chore: update to gpt-4-turbo

prediction_prophet/app.py

@@ -19,7 +19,7 @@
 def research(
     goal: str,
     tavily_api_key: SecretStr,
-    model: str = "gpt-4-0125-preview",
+    model: str = "gpt-4-turbo",
     initial_subqueries_limit: int = 20,
     subqueries_limit: int = 4,
     scrape_content_split_chunk_size: int = 800,
@@ -144,7 +144,7 @@ def research(
             )
 
             with st.status("Making prediction"):
-                prediction = _make_prediction(market_question=question, additional_information=report, engine="gpt-4-0125-preview", temperature=0.0)
+                prediction = _make_prediction(market_question=question, additional_information=report, engine="gpt-4-turbo", temperature=0.0)
 
                 if prediction.outcome_prediction == None:
                     st.container().error("The agent failed to generate a prediction")

prediction_prophet/autonolas/research.py

@@ -52,7 +52,7 @@
 ]
 TOOL_TO_ENGINE = {
     "prediction-sentence-embedding-conservative": "gpt-3.5-turbo",
-    "prediction-sentence-embedding-bold": "gpt-4",
+    "prediction-sentence-embedding-bold": "gpt-4-turbo",
 }
 
 

prediction_prophet/deployment/models.py

@@ -105,7 +105,7 @@ class DeployableAgentER_PredictionProphetGPT3(DeployableAgentER):
 
 
 class DAPredictionProphetGPT4(DeployableAgentER):
-    agent = PredictionProphetAgent(model="gpt-4-0125-preview")
+    agent = PredictionProphetAgent(model="gpt-4-turbo")
     # Limit to just 1, because so far it seems that 20x higher costs aren't justified by the prediction performance.
     max_markets_per_run = 1
 

prediction_prophet/functions/debate_prediction.py

@@ -92,7 +92,7 @@ def make_debated_prediction(prompt: str, additional_information: str, api_key: S
 
     prediction_chain = (
         prediction_prompt |
-        ChatOpenAI(model="gpt-4-0125-preview", api_key=secretstr_to_v1_secretstr(api_key)) |
+        ChatOpenAI(model="gpt-4-turbo", api_key=secretstr_to_v1_secretstr(api_key)) |
         StrOutputParser()
     )
 
@@ -106,7 +106,7 @@ def make_debated_prediction(prompt: str, additional_information: str, api_key: S
         ConversableAgent(
             name=f"Predictor_Agent_{i}",
             system_message=PREDICTION_PROMPT,
-            llm_config={"config_list": [{"model": "gpt-4-0125-preview", "api_key": api_key.get_secret_value()}]},
+            llm_config={"config_list": [{"model": "gpt-4-turbo", "api_key": api_key.get_secret_value()}]},
             human_input_mode="NEVER")
         for i in range(2) ]
 

prediction_prophet/functions/evaluate_question.py

@@ -41,7 +41,7 @@
 @persistent_inmemory_cache
 def is_predictable(
     question: str,
-    engine: str = "gpt-4-0125-preview",
+    engine: str = "gpt-4-turbo",
     prompt_template: str = QUESTION_EVALUATE_PROMPT,
     api_key: SecretStr | None = None
 ) -> tuple[bool, str]:

prediction_prophet/functions/is_predictable_and_binary.py

@@ -42,7 +42,7 @@
 @persistent_inmemory_cache
 def is_predictable_and_binary(
     question: str,
-    engine: str = "gpt-4-0125-preview",
+    engine: str = "gpt-4-turbo",
     prompt_template: str = QUESTION_EVALUATE_PROMPT,
     api_key: SecretStr | None = None
 ) -> tuple[bool, str]:

prediction_prophet/functions/rephrase_question.py

@@ -25,7 +25,7 @@ class RephrasedQuestion(BaseModel):
 
 def rephrase_question(
     question: str,
-    engine: str = "gpt-4-0125-preview"
+    engine: str = "gpt-4-turbo"
 ) -> RephrasedQuestion:
     """
     Rephrase the original question, by asking it in negation and universally, for example:

prediction_prophet/functions/rerank_results.py

@@ -19,7 +19,7 @@ def rerank_results(results: list[str], goal: str) -> list[str]:
 
     rerank_results_chain = (
         rerank_results_prompt |
-        ChatOpenAI(model="gpt-4-0125-preview") |
+        ChatOpenAI(model="gpt-4-turbo") |
         CommaSeparatedListOutputParser()
     )
 

prediction_prophet/functions/research.py

@@ -13,7 +13,7 @@
 def research(
     goal: str,
     use_summaries: bool,
-    model: str = "gpt-4-0125-preview",
+    model: str = "gpt-4-turbo",
     initial_subqueries_limit: int = 20,
     subqueries_limit: int = 4,
     scrape_content_split_chunk_size: int = 800,

prediction_prophet/main.py

@@ -40,7 +40,7 @@ def research(
     start = time.time()
 
     with get_openai_callback() as cb:
-      report = prophet_research(goal=prompt, use_summaries=False, model="gpt-4-0125-preview")
+      report = prophet_research(goal=prompt, use_summaries=False, model="gpt-4-turbo")
 
     end = time.time()
 
@@ -66,7 +66,7 @@ def predict(prompt: str, path: str | None = None) -> None:
         else:
             logger = logging.getLogger("research")
             logger.setLevel(logging.INFO)
-            report = prophet_research(goal=prompt, model="gpt-4-0125-preview", use_summaries=False, logger=logger)
+            report = prophet_research(goal=prompt, model="gpt-4-turbo", use_summaries=False, logger=logger)
 
         prediction = make_debated_prediction(prompt=prompt, additional_information=report)
 

scripts/benchmark.py

@@ -83,7 +83,7 @@ def main(
                 agent_name="prediction_prophet_gpt-3.5-turbo-0125_tavilyrawcontent",
                 use_tavily_raw_content=True,
             ),
-            # PredictionProphetAgent(model="gpt-4-0125-preview", max_workers=max_workers, agent_name="prediction_prophet_gpt-4-0125-preview"),  # Too expensive to be enabled by default.
+            # PredictionProphetAgent(model="gpt-4-turbo", max_workers=max_workers, agent_name="prediction_prophet_gpt-4-turbo"),  # Too expensive to be enabled by default.
         ],
         cache_path=cache_path,
         only_cached=only_cached,