From 381b7201d311062179e0690209f187d069289ffe Mon Sep 17 00:00:00 2001
From: amumu96 <15667065080@163.com>
Date: Mon, 29 Apr 2024 12:23:40 +0800
Subject: [PATCH] fix metric empty

---
 xinference/core/model.py | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/xinference/core/model.py b/xinference/core/model.py
index 63616cf3a7..251e90ce92 100644
--- a/xinference/core/model.py
+++ b/xinference/core/model.py
@@ -25,6 +25,7 @@
     AsyncGenerator,
     Callable,
     Dict,
+    Generator,
     Iterator,
     List,
     Optional,
@@ -379,8 +380,13 @@ async def chat(self, prompt: str, *args, **kwargs):
             raise AttributeError(f"Model {self._model.model_spec} is not for chat.")
         finally:
             # For the non stream result.
-            if response is not None and isinstance(response, dict):
-                usage = response["usage"]
+            record = None
+            if isinstance(response, Generator) or isinstance(response, AsyncGenerator):
+                record = response
+            elif isinstance(response, bytes):
+                record = json.loads(response)
+            if record and isinstance(record, dict):
+                usage = record["usage"]
                 # Some backends may not have a valid usage, we just skip them.
                 completion_tokens = usage["completion_tokens"]
                 prompt_tokens = usage["prompt_tokens"]