fix bug in vllm

dataelement · Nov 26, 2023 · 35d8c47 · 35d8c47
1 parent 1558494
commit 35d8c47
Show file tree

Hide file tree

Showing 3 changed files with 5 additions and 5 deletions.
diff --git a/python/pybackend_libs/requirements.txt b/python/pybackend_libs/requirements.txt
@@ -19,7 +19,7 @@ transformers_stream_generator==0.0.4
 SwissArmyTransformer==0.4.8
 cpm_kernels==1.0.11
 accelerate==0.24.1
-vllm==0.2.1
+vllm==0.2.1.post1
 
 anyio==3.7.1
 starlette==0.27.0

diff --git a/src/tests/regression/test_close_models.py b/src/tests/regression/test_close_models.py
@@ -79,7 +79,7 @@ def test_infer(name):
 
 def test_closed_model():
     RT_EP = os.environ.get('RT_EP', '192.168.106.12:9001')
-    ep_prefix = f'http://{RT_EP}/v2'
+    ep_prefix = f'http://{RT_EP}'
     repo_ep = f'http://{RT_EP}/v2/repository'
 
     test_config = 'config/closed_models.json'
@@ -97,7 +97,7 @@ def test_closed_model():
     for name in model_names:
         load_ep = f'{repo_ep}/models/{name}/load'
         unload_ep = f'{repo_ep}/models/{name}/unload'
-        ready_ep = f'{ep_prefix}/models/{name}/ready'
+        ready_ep = f'{ep_prefix}/v2/models/{name}/ready'
         load_params = load_params_map.get(name)
 
         print(f'load model {name}...')

diff --git a/src/tests/regression/test_load_unload.py b/src/tests/regression/test_load_unload.py
@@ -37,7 +37,7 @@ def call_llm(model, ep):
 
 def test_load_unload():
     RT_EP = os.environ.get('RT_EP', '192.168.106.12:9001')
-    ep_prefix = f'http://{RT_EP}/v2'
+    ep_prefix = f'http://{RT_EP}'
     repo_ep = f'http://{RT_EP}/v2/repository'
 
     test_config = 'config/load_unload.json'
@@ -55,7 +55,7 @@ def test_load_unload():
     for name in model_names:
         load_ep = f'{repo_ep}/models/{name}/load'
         unload_ep = f'{repo_ep}/models/{name}/unload'
-        ready_ep = f'{ep_prefix}/models/{name}/ready'
+        ready_ep = f'{ep_prefix}/v2/models/{name}/ready'
         load_params = load_params_map.get(name)
         infer_ep = f'{ep_prefix}/v2.1/models/{name}/infer'