init

experiments/baseline_lm_lc.py

@@ -19,29 +19,29 @@ prompt_dict = {
 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}"),
+    ("google/flan-t5-base", EncoderDecoderLM, 128, "Flan-T5\textsubscript{BASE}"),
+    ("google/flan-t5-large", EncoderDecoderLM, 32, "Flan-T5\textsubscript{LARGE}"),
+    ("google/flan-t5-xl", EncoderDecoderLM, 1, "Flan-T5\textsubscript{XL}"),
+    ("google/flan-t5-xxl", EncoderDecoderLM, 1, "Flan-T5\textsubscript{XXL}"),
+    ("google/flan-ul2", EncoderDecoderLM, 1, "Flan-UL2"),
     ("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}"),
+    ("t5-base", EncoderDecoderLM, 128, "T5\textsubscript{BASE}"),
+    ("t5-large", EncoderDecoderLM, 32, "T5\textsubscript{LARGE}"),
+    ("t5-3b", EncoderDecoderLM, 1, "T5\textsubscript{XL}"),
+    ("t5-11b", EncoderDecoderLM, 1, "T5\textsubscript{XXL}"),
     ("facebook/opt-125m", LM, 256, "OPT\textsubscript{125M}"),
-    ("facebook/opt-iml-30b", LM, 1, "OPT-IML\textsubscript{30B}"),
+    ("facebook/opt-350m", LM, 128, "OPT\textsubscript{350M}"),
+    ("facebook/opt-1.3b", LM, 1, "OPT\textsubscript{1.3B}"),
+    ("facebook/opt-2.7b", LM, 1, "OPT\textsubscript{2.7B}"),
+    ("facebook/opt-6.7b", LM, 1, "OPT\textsubscript{6.7B}"),
+    ("facebook/opt-13b", LM, 1, "OPT\textsubscript{13B}"),
+    ("facebook/opt-30b", LM, 1, "OPT\textsubscript{30B}"),
+    # ("facebook/opt-66b", LM, 1, "OPT\textsubscript{66B}"),
     ("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}"),
+    ("facebook/opt-iml-30b", LM, 1, "OPT-IML\textsubscript{30B}"),
+    ("facebook/opt-iml-max-1.3b", LM, 1, "OPT-IML\textsubscript{M-1.3B}"),
+    ("facebook/opt-iml-max-30b", LM, 1, "OPT-IML\textsubscript{M-30B}"),
     ("davinci", OpenAI, None, "GPT-3\textsubscript{davinci}")
 ]:
     os.makedirs(f"results/lm_lc/{os.path.basename(lm)}", exist_ok=True)
@@ -98,3 +98,4 @@ df.to_csv("results/lm_lc/lm.csv")
 df = (100 * df).round()
 print(df.to_markdown())
 print(df.to_latex(escape=False))
+

experiments/baseline_lm_qa.py

@@ -19,29 +19,29 @@ prompt_dict = {
 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}"),
+    ("google/flan-t5-base", EncoderDecoderLM, 128, "Flan-T5\textsubscript{BASE}"),
+    ("google/flan-t5-large", EncoderDecoderLM, 32, "Flan-T5\textsubscript{LARGE}"),
+    ("google/flan-t5-xl", EncoderDecoderLM, 1, "Flan-T5\textsubscript{XL}"),
+    ("google/flan-t5-xxl", EncoderDecoderLM, 1, "Flan-T5\textsubscript{XXL}"),
+    ("google/flan-ul2", EncoderDecoderLM, 1, "Flan-UL2"),
     ("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}"),
+    ("t5-base", EncoderDecoderLM, 128, "T5\textsubscript{BASE}"),
+    ("t5-large", EncoderDecoderLM, 32, "T5\textsubscript{LARGE}"),
+    ("t5-3b", EncoderDecoderLM, 1, "T5\textsubscript{XL}"),
+    ("t5-11b", EncoderDecoderLM, 1, "T5\textsubscript{XXL}"),
     ("facebook/opt-125m", LM, 256, "OPT\textsubscript{125M}"),
-    ("facebook/opt-iml-30b", LM, 1, "OPT-IML\textsubscript{30B}"),
+    ("facebook/opt-350m", LM, 128, "OPT\textsubscript{350M}"),
+    ("facebook/opt-1.3b", LM, 1, "OPT\textsubscript{1.3B}"),
+    ("facebook/opt-2.7b", LM, 1, "OPT\textsubscript{2.7B}"),
+    ("facebook/opt-6.7b", LM, 1, "OPT\textsubscript{6.7B}"),
+    ("facebook/opt-13b", LM, 1, "OPT\textsubscript{13B}"),
+    ("facebook/opt-30b", LM, 1, "OPT\textsubscript{30B}"),
+    # ("facebook/opt-66b", LM, 1, "OPT\textsubscript{66B}"),
     ("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}"),
+    ("facebook/opt-iml-30b", LM, 1, "OPT-IML\textsubscript{30B}"),
+    ("facebook/opt-iml-max-1.3b", LM, 1, "OPT-IML\textsubscript{M-1.3B}"),
+    ("facebook/opt-iml-max-30b", LM, 1, "OPT-IML\textsubscript{M-30B}"),
     ("davinci", OpenAI, None, "GPT-3\textsubscript{davinci}")
 ]:
     os.makedirs(f"results/lm_qa/{os.path.basename(lm)}", exist_ok=True)

experiments/baseline_lm_qa_fewshot.py

@@ -26,9 +26,9 @@ for shots in shots_num:
 
 full_result = []
 for lm, ppl_class, batch, pretty_name in [
-            ("google/flan-ul2", EncoderDecoderLM, 1, "Flan-UL2"),
+            # ("google/flan-ul2", EncoderDecoderLM, 1, "Flan-UL2"),
             # ("google/flan-t5-xxl", EncoderDecoderLM, 1, "Flan-T5\textsubscript{XXL}"),
-            # ("facebook/opt-13b", LM, 1, "OPT\textsubscript{13B}"),
+            ("facebook/opt-13b", LM, 1, "OPT\textsubscript{13B}"),
             ("davinci", OpenAI, None, "GPT-3\textsubscript{davinci}")
         ]:
     scorer = None

experiments/results/lm_lc/lm.csv

@@ -1,24 +1,24 @@
 model,competitor/rival of,friend/ally of,influenced by,known for,similar to,average
-Flan-UL2,0.6605035317658581,0.35791417582293494,0.5329592730844052,0.5617353599147163,0.4369621165115711,0.5100148914198971
-Flan-T5	extsubscript{XXL},0.6558600331708642,0.39097532739732255,0.5137148714722951,0.48498322027473,0.46526989625821613,0.5021606697146856
-Flan-T5	extsubscript{XL},0.4948996416160854,0.2585068655322593,0.40334740581651923,0.3234891964665975,0.38715447342514303,0.3734795165713209
-Flan-T5	extsubscript{LARGE},0.500097970006437,0.2791015777434352,0.41810593292697185,0.3014720569753186,0.2944442528504654,0.3586443581005256
-Flan-T5	extsubscript{BASE},0.4566419932654189,0.3899872752372662,0.3459207357378026,0.23634179781685494,0.10649793272688465,0.3070779469568455
 Flan-T5	extsubscript{SMALL},0.47676097522939054,0.40833571261581375,0.30934590563428976,0.22436672671700258,0.16942071664102182,0.3176460073675037
-T5	extsubscript{XXL},0.3078963930515155,0.10340583387590391,0.2646516426827772,0.21968933744723163,0.168830348478407,0.212894711107167
-T5	extsubscript{XL},0.5446721421811761,0.36070079168059405,0.5455346746040511,0.38386092796999827,0.3385425126933779,0.43466220982583953
-T5	extsubscript{LARGE},0.36104055160378007,0.16276615505429307,0.3156410828516156,0.174686170687364,0.29006506622651235,0.260839805284713
-T5	extsubscript{BASE},0.4484817273521575,0.4231564950166599,0.4386362852295569,0.25739482237701783,0.22215778149637408,0.35796542229435324
+Flan-T5	extsubscript{BASE},0.4566419932654189,0.3899872752372662,0.3459207357378026,0.23634179781685494,0.10649793272688465,0.3070779469568455
+Flan-T5	extsubscript{LARGE},0.500097970006437,0.2791015777434352,0.41810593292697185,0.3014720569753186,0.2944442528504654,0.3586443581005256
+Flan-T5	extsubscript{XL},0.4948996416160854,0.2585068655322593,0.40334740581651923,0.3234891964665975,0.38715447342514303,0.3734795165713209
+Flan-T5	extsubscript{XXL},0.6558600331708642,0.39097532739732255,0.5137148714722951,0.48498322027473,0.46526989625821613,0.5021606697146856
+Flan-UL2,0.6605035317658581,0.35791417582293494,0.5329592730844052,0.5617353599147163,0.4369621165115711,0.5100148914198971
 T5	extsubscript{SMALL},0.31788332936795455,0.4067532853282234,0.3070207273509307,0.17679576870495456,0.19644313785640424,0.2809792497216935
-OPT	extsubscript{30B},0.6813139170132756,0.3770885630540297,0.5854813870413097,0.6989690065205573,0.5716333147202012,0.5828972376698747
-OPT	extsubscript{13B},0.7212445905930207,0.4058964588456745,0.5814366235323154,0.7037084427908455,0.5978337043167529,0.6020239640157218
-OPT	extsubscript{6.7B},0.7080054980914848,0.42069408377401935,0.5996118917215304,0.6223886893435421,0.5280282738485872,0.5757456873558329
-OPT	extsubscript{2.7B},0.6617839082166836,0.4307367076820927,0.5980792179334641,0.588005105364532,0.4801860336326363,0.5517581945658818
-OPT	extsubscript{1.3B},0.6174316679600876,0.3921023243923869,0.5616838228100176,0.48487821765846983,0.47457379958043716,0.5061339664802799
-OPT	extsubscript{350M},0.48169469248657154,0.3571499792303913,0.5061636493945053,0.40226502289268884,0.272974281569549,0.40404952511474124
+T5	extsubscript{BASE},0.4484817273521575,0.4231564950166599,0.4386362852295569,0.25739482237701783,0.22215778149637408,0.35796542229435324
+T5	extsubscript{LARGE},0.36104055160378007,0.16276615505429307,0.3156410828516156,0.174686170687364,0.29006506622651235,0.260839805284713
+T5	extsubscript{XL},0.5446721421811761,0.36070079168059405,0.5455346746040511,0.38386092796999827,0.3385425126933779,0.43466220982583953
+T5	extsubscript{XXL},0.3078963930515155,0.10340583387590391,0.2646516426827772,0.21968933744723163,0.168830348478407,0.212894711107167
 OPT	extsubscript{125M},0.48600529320435076,0.409285168988368,0.5499158787006698,0.2905279206532933,0.22503489216228176,0.3921538307417927
-OPT-IML	extsubscript{30B},0.6440378906082419,0.3558763182428186,0.5676350829822163,0.6966494032704463,0.5200321481017788,0.5568461686411004
+OPT	extsubscript{350M},0.48169469248657154,0.3571499792303913,0.5061636493945053,0.40226502289268884,0.272974281569549,0.40404952511474124
+OPT	extsubscript{1.3B},0.6174316679600876,0.3921023243923869,0.5616838228100176,0.48487821765846983,0.47457379958043716,0.5061339664802799
+OPT	extsubscript{2.7B},0.6617839082166836,0.4307367076820927,0.5980792179334641,0.588005105364532,0.4801860336326363,0.5517581945658818
+OPT	extsubscript{6.7B},0.7080054980914848,0.42069408377401935,0.5996118917215304,0.6223886893435421,0.5280282738485872,0.5757456873558329
+OPT	extsubscript{13B},0.7212445905930207,0.4058964588456745,0.5814366235323154,0.7037084427908455,0.5978337043167529,0.6020239640157218
+OPT	extsubscript{30B},0.6813139170132756,0.3770885630540297,0.5854813870413097,0.6989690065205573,0.5716333147202012,0.5828972376698747
 OPT-IML	extsubscript{1.3B},0.6061643551928231,0.39118374464983446,0.5777955887284701,0.5394270768056251,0.4168298148649338,0.5062801160483372
-OPT-IML	extsubscript{MAX-30B},0.6125833091329617,0.3685743323310436,0.5747227646948349,0.6765032194870758,0.5111168415448232,0.5487000934381479
+OPT-IML	extsubscript{30B},0.6440378906082419,0.3558763182428186,0.5676350829822163,0.6966494032704463,0.5200321481017788,0.5568461686411004
 OPT-IML	extsubscript{MAX-1.3B},0.5909364112710048,0.3755292932389407,0.590692477920735,0.5158492166090246,0.40518685793640347,0.49563885139522174
+OPT-IML	extsubscript{MAX-30B},0.6125833091329617,0.3685743323310436,0.5747227646948349,0.6765032194870758,0.5111168415448232,0.5487000934381479
 GPT-3	extsubscript{davinci},0.70571789216601,0.3956531368425896,0.6475957959742583,0.7399152540158276,0.5197407005278296,0.6017245559053029

experiments/results/lm_qa/lm.csv

@@ -1,24 +1,24 @@
 model,competitor/rival of,friend/ally of,influenced by,known for,similar to,average
-Flan-UL2,0.8139011664177609,0.5668563310448644,0.5182231753464608,0.6982912623610599,0.6226590592054422,0.6439861988751175
-Flan-T5	extsubscript{XXL},0.7759252008862759,0.6082696110347289,0.4862164607771357,0.7200172582345266,0.6973966793865924,0.6575650420638519
-Flan-T5	extsubscript{XL},0.7861767483358856,0.5003943849660698,0.4883173453354119,0.6349030920632762,0.6740659274412318,0.6167714996283751
-Flan-T5	extsubscript{LARGE},0.7161401564757297,0.4571362263336,0.3097571107969417,0.5286213530232154,0.6074813663159394,0.5238272425890853
-Flan-T5	extsubscript{BASE},0.5034269487785833,0.36490001336083383,0.5049076045340413,0.2302182361504099,0.3121104596406103,0.3831126524928957
 Flan-T5	extsubscript{SMALL},0.41059965609373317,0.10640086698607491,0.2820643122067104,0.04446383513998588,0.17695351547539834,0.20409643718038054
-T5	extsubscript{XXL},0.025803853405636974,-0.06345919381112318,-0.023476076558673384,0.11729746805499064,-0.13017244334921085,-0.014801278451675962
-T5	extsubscript{XL},-0.00413988385766918,0.16291281904680144,-0.08980720752317983,0.16149879665432612,-0.19343151292559696,0.007406602278936319
-T5	extsubscript{LARGE},-0.030600997174729923,0.09892100336814785,-0.1485721634948909,-0.19286116899275968,0.037125937112284256,-0.04719747783638968
-T5	extsubscript{BASE},-0.013529311163722992,-0.11311653401145834,-0.007543745620287084,-0.1494091772458263,0.15490064904658374,-0.02573962379894219
+Flan-T5	extsubscript{BASE},0.5034269487785833,0.36490001336083383,0.5049076045340413,0.2302182361504099,0.3121104596406103,0.3831126524928957
+Flan-T5	extsubscript{LARGE},0.7161401564757297,0.4571362263336,0.3097571107969417,0.5286213530232154,0.6074813663159394,0.5238272425890853
+Flan-T5	extsubscript{XL},0.7861767483358856,0.5003943849660698,0.4883173453354119,0.6349030920632762,0.6740659274412318,0.6167714996283751
+Flan-T5	extsubscript{XXL},0.7759252008862759,0.6082696110347289,0.4862164607771357,0.7200172582345266,0.6973966793865924,0.6575650420638519
+Flan-UL2,0.8139011664177609,0.5668563310448644,0.5182231753464608,0.6982912623610599,0.6226590592054422,0.6439861988751175
 T5	extsubscript{SMALL},0.14713232588586322,-0.14745134657341868,0.21085105449290023,0.0024771071745011454,0.053006093385151415,0.05320304687299946
-OPT	extsubscript{30B},0.5971334332930004,0.37923448883915223,0.5059991673294445,0.638926601404518,0.5040249505787286,0.5250637282889687
-OPT	extsubscript{13B},0.6196680588275295,0.37709628221153013,0.4580302159917221,0.6335476037442813,0.48698647702478176,0.5150657275599689
-OPT	extsubscript{6.7B},0.5604634517413575,0.32831892596624523,0.4269879644402534,0.49748807730207684,0.40472353102397157,0.4435963900947809
-OPT	extsubscript{2.7B},0.5717222286656166,0.3371033272017468,0.5281967696551456,0.4209793528179663,0.3906219522536655,0.44972472611882824
-OPT	extsubscript{1.3B},0.4868332699758846,0.33257990090648853,0.48834725116542294,0.338356612359798,0.38402328025886945,0.40602806293329274
-OPT	extsubscript{350M},0.4394508054523349,0.3818975981768043,0.4192647838399,0.2634802012739137,0.2726155768631501,0.35534179312122055
+T5	extsubscript{BASE},-0.013529311163722992,-0.11311653401145834,-0.007543745620287084,-0.1494091772458263,0.15490064904658374,-0.02573962379894219
+T5	extsubscript{LARGE},-0.030600997174729923,0.09892100336814785,-0.1485721634948909,-0.19286116899275968,0.037125937112284256,-0.04719747783638968
+T5	extsubscript{XL},-0.00413988385766918,0.16291281904680144,-0.08980720752317983,0.16149879665432612,-0.19343151292559696,0.007406602278936319
+T5	extsubscript{XXL},0.025803853405636974,-0.06345919381112318,-0.023476076558673384,0.11729746805499064,-0.13017244334921085,-0.014801278451675962
 OPT	extsubscript{125M},0.44059460841507236,0.35041115473250656,0.5006310708424613,0.16778463509135502,0.18441905717732507,0.32876810525174405
-OPT-IML	extsubscript{30B},0.5508094133021331,0.3508511467100317,0.3788470545799674,0.555669072402595,0.4178909829546972,0.4508135339898849
+OPT	extsubscript{350M},0.4394508054523349,0.3818975981768043,0.4192647838399,0.2634802012739137,0.2726155768631501,0.35534179312122055
+OPT	extsubscript{1.3B},0.4868332699758846,0.33257990090648853,0.48834725116542294,0.338356612359798,0.38402328025886945,0.40602806293329274
+OPT	extsubscript{2.7B},0.5717222286656166,0.3371033272017468,0.5281967696551456,0.4209793528179663,0.3906219522536655,0.44972472611882824
+OPT	extsubscript{6.7B},0.5604634517413575,0.32831892596624523,0.4269879644402534,0.49748807730207684,0.40472353102397157,0.4435963900947809
+OPT	extsubscript{13B},0.6196680588275295,0.37709628221153013,0.4580302159917221,0.6335476037442813,0.48698647702478176,0.5150657275599689
+OPT	extsubscript{30B},0.5971334332930004,0.37923448883915223,0.5059991673294445,0.638926601404518,0.5040249505787286,0.5250637282889687
 OPT-IML	extsubscript{1.3B},0.4877807485494955,0.2937602578367721,0.45971989538734637,0.2690262485509278,0.3385350396786613,0.3697644380006406
-OPT-IML	extsubscript{MAX-30B},0.5590721093314605,0.34693753385730824,0.42109651592807673,0.467013454355661,0.48242793804762935,0.4553095103040272
-OPT-IML	extsubscript{MAX-1.3B},0.4887452988091174,0.2824902878861287,0.42677862363017605,0.22738132064323313,0.3703999744304286,0.3591591010798168
+OPT-IML	extsubscript{30B},0.5508094133021331,0.3508511467100317,0.3788470545799674,0.555669072402595,0.4178909829546972,0.4508135339898849
+OPT-IML	extsubscript{M-1.3B},0.4887452988091174,0.2824902878861287,0.42677862363017605,0.22738132064323313,0.3703999744304286,0.3591591010798168
+OPT-IML	extsubscript{M-30B},0.5590721093314605,0.34693753385730824,0.42109651592807673,0.467013454355661,0.48242793804762935,0.4553095103040272
 GPT-3	extsubscript{davinci},0.6817407091635507,0.38019166436920687,0.5366302137182614,0.6383443141688934,0.4213061506802032,0.5316426104200231