Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -47,36 +47,12 @@ MODEL_IDS = [
|
|
47 |
# Add here wherever model is working for you
|
48 |
]
|
49 |
|
50 |
-
def is_model_overloaded(model_url):
|
51 |
-
"""Verify if the model is overloaded doing a test call."""
|
52 |
-
try:
|
53 |
-
response = requests.post(model_url, json={"inputs": "Test"})
|
54 |
-
if verbose:
|
55 |
-
print(f"URL: {model_url}")
|
56 |
-
print(f"Response Status Code: {response.status_code}")
|
57 |
-
|
58 |
-
if response.status_code in [503, 404, 424]:
|
59 |
-
return True
|
60 |
-
return False
|
61 |
-
except requests.RequestException as e:
|
62 |
-
if verbose:
|
63 |
-
print(f"Error accessing {model_url}: {e}")
|
64 |
-
return True # If there is an error, consider it overloaded
|
65 |
-
|
66 |
|
67 |
def get_available_model():
|
68 |
"""Select the first model available from the list."""
|
69 |
-
|
70 |
-
print("trying",model_url)
|
71 |
-
if not is_model_overloaded(model_url):
|
72 |
-
return model_url
|
73 |
-
return MODEL_IDS[0] # if all are failing, use the first model by default
|
74 |
-
|
75 |
-
if verbose: print("Checking available models.")
|
76 |
|
77 |
selected_model_id = get_available_model()
|
78 |
-
|
79 |
-
if verbose: print("Using model {selected_model_id}")
|
80 |
|
81 |
model = HfApiModel(
|
82 |
max_tokens=1048,
|
|
|
47 |
# Add here wherever model is working for you
|
48 |
]
|
49 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
50 |
|
51 |
def get_available_model():
|
52 |
"""Select the first model available from the list."""
|
53 |
+
return MODEL_IDS[0]
|
|
|
|
|
|
|
|
|
|
|
|
|
54 |
|
55 |
selected_model_id = get_available_model()
|
|
|
|
|
56 |
|
57 |
model = HfApiModel(
|
58 |
max_tokens=1048,
|