chore: set max duration to 120s

app.py

@@ -18,7 +18,6 @@ moderator_model = AutoModelForSequenceClassification.from_pretrained(moderator_m
 moderator_model.to("cuda")
 moderator_tokenizer = AutoTokenizer.from_pretrained(moderator_model_name)
 
-@spaces.GPU
 def generate_responses(model, tokenizer, prompts):
     messages = [[{"role": "user", "content": message}] for message in prompts]
 
@@ -41,7 +40,6 @@ def generate_responses(model, tokenizer, prompts):
 
     return responses
 
-@spaces.GPU
 def classify_pairs(model, tokenizer, prompts, responses):
     texts = [
         prompt + "[SEP]" + response for prompt, response in zip(prompts, responses)
@@ -59,7 +57,7 @@ def classify_pairs(model, tokenizer, prompts, responses):
     return unsafety_scores
 
 
-@spaces.GPU
+@spaces.GPU(duration=120)
 def generate(submission: list[dict[str, str]], team_id: str) -> list[dict[str, str | float]]:
     ids = [s["id"] for s in submission]
     prompts = [s["prompt"] for s in submission]