xinference-client 2.4.0__tar.gz → 2.7.0__tar.gz

{xinference_client-2.4.0 → xinference_client-2.7.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: xinference-client
-Version: 2.4.0
+Version: 2.7.0
 Summary: Client for Xinference
 Home-page: https://github.com/xorbitsai/inference-client
 Author: Qin Xuye

{xinference_client-2.4.0 → xinference_client-2.7.0}/xinference_client/_version.py

@@ -8,11 +8,11 @@ import json
 
 version_json = '''
 {
- "date": "2026-03-29T11:20:56+0800",
+ "date": "2026-04-25T15:17:51+0800",
  "dirty": false,
  "error": null,
- "full-revisionid": "46f9ecb0ab0fc671d0ea4e765c813d7b7388d38b",
- "version": "2.4.0"
+ "full-revisionid": "a7d295c394c5dd408b1d7167d692a3326b3f4e03",
+ "version": "2.7.0"
 }
 '''  # END VERSION_JSON
 

{xinference_client-2.4.0 → xinference_client-2.7.0}/xinference_client/client/restful/async_restful_client.py

@@ -1432,6 +1432,21 @@ class AsyncClient:
             )
         await _release_response(response)
 
+    async def terminate_model_replica(self, model_uid: str, replica_id: int) -> int:
+        """Terminate a specific replica of a running model."""
+
+        url = f"{self.base_url}/v1/models/{model_uid}/replicas/{replica_id}"
+
+        response = await self.session.delete(url, headers=self._headers)
+        if response.status != 200:
+            raise RuntimeError(
+                f"Failed to terminate model replica, detail: {await _get_error_string(response)}"
+            )
+
+        response_data = await response.json()
+        await _release_response(response)
+        return response_data["remaining_replicas"]
+
     async def get_launch_model_progress(self, model_uid: str) -> dict:
         """
         Get progress of the specific model.
@@ -1575,7 +1590,7 @@ class AsyncClient:
                 model_uid, self.base_url, auth_headers=self._headers
             )
         else:
-            raise ValueError(f"Unknown model type:{desc['model_type']}")
+            raise ValueError(f"Unknown model type: {desc['model_type']}")
 
     async def describe_model(self, model_uid: str):
         """

{xinference_client-2.4.0 → xinference_client-2.7.0}/xinference_client/client/restful/restful_client.py

@@ -1329,6 +1329,20 @@ class Client:
                 f"Failed to terminate model, detail: {_get_error_string(response)}"
             )
 
+    def terminate_model_replica(self, model_uid: str, replica_id: int) -> int:
+        """Terminate a specific replica of a running model."""
+
+        url = f"{self.base_url}/v1/models/{model_uid}/replicas/{replica_id}"
+
+        response = self.session.delete(url, headers=self._headers)
+        if response.status_code != 200:
+            raise RuntimeError(
+                f"Failed to terminate model replica, detail: {_get_error_string(response)}"
+            )
+
+        response_data = response.json()
+        return response_data["remaining_replicas"]
+
     def get_launch_model_progress(self, model_uid: str) -> dict:
         """
         Get progress of the specific model.
@@ -1467,7 +1481,7 @@ class Client:
                 model_uid, self.base_url, auth_headers=self._headers
             )
         else:
-            raise ValueError(f"Unknown model type:{desc['model_type']}")
+            raise ValueError(f"Unknown model type: {desc['model_type']}")
 
     def describe_model(self, model_uid: str):
         """

{xinference_client-2.4.0 → xinference_client-2.7.0}/xinference_client.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: xinference-client
-Version: 2.4.0
+Version: 2.7.0
 Summary: Client for Xinference
 Home-page: https://github.com/xorbitsai/inference-client
 Author: Qin Xuye