[fix] Fix setting default rolling batch and extract lora adapter (#2931)

xyang16 · web-flow · commit 28ac35884faf · 2025-10-30T09:05:32.000-07:00
diff --git a/engines/python/setup/djl_python/async_utils.py b/engines/python/setup/djl_python/async_utils.py
@@ -128,7 +128,7 @@ def _extract_lora_adapter(raw_request, decoded_payload):
             SAGEMAKER_ADAPTER_IDENTIFIER_HEADER)
         logging.debug(f"Found adapter in headers: {adapter_name}")
     elif "adapters" in decoded_payload:
-        adapter_name = decoded_payload.get("adapters")
+        adapter_name = decoded_payload.pop("adapters")
         logging.debug(f"Found adapter in payload: {adapter_name}")
 
     return adapter_name
diff --git a/wlm/src/main/java/ai/djl/serving/wlm/LmiConfigRecommender.java b/wlm/src/main/java/ai/djl/serving/wlm/LmiConfigRecommender.java
@@ -82,7 +82,6 @@ private static void setRollingBatch(
             rollingBatch = "trtllm";
         } else {
             rollingBatch = "disable";
-            lmiProperties.setProperty("option.async_mode", "true");
         }
         lmiProperties.setProperty("option.rolling_batch", rollingBatch);
     }

Original file line number	Diff line number	Diff line change
`@@ -82,7 +82,6 @@ private static void setRollingBatch(`
`82`	`82`	`rollingBatch = "trtllm";`
`83`	`83`	`} else {`
`84`	`84`	`rollingBatch = "disable";`
`85`		`- lmiProperties.setProperty("option.async_mode", "true");`
`86`	`85`	`}`
`87`	`86`	`lmiProperties.setProperty("option.rolling_batch", rollingBatch);`
`88`	`87`	`}`