Skip to content

Build and run agents you can see, understand and trust.

fix(embedding): report the maximum batch time instead of the sum (#2967)

ANIRUDDHA ADAKcf9278bbbd75
diff --git a/src/agentscope/embedding/_embedding_base.py b/src/agentscope/embedding/_embedding_base.py
index 94e5454a0..a103a795c 100644
--- a/src/agentscope/embedding/_embedding_base.py
+++ b/src/agentscope/embedding/_embedding_base.py
@@ -290,7 +290,7 @@ class EmbeddingModelBase(Generic[InputT]):
if resp.source:
sources.add(resp.source)
if resp.usage:
- total_time += resp.usage.time
+ total_time = max(total_time, resp.usage.time)
if resp.usage.tokens:
total_tokens += resp.usage.tokens
diff --git a/tests/embedding_base_test.py b/tests/embedding_base_test.py
index 640fdab0e..8de90fde6 100644
--- a/tests/embedding_base_test.py
+++ b/tests/embedding_base_test.py
@@ -62,8 +62,8 @@ class EmbeddingModelBaseTest(TestCase):
},
)
- def test_merge_live_api_batches_reports_api(self) -> None:
- """Preserve vector order and sum tokens and time across API calls."""
+ def test_merge_live_api_batches_uses_max_time(self) -> None:
+ """Preserve vector order, sum tokens, and take the maximum time."""
merged = EmbeddingModelBase._merge_responses(
[
_resp("api", 1.0, tokens=7, time=0.25),
@@ -78,7 +78,7 @@ class EmbeddingModelBaseTest(TestCase):
"id": AnyString(),
"created_at": AnyString(),
"type": "embedding",
- "usage": {"tokens": 12, "time": 0.75, "type": "embedding"},
+ "usage": {"tokens": 12, "time": 0.5, "type": "embedding"},
"source": "api",
},
)
@@ -103,3 +103,19 @@ class EmbeddingModelBaseTest(TestCase):
"source": "api",
},
)
+
+ def test_merge_empty_responses_reports_api(self) -> None:
+ """An empty response list must not count as a cache hit."""
+ merged = EmbeddingModelBase._merge_responses([])
+
+ self.assertEqual(
+ asdict(merged),
+ asdict(
+ EmbeddingResponse(
+ embeddings=[],
+ id=AnyString(),
+ created_at=AnyString(),
+ usage=EmbeddingUsage(tokens=0, time=0.0),
+ ),
+ ),
+ )