Fix deprecated OpenAI v0 API and add requirements.txt

README.md

@@ -4,24 +4,35 @@
 
 This repo contains the official code for running `LLM persona` experiments and subsequent analyses in the PersonaLLM paper.
 
+## Setup
+
+```bash
+pip install -r requirements.txt
+```
+
+Set your API keys as environment variables before running any scripts:
+
+```bash
+export OPENAI_API_KEY="your-openai-api-key"
+export REPLICATE_API_TOKEN="your-replicate-token"  # only needed for llama-2
+```
 
 ## Simulate `LLM personas`
 
 We first create 10 personas for each of 32 personality types.
 
 ```bash
-conda activate audiencenlp
-python3.9 run_bfi.py --model "GPT-3.5-turbo-0613"
-python3.9 run_bfi.py --model "GPT-4-0613"
-python3.9 run_bfi.py --model "llama-2"
+python run_bfi.py --model "gpt-3.5-turbo-0613"
+python run_bfi.py --model "gpt-4-0613"
+python run_bfi.py --model "llama-2"
 ```
 
 ## Generate stories with `LLM personas`
 
 ```bash
-python3.9 run_creative_writing.py --model "GPT-3.5-turbo-0613"
-python3.9 run_creative_writing.py --model "GPT-4-0613"
-python3.9 run_creative_writing.py --model "llama-2"
+python run_creative_writing.py --model "gpt-3.5-turbo-0613"
+python run_creative_writing.py --model "gpt-4-0613"
+python run_creative_writing.py --model "llama-2"
 ```
 
 ## References

requirements.txt

@@ -0,0 +1,7 @@
+openai>=1.0.0
+replicate
+tqdm
+pandas
+numpy
+scipy
+tenacity

run_bfi.py

@@ -1,7 +1,6 @@
 import argparse
 import random
 import json
-import openai
 import os
 import pandas as pd
 import sys
@@ -11,13 +10,13 @@
 import json
 import numpy as np
 import itertools
+from openai import OpenAI
 from gpt import is_answer_in_valid_form
 from tenacity import retry, stop_after_attempt, wait_random_exponential
 import multiprocessing
 import replicate
 
-openai.organization = ""
-openai.api_key = ""
+client = OpenAI()  # reads OPENAI_API_KEY from environment
 
 
 def construct_big_five_words(persona_type):
@@ -31,7 +30,7 @@ def construct_big_five_words(persona_type):
     return ", ".join(options)
 
 def run_gpt_query(model_name, temperature, system_prompt, user_prompt):
-    response = openai.ChatCompletion.create(
+    response = client.chat.completions.create(
         model=model_name,
         messages=[
             {"role": "system", "content": system_prompt},
@@ -80,7 +79,7 @@ def generate_bfi_response(model_name, temperature, persona_type, prompt_file):
     while True:
         if model_name.lower().startswith("gpt"):
             response = run_gpt_query(model_name, temperature, system_prompt, user_prompt)
-            response = response["choices"][0]['message']['content'].strip("\n")
+            response = response.choices[0].message.content.strip("\n")
         elif model_name.lower().startswith("llama"):
             # print("llama-2 generating")
             # print(system_prompt, user_prompt)

run_creative_writing.py

@@ -1,7 +1,6 @@
 import argparse
 import random
 import json
-import openai
 import os
 import pandas as pd
 import sys
@@ -11,13 +10,13 @@
 import json
 import numpy as np
 import itertools
+from openai import OpenAI
 # from gpt import run_completion_query
 from tenacity import retry, stop_after_attempt, wait_random_exponential
 import multiprocessing
 import replicate
 
-openai.organization = ""
-openai.api_key = ""
+client = OpenAI()  # reads OPENAI_API_KEY from environment
 
 def construct_big_five_words(persona_type: list):
     """Construct the list of personality traits
@@ -30,7 +29,7 @@ def construct_big_five_words(persona_type: list):
     return ", ".join(options)
 
 def run_gpt_query(model_name, temperature, system_prompt, prev_user_prompt, prev_assistant_prompt, user_prompt):
-    response = openai.ChatCompletion.create(
+    response = client.chat.completions.create(
         model=model_name,
         messages=[
             {"role": "system", "content": system_prompt},
@@ -85,7 +84,7 @@ def generate_bfi_story(model_name, temperature, persona_type, prompt_file, json_
 
     if model_name.lower().startswith("gpt"):
         response = run_gpt_query(model_name, temperature, system_prompt, prev_user_prompt, prev_assistant_prompt, user_prompt)
-        response = response["choices"][0]['message']['content'].strip("\n")
+        response = response.choices[0].message.content.strip("\n")
     elif model_name.lower().startswith("llama"):
         response = run_llama2_query(temperature, system_prompt, prev_user_prompt, prev_assistant_prompt, user_prompt)
         response = "".join([each for each in response]).strip("\n").strip()