Skip to content

Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.

locomo数据导入优化 (#2852)

fujiajie66679cb57107413
diff --git a/benchmark/locomo/vikingbot/import_to_ov.py b/benchmark/locomo/vikingbot/import_to_ov.py
index 7b8879dc4..95746f927 100644
--- a/benchmark/locomo/vikingbot/import_to_ov.py
+++ b/benchmark/locomo/vikingbot/import_to_ov.py
@@ -89,6 +89,28 @@ def load_locomo_data(
return data
+def build_message_text(msg: Dict[str, Any]) -> str:
+ """Preserve LoCoMo image metadata as labelled text for memory extraction."""
+ text = (msg.get("text") or "").strip()
+
+ blip_caption_value = msg.get("blip_caption")
+ blip_caption = str(blip_caption_value or "").strip()
+
+ query = str(msg.get("query") or "").strip()
+
+ image_parts = []
+ if blip_caption:
+ image_parts.append(f"image description: {blip_caption}")
+ if query:
+ image_parts.append(f"image search/query text: {query}")
+
+ if not image_parts:
+ return text
+
+ image_note = f"(attached image; {'; '.join(image_parts)})"
+ return f"{text}\n{image_note}" if text else image_note
+
+
def build_session_messages(
item: Dict[str, Any],
session_range: Optional[Tuple[int, int]] = None,
@@ -126,7 +148,7 @@ def build_session_messages(
messages = []
for idx, msg in enumerate(conv[sk]):
speaker = msg.get("speaker", "unknown")
- text = msg.get("text", "")
+ text = build_message_text(msg)
if group_chat:
messages.append(
{