adjust zeroGPU

sotopia_generate.py

@@ -37,7 +37,6 @@ OutputType = TypeVar("OutputType", bound=object)
 log = logging.getLogger("generate")
 logging_handler = LoggingCallbackHandler("langchain")
 
-# @spaces.GPU
 def generate_action(
     model_name: str,
     history: str,
@@ -82,7 +81,6 @@ def generate_action(
     #     print(e)
     #     return AgentAction(action_type="none", argument="")
 
-@spaces.GPU(duration=1200)
 @cache
 def prepare_model(model_name):
     compute_type = torch.float16
@@ -151,7 +149,7 @@ def obtain_chain_hf(
     chain = LLMChain(llm=hf, prompt=chat_prompt_template)
     return chain
 
-
+@spaces.GPU
 def generate(
     model_name: str,
     template: str,

sotopia_space/chat.py

@@ -10,7 +10,7 @@ from sotopia_generate import prepare_model, generate_action
 from sotopia_space.constants import MODEL_OPTIONS
 
 DEPLOYED = os.getenv("DEPLOYED", "true").lower() == "true"
-DEFAULT_MODEL_SELECTION = "gpt-3.5-turbo"
+DEFAULT_MODEL_SELECTION = "cmu-lti/sotopia-pi-mistral-7b-BC_SR"
 TEMPERATURE = 0.7
 TOP_P = 1
 MAX_TOKENS = 1024