#38 Groundwork for migration to new openai >=v1.0.0 python API

factscore/atomic_facts.py

@@ -6,7 +6,6 @@
 import spacy
 import sys
 import nltk
-import openai
 from rank_bm25 import BM25Okapi
 import os
 import time

factscore/openai_lm.py

@@ -1,16 +1,18 @@
 from factscore.lm import LM
-import openai
+from openai import OpenAI, BadRequestError
 import sys
 import time
 import os
 import numpy as np
 import logging
 
+
 class OpenAIModel(LM):
 
     def __init__(self, model_name, cache_file=None, key_path="api.key"):
         self.model_name = model_name
         self.key_path = key_path
+        self.client = None  # Initialized with load_model() method
         self.temp = 0.7
         self.save_interval = 100
         super().__init__(cache_file)
@@ -21,7 +23,7 @@ def load_model(self):
         assert os.path.exists(key_path), f"Please place your OpenAI APT Key in {key_path}."
         with open(key_path, 'r') as f:
             api_key = f.readline()
-        openai.api_key = api_key.strip()
+        self.client = OpenAI(api_key=api_key.strip())
         self.model = self.model_name
 
     def _generate(self, prompt, max_sequence_length=2048, max_output_length=128):
@@ -33,65 +35,71 @@ def _generate(self, prompt, max_sequence_length=2048, max_output_length=128):
             # Construct the prompt send to ChatGPT
             message = [{"role": "user", "content": prompt}]
             # Call API
-            response = call_ChatGPT(message, temp=self.temp, max_len=max_sequence_length)
+            response = call_ChatGPT(message, self.client, temp=self.temp, max_len=max_sequence_length)
             # Get the output from the response
             output = response["choices"][0]["message"]["content"]
             return output, response
         elif self.model_name == "InstructGPT":
             # Call API
-            response = call_GPT3(prompt, temp=self.temp)
+            response = call_GPT3(prompt, self.client, temp=self.temp)
             # Get the output from the response
-            output = response["choices"][0]["text"]
+            output = response.choices[0].message.content
             return output, response
         else:
             raise NotImplementedError()
 
-def call_ChatGPT(message, model_name="gpt-3.5-turbo", max_len=1024, temp=0.7, verbose=False):
+
+def call_ChatGPT(message, openai_client, model_name="gpt-3.5-turbo", max_len=1024, temp=0.7, verbose=False):
     # call GPT-3 API until result is provided and then return it
     response = None
     received = False
     num_rate_errors = 0
     while not received:
         try:
-            response = openai.ChatCompletion.create(model=model_name,
-                                                    messages=message,
-                                                    max_tokens=max_len,
-                                                    temperature=temp)
+            response = openai_client.chat.completions.create(
+                model=model_name,
+                messages=message,
+                max_tokens=max_len,
+                temperature=temp)
             received = True
         except:
             # print(message)
             num_rate_errors += 1
             error = sys.exc_info()[0]
-            if error == openai.error.InvalidRequestError:
+            if error == BadRequestError:
                 # something is wrong: e.g. prompt too long
-                logging.critical(f"InvalidRequestError\nPrompt passed in:\n\n{message}\n\n")
+                logging.critical(f"BadRequestError\nPrompt passed in:\n\n{message}\n\n")
                 assert False
-            
+
             logging.error("API error: %s (%d). Waiting %dsec" % (error, num_rate_errors, np.power(2, num_rate_errors)))
             time.sleep(np.power(2, num_rate_errors))
     return response
 
 
-def call_GPT3(prompt, model_name="text-davinci-003", max_len=512, temp=0.7, num_log_probs=0, echo=False, verbose=False):
+def call_GPT3(prompt, openai_client, model_name="gpt-3.5-turbo-0125", max_len=512, temp=0.7, num_log_probs=0):
     # call GPT-3 API until result is provided and then return it
     response = None
     received = False
     num_rate_errors = 0
     while not received:
         try:
-            response = openai.Completion.create(model=model_name,
-                                                prompt=prompt,
-                                                max_tokens=max_len,
-                                                temperature=temp,
-                                                logprobs=num_log_probs,
-                                                echo=echo)
+            response = openai_client.chat.completions.create(
+                model=model_name,
+                messages=[
+                    {"role": "user", "content": prompt},
+                ],
+                max_tokens=max_len,
+                temperature=temp,
+                logprobs=num_log_probs > 0,  # Needs to be True if num_log_probs > 0
+                top_logprobs=num_log_probs,
+            )
             received = True
         except:
             error = sys.exc_info()[0]
             num_rate_errors += 1
-            if error == openai.error.InvalidRequestError:
+            if error == BadRequestError:
                 # something is wrong: e.g. prompt too long
-                logging.critical(f"InvalidRequestError\nPrompt passed in:\n\n{prompt}\n\n")
+                logging.critical(f"BadRequestError\nPrompt passed in:\n\n{prompt}\n\n")
                 assert False
             logging.error("API error: %s (%d)" % (error, num_rate_errors))
             time.sleep(np.power(2, num_rate_errors))

preprocessing/preprocess_acl.py

@@ -1,7 +1,7 @@
 import pandas as pd
 import tqdm
 import json
-import openai
+from openai import OpenAI
 from factscore.openai_lm import call_ChatGPT
 from factscore.factscorer import FactScorer
 
@@ -49,12 +49,12 @@
 
 with open("api.key", 'r') as f:
     api_key = f.readline()
-openai.api_key = api_key.strip()
+openai_client = OpenAI(api_key=api_key.strip())
 
 responses = []
 for ptitle, prompt in tqdm.tqdm(zip(prompt_titles, prompts_list)):
     message = [{"role": "user", "content": prompt}]
-    response = call_ChatGPT(message, model_name="gpt-3.5-turbo-0301")
+    response = call_ChatGPT(message, openai_client, model_name="gpt-3.5-turbo-0301")
     responses.append({
         "topic": ptitle,
         "output": response["choices"][0]["message"]["content"]