init

experiments/baseline_gpt4.py

@@ -4,9 +4,9 @@ import pandas as pd
 import openai
 from datasets import load_dataset
 
-data = load_dataset("cardiffnlp/relentless_full", split="test")
+data = load_dataset("cardiffnlp/relentless", split="test")
 openai.api_key = os.getenv("OPENAI_API_KEY", None)
-pretty_name = {"is competitor/rival of": "Rival", "is friend/ally of": "Ally", "is influenced by": "Inf", "is known for": "Know", "is similar to": "Sim"}
+pretty_name = {"competitor/rival of": "Rival", "friend/ally of": "Ally", "influenced by": "Inf", "known for": "Know", "similar to": "Sim"}
 pretty_model = {"gpt-3.5-turbo": "GPT-3.5", "gpt-4": "GPT-4"}
 
 
@@ -22,16 +22,16 @@ def get_reply(model, text):
 
 
 prompt_dict = {
-    "is friend/ally of": "entities that are friends or allies",
-    "is competitor/rival of": "entities that are competitors or rivals",
-    "is known for": "what entities are known for",
-    "is influenced by": "what has influenced different entities",
-    "is similar to": "entities that are similar"
+    "friend/ally of": "entities that are friends or allies",
+    "competitor/rival of": "entities that are competitors or rivals",
+    "known for": "what entities are known for",
+    "influenced by": "what has influenced different entities",
+    "similar to": "entities that are similar"
 }
 
 
 def get_prompt(_data):
-    ref = "\n".join([str(_i) for _i in _data["positive_examples"]])
+    ref = "\n".join([str(_i) for _i in _data["prototypical_examples"]])
     prefix = f'Consider the following reference list of {prompt_dict[_data["relation_type"]]}, \n{ref}\n' \
              f'Now sort the entity pairs from the following list based on the extent to which they also represent ' \
              f'{prompt_dict[_data["relation_type"]]} in descending order. Do not include the pairs from the reference list. ' \
@@ -41,14 +41,14 @@ def get_prompt(_data):
 
 
 if __name__ == '__main__':
-    os.makedirs('experiments/results/chat', exist_ok=True)
+    os.makedirs('results/chat', exist_ok=True)
 
     full_result = []
     valid_count = []
     for target_model in ['gpt-3.5-turbo', 'gpt-4']:
 
         for d in data:
-            output_file = f"experiments/results/chat/{target_model}.{d['relation_type'].replace(' ', '_').replace('/', '-')}.json"
+            output_file = f"results/chat/{target_model}.{d['relation_type'].replace(' ', '_').replace('/', '-')}.json"
             if not os.path.exists(output_file):
                 print(target_model, d['relation_type'])
                 i = get_prompt(d)
@@ -96,5 +96,5 @@ if __name__ == '__main__':
     df.index = [pretty_model[m] for m in df.index]
     print(df.to_latex())
     df = df.T
-    # df.to_csv("experiments/results/chat/chat.csv")
+    # df.to_csv("results/chat/chat.csv")
 

experiments/baseline_lm_lc.py

@@ -10,11 +10,11 @@ from lmppl import EncoderDecoderLM, LM, OpenAI
 OPENAI_API_KEY = os.getenv("OPENAI_API_KEY", None)
 
 prompt_dict = {
-    "is friend/ally of": "Complete the following list with examples of entities that are friends or allies",
-    "is competitor/rival of": "Complete the following list with examples of entities that are competitors or rivals",
-    "is known for": "Complete the following list with examples of what entities are known for",
-    "is influenced by": "Complete the following list with examples of what has influenced different entities",
-    "is similar to": "Complete the following list with examples of entities that are similar"
+    "friend/ally of": "Complete the following list with examples of entities that are friends or allies",
+    "competitor/rival of": "Complete the following list with examples of entities that are competitors or rivals",
+    "known for": "Complete the following list with examples of what entities are known for",
+    "influenced by": "Complete the following list with examples of what has influenced different entities",
+    "similar to": "Complete the following list with examples of entities that are similar"
 }
 data = load_dataset("cardiffnlp/relentless", split="test")
 full_result = []

experiments/baseline_lm_qa.py

@@ -10,39 +10,39 @@ from lmppl import EncoderDecoderLM, LM, OpenAI
 OPENAI_API_KEY = os.getenv("OPENAI_API_KEY", None)
 
 prompt_dict = {
-    "is friend/ally of": "entities that are friends or allies",
-    "is competitor/rival of": "entities that are competitors or rivals",
-    "is known for": "examples of what entities are known for",
-    "is influenced by": "what has influenced different entities",
-    "is similar to": "examples of entities that are similar"
+    "friend/ally of": "entities that are friends or allies",
+    "competitor/rival of": "entities that are competitors or rivals",
+    "known for": "examples of what entities are known for",
+    "influenced by": "what has influenced different entities",
+    "similar to": "examples of entities that are similar"
 }
 data = load_dataset("cardiffnlp/relentless", split="test")
 full_result = []
 for lm, ppl_class, batch, pretty_name in [
-    ("google/flan-ul2", EncoderDecoderLM, 1, "Flan-UL2"),
-    ("google/flan-t5-xxl", EncoderDecoderLM, 1, "Flan-T5\textsubscript{XXL}"),
-    ("google/flan-t5-xl", EncoderDecoderLM, 1, "Flan-T5\textsubscript{XL}"),
-    ("google/flan-t5-large", EncoderDecoderLM, 32, "Flan-T5\textsubscript{LARGE}"),
-    ("google/flan-t5-base", EncoderDecoderLM, 128, "Flan-T5\textsubscript{BASE}"),
-    ("google/flan-t5-small", EncoderDecoderLM, 256, "Flan-T5\textsubscript{SMALL}"),
-    ("t5-11b", EncoderDecoderLM, 1, "T5\textsubscript{XXL}"),
-    ("t5-3b", EncoderDecoderLM, 1, "T5\textsubscript{XL}"),
-    ("t5-large", EncoderDecoderLM, 32, "T5\textsubscript{LARGE}"),
-    ("t5-base", EncoderDecoderLM, 128, "T5\textsubscript{BASE}"),
-    ("t5-small", EncoderDecoderLM, 256, "T5\textsubscript{SMALL}"),
-    # ("facebook/opt-66b", LM, 1, "OPT\textsubscript{66B}"),
-    ("facebook/opt-30b", LM, 1, "OPT\textsubscript{30B}"),
-    ("facebook/opt-13b", LM, 1, "OPT\textsubscript{13B}"),
-    ("facebook/opt-6.7b", LM, 1, "OPT\textsubscript{6.7B}"),
-    ("facebook/opt-2.7b", LM, 1, "OPT\textsubscript{2.7B}"),
-    ("facebook/opt-1.3b", LM, 1, "OPT\textsubscript{1.3B}"),
-    ("facebook/opt-350m", LM, 128, "OPT\textsubscript{350M}"),
-    ("facebook/opt-125m", LM, 256, "OPT\textsubscript{125M}"),
-    ("facebook/opt-iml-30b", LM, 1, "OPT-IML\textsubscript{30B}"),
-    ("facebook/opt-iml-1.3b", LM, 1, "OPT-IML\textsubscript{1.3B}"),
-    ("facebook/opt-iml-max-30b", LM, 1, "OPT-IML\textsubscript{MAX-30B}"),
-    ("facebook/opt-iml-max-1.3b", LM, 1, "OPT-IML\textsubscript{MAX-1.3B}"),
-    # ("davinci", OpenAI, None, "GPT-3\textsubscript{davinci}")
+    # ("google/flan-ul2", EncoderDecoderLM, 1, "Flan-UL2"),
+    # ("google/flan-t5-xxl", EncoderDecoderLM, 1, "Flan-T5\textsubscript{XXL}"),
+    # ("google/flan-t5-xl", EncoderDecoderLM, 1, "Flan-T5\textsubscript{XL}"),
+    # ("google/flan-t5-large", EncoderDecoderLM, 32, "Flan-T5\textsubscript{LARGE}"),
+    # ("google/flan-t5-base", EncoderDecoderLM, 128, "Flan-T5\textsubscript{BASE}"),
+    # ("google/flan-t5-small", EncoderDecoderLM, 256, "Flan-T5\textsubscript{SMALL}"),
+    # ("t5-11b", EncoderDecoderLM, 1, "T5\textsubscript{XXL}"),
+    # ("t5-3b", EncoderDecoderLM, 1, "T5\textsubscript{XL}"),
+    # ("t5-large", EncoderDecoderLM, 32, "T5\textsubscript{LARGE}"),
+    # ("t5-base", EncoderDecoderLM, 128, "T5\textsubscript{BASE}"),
+    # ("t5-small", EncoderDecoderLM, 256, "T5\textsubscript{SMALL}"),
+    # # ("facebook/opt-66b", LM, 1, "OPT\textsubscript{66B}"),
+    # ("facebook/opt-30b", LM, 1, "OPT\textsubscript{30B}"),
+    # ("facebook/opt-13b", LM, 1, "OPT\textsubscript{13B}"),
+    # ("facebook/opt-6.7b", LM, 1, "OPT\textsubscript{6.7B}"),
+    # ("facebook/opt-2.7b", LM, 1, "OPT\textsubscript{2.7B}"),
+    # ("facebook/opt-1.3b", LM, 1, "OPT\textsubscript{1.3B}"),
+    # ("facebook/opt-350m", LM, 128, "OPT\textsubscript{350M}"),
+    # ("facebook/opt-125m", LM, 256, "OPT\textsubscript{125M}"),
+    # ("facebook/opt-iml-30b", LM, 1, "OPT-IML\textsubscript{30B}"),
+    # ("facebook/opt-iml-1.3b", LM, 1, "OPT-IML\textsubscript{1.3B}"),
+    # ("facebook/opt-iml-max-30b", LM, 1, "OPT-IML\textsubscript{MAX-30B}"),
+    # ("facebook/opt-iml-max-1.3b", LM, 1, "OPT-IML\textsubscript{MAX-1.3B}"),
+    ("davinci", OpenAI, None, "GPT-3\textsubscript{davinci}")
 ]:
     os.makedirs(f"results/lm_qa/{os.path.basename(lm)}", exist_ok=True)
     scorer = None

experiments/results/chat/gpt-3.5-turbo.competitor-rival_of.json

@@ -0,0 +1,19 @@
+1. Liverpool FC - Manchester United
+2. Apple - Microsoft
+3. Microsoft - Google
+4. Netflix - Disney Plus
+5. PyTorch - TensorFlow
+6. Razer - Dell
+7. Spotify - Apple
+8. Manchester United - Arsenal
+9. Saudi Arabia - Israel
+10. Coca-Cola Company - Pepsi
+11. Twitter - Facebook
+12. Arsenal - Tottenham Hotspur
+13. Nintendo - Xbox
+14. Liverpool FC - Manchester City
+15. Nike - Adidas
+16. Manchester City - Manchester United
+17. Amazon - Ebay
+18. McDonald's - Burger King
+19. Sprite - 7 Up

experiments/results/chat/gpt-3.5-turbo.friend-ally_of.json

@@ -0,0 +1,79 @@
+1. ['Australia', 'New Zealand']
+2. ['Aznar', 'Bush']
+3. ['Extinction Rebellion', 'Greta Thunberg']
+4. ['Elsa', 'Anna']
+5. ['CIA', 'MI6']
+6. ['UK', 'Commonwealth']
+7. ['Singapore', 'Israel']
+8. ['India', 'US']
+9. ['US', 'Canada']
+10. ['UN', 'NATO']
+11. ['Germany', 'France']
+12. ['Spain', 'Portugal']
+13. ['France', 'Belgium']
+14. ['Malaysia', 'Singapore']
+15. ['Islamic State', 'Denys Prokopenko']
+16. ['China', 'North Korea']
+17. ['Armenia', 'Azerbaijan']
+18. ['Russia', 'Georgia']
+19. ['Japan', 'Taiwan']
+20. ['South Korea', 'Japan']
+21. ['UK', 'Ireland']
+22. ['Turkey', 'All Nippon Airways']
+23. ['Pedro Sánchez', 'Pablo Iglesias']
+24. ['Hillary Clinton', 'Barack Obama']
+25. ['Rishi Sunak', 'Leo Varadkar']
+26. ['Boris Johnson', 'Emmanuel Macron']
+27. ['Di Maio', 'Salvini']
+28. ['Brazil', 'India']
+29. ['Kylo Ren', 'Rey']
+30. ['Keir Starmer', 'Jeremy Corbyn']
+31. ['Margaret Thatcher', 'Ronald Reagan']
+32. ['Singapore', 'Malaysia'] (Removing duplicate)
+33. ['UK', 'Australia'] (Removing duplicate)
+34. ['Cersei Lannister', 'Euron Greyjoy']
+35. ['FTX', 'Alameda Research']
+36. ['Sophia Loren', 'Marlon Brando']
+37. ['Paul Rudd', 'Memento']
+38. ['Jean-Michel Basquiat', 'Andy Warhol']
+39. ['India', 'Brazil']
+40. ['Nikon', 'Tokina']
+41. ['Google', 'Samsung']
+42. ['IMF', 'The World Bank']
+43. ['Instagram', 'WhatsApp']
+44. ['Windows', 'Xbox']
+45. ['Johnny Cash', 'Waylon Jennings']
+46. ['Oman', 'Iran']
+47. ['China', 'Huawei']
+48. ['Amazon', 'Royal Mail']
+49. ['Red Bull', 'GoPro']
+50. ['HSBC', 'BlackRock']
+51. ['Tata Motors', 'Jaguar']
+52. ['KGB', 'CIA']
+53. ['JP Morgan', 'Morgan Stanley']
+54. ['Eva Perón', 'Interpol']
+55. ['Eastern Orthodoxy', 'Oriental Orthodoxy']
+56. ['Darth Vader', 'Emperor Palpatine']
+57. ['Doja Cat', 'Anthony Albanese']
+58. ['Thomas Jefferson', 'Kid Cudi']
+59. ['Liam Gallagher', 'Noel Gallagher']
+60. ['Quentin Tarantino', 'Edgar Wright']
+61. ['Rishi Sunak', 'Joe Biden']
+62. ['Macbeth', 'Banquo']
+63. ['Ron Weasley', 'Neville Longbottom']
+64. ['Bob Marley', 'Abu Bakr']
+65. ['Noah Schnapp', 'Galatasaray S.K.']
+66. ['Kendall Jenner', 'Bergen']
+67. ['Porter Wagoner', 'Dolly Parton']
+68. ['Stephen Hawking', 'Brian Cox']
+69. ['Johnny Knoxville', 'Catherine Zeta-Jones']
+70. ['Mark Drakeford', 'Rishi Sunak']
+71. ['J.R.R. Tolkien', 'C.S. Lewis']
+72. ['Beatles', 'Rolling Stones']
+73. ['Benedict Cumberbatch', 'Hanukkah']
+74. ['United States', 'United Kingdom']
+75. ['Linus Sebastian', 'Marques Brownlee']
+76. ['Saturn', 'Rachel Bilson']
+77. ['Huawei', 'China']
+78. ['Achilles', 'Jonathan Bailey']
+79. ['The Beatles', 'Queen']