a

ape/ape.py

@@ -6,8 +6,6 @@ from ape.llm import LLMGPT
 from functools import partial
 from itertools import chain
 
-LLM = None
-
 
 def load_task(task, file):
     global instance
@@ -36,9 +34,7 @@ def esttimate_cost(instance):
 
 
 def generate(gen_prompt, instance, openai_key):
-    global LLM
-    if LLM is None:
-        LLM = LLMGPT(openai_key)
+    LLM = LLMGPT(openai_key)
 
     instructions = []
     train_iter = instance.get_train_iter()
@@ -50,9 +46,7 @@ def generate(gen_prompt, instance, openai_key):
 
 
 def single_test(test_prompt, instruction, input, openai_key):
-    global LLM
-    if LLM is None:
-        LLM = LLMGPT(openai_key)
+    LLM = LLMGPT(openai_key)
     output = []
     for i in input.split('\n'):
         output.append(LLM.generate_output(test_prompt, instruction, i)['text'])
@@ -60,9 +54,7 @@ def single_test(test_prompt, instruction, input, openai_key):
 
 
 def score_single(eval_prompt, instance, instruction, openai_key):
-    global LLM
-    if LLM is None:
-        LLM = LLMGPT(openai_key)
+    LLM = LLMGPT(openai_key)
 
     score = LLM.generate_logprobs(eval_prompt, instruction, instance.eval_samples)
     return score

ape/llm.py

@@ -21,7 +21,6 @@ Cost = {
 
 class LLMGPT(object):
     encoding = tiktoken.get_encoding("cl100k_base")
-
     def __init__(self, openai_key):
         self.gen_llm = ChatOpenAI(openai_api_key=openai_key, max_tokens=2000, temperature=0.7, verbose=True)
         self.eval_llm = OpenAI(openai_api_key=openai_key, max_tokens=0, temperature=0.7, echo=True, logprobs=1)