with_no_max_input_length():
+    """Regression for #1298: text > 75,000 chars must not raise ValueError.
+
+    When max_input_length is None, _embed() must fall back to
+    max_total_input_length_per_request (75,000) as the chunking bound so that
+    cluster_texts() never receives a text that exceeds its hard limit.
+    Prior to the fix, this raised ValueError("Text length ... exceeds
+    max_total_length_per_cluster ..."), which surfaced as HTTP 500.
+    """
+    mock_client = AsyncMock(spec=openai.AsyncOpenAI)
+
+    mock_embedding = MagicMock()
+    mock_embedding.embedding = [0.1] * 256
+
+    mock_response = MagicMock()
+    mock_response.data = [mock_embedding]
+    mock_response.usage.prompt_tokens = 10
+    mock_response.usage.total_tokens = 10
+
+    mock_client.embeddings.create = AsyncMock(return_value=mock_response)
+
+    embedder = OpenAIEmbedder(
+        OpenAIEmbedderParams(
+            client=mock_client,
+            model="test-model",
+            dimensions=256,
+            max_input_length=None,  # default — reproduces the crash scenario
+        )
+    )
+
+    # 80,000 chars exceeds the 75,000-char cluster hard limit by 5,000.
+    long_text = "x" * 80_000
+
+    # Before the fix this raised:
+    #   ValueError: Text length 80000 exceeds max_total_length_per_cluster 75000
+    result = await embedder.ingest_embed([long_text])
+
+    assert len(result) == 1
+    assert len(result[0]) == 256
+    # The text must have been split: balanced chunks of 80,000 chars at limit
+    # 75,000 produce two chunks of 40,000 each, each in its own cluster.
+    assert mock_client.embeddings.create.call_count >= 2

{memmachine_server-0.3.4 → memmachine_server-0.3.5}/src/memmachine_server/common/embedder/openai_embedder.py

@@ -150,11 +150,11 @@ class OpenAIEmbedder(Embedder):
             for input_text in inputs
         ]
 
+        effective_max = (
+            self._max_input_length or self.max_total_input_length_per_request
+        )
         inputs_chunks = [
-            chunk_text_balanced(input_text, self._max_input_length)
-            if self._max_input_length is not None
-            else [input_text]
-            for input_text in inputs
+            chunk_text_balanced(input_text, effective_max) for input_text in inputs
         ]
 
         chunks = [chunk for input_chunks in inputs_chunks for chunk in input_chunks]

{memmachine_server-0.3.4 → memmachine_server-0.3.5/src/memmachine_server.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: memmachine-server
-Version: 0.3.4
+Version: 0.3.5
 Summary: MemMachine Server - The complete MemMachine memory system server with episodic and profile memory
 Author-email: MemMachine <noreply@memmachine.ai>
 License-Expression: Apache-2.0