fix bug

lazyllm/components/deploy/utils.py

@@ -1,5 +1,5 @@
 import os
-import datetime
+from datetime import datetime
 import random
 import lazyllm
 

lazyllm/engine/engine.py

@@ -478,7 +478,7 @@ def make_shared_llm(llm: str, local: bool = True, prompt: Optional[str] = None,
 def make_online_llm(source: str, base_model: Optional[str] = None, prompt: Optional[str] = None,
                     api_key: Optional[str] = None, secret_key: Optional[str] = None,
                     stream: bool = False, token: Optional[str] = None, base_url: Optional[str] = None):
-    if source.lower() == 'lazyllm':
+    if source and source.lower() == 'lazyllm':
         return make_shared_llm(base_model, False, prompt, token, stream)
     else:
         return lazyllm.OnlineChatModule(base_model, source, base_url, stream,

lazyllm/tools/infer_service/client.py

@@ -1,3 +1,4 @@
+import time
 from urllib.parse import urljoin
 import requests
 import lazyllm
@@ -124,3 +125,28 @@ def get_infra_handle(self, token, job_id):
         if not (deployer := getattr(lazyllm.deploy, deploy_method, None)):
             deployer = type(lazyllm.deploy.auto(base_model))
         return lazyllm.TrainableModule(base_model).deploy_method(deployer, url=url)
+
+    def wait_ready(self, token, job_id, timeout=1800):
+        '''
+        This method to wait for a specific job based on the provided job ID.
+
+        Parameters:
+        - token(str): A string representing the authentication token required to access the job details.
+        - job_id(str): The unique identifier for the job whose details need to be retrieved.
+
+        Returns:
+        - infer service status.
+        '''
+        def get_status():
+            response = requests.get(urljoin(self.url, f'jobs/{job_id}'), headers={'token': token})
+            response.raise_for_status()
+            response = response.json()
+            return self.uniform_status(response['status'])
+
+        n = 0
+        while (status := get_status()) != 'Running':
+            if status in ('Invalid', 'Cancelled', 'Failed'):
+                raise RuntimeError(f'Deploy service failed. status is {status}')
+            if n > timeout: raise TimeoutError('Inference service has not started after 1800 seconds.')
+            time.sleep(10)
+            n += 10

tests/requirements.txt

@@ -3,7 +3,8 @@ docx2txt
 olefile
 pytest-rerunfailures
 pytest-order
-pymilvus>=2.4.7, <2.4.11
+pymilvus==2.4.10
+milvus-lite==2.4.10
 openpyxl
 nbconvert
 python-pptx