# Few-shot examples for generating and simulating neuron explanations. from __future__ import annotations from dataclasses import dataclass from enum import Enum from neuron_explainer.activations.activations import ActivationRecord from neuron_explainer.fast_dataclasses import FastDataclass @dataclass class Example(FastDataclass): activation_records: list[ActivationRecord] explanation: str first_revealed_activation_indices: list[int] """ For each activation record, the index of the first token for which the activation value in the prompt should be an actual number rather than "unknown". Examples all start with the activations rendered as "unknown", then transition to revealing specific normalized activation values. The goal is to lead the model to predict that activation sequences will eventually transition to predicting specific activation values instead of just "unknown". This lets us cheat and get predictions of activation values for every token in a single round of inference by having the activations in the sequence we're predicting always be "unknown" in the prompt: the model will always think that maybe the next token will be a real activation. """ token_index_to_score: int | None = None """ If the prompt is used as an example for one-token-at-a-time scoring, this is the index of the token to score. """ class FewShotExampleSet(Enum): """Determines which few-shot examples to use when sampling explanations.""" ORIGINAL = "original" COLANGV2 = "colangv2" TEST = "test" @classmethod def from_string(cls, string: str) -> FewShotExampleSet: for example_set in FewShotExampleSet: if example_set.value == string: return example_set raise ValueError(f"Unrecognized example set: {string}") def get_examples(self) -> list[Example]: """Returns regular examples for use in a few-shot prompt.""" if self is FewShotExampleSet.ORIGINAL: return ORIGINAL_EXAMPLES elif self is FewShotExampleSet.COLANGV2: return COLANGV2_EXAMPLES elif self is FewShotExampleSet.TEST: return TEST_EXAMPLES else: raise ValueError(f"Unhandled example set: {self}") def get_single_token_prediction_example(self) -> Example: """ Returns an example suitable for use in a subprompt for predicting a single token's normalized activation, for use with the "one token at a time" scoring approach. """ if self is FewShotExampleSet.COLANGV2: return COLANGV2_SINGLE_TOKEN_EXAMPLE elif self is FewShotExampleSet.TEST: return TEST_SINGLE_TOKEN_EXAMPLE else: raise ValueError(f"Unhandled example set: {self}") TEST_EXAMPLES = [ Example( activation_records=[ ActivationRecord( tokens=["a", "b", "c"], activations=[1.0, 0.0, 0.0], ), ActivationRecord( tokens=["d", "e", "f"], activations=[0.0, 1.0, 0.0], ), ], explanation="vowels", first_revealed_activation_indices=[0, 1], ), ] TEST_SINGLE_TOKEN_EXAMPLE = Example( activation_records=[ ActivationRecord( activations=[0.0, 0.0, 1.0], tokens=["g", "h", "i"], ), ], first_revealed_activation_indices=[], token_index_to_score=2, explanation="test explanation", ) ORIGINAL_EXAMPLES = [ Example( activation_records=[ ActivationRecord( tokens=[ "t", "urt", "ur", "ro", " is", " fab", "ulously", " funny", " and", " over", " the", " top", " as", " a", " '", "very", " sneaky", "'", " but", "ler", " who", " excel", "s", " in", " the", " art", " of", " impossible", " disappearing", "/", "re", "app", "earing", " acts", ], activations=[ -0.71, -1.85, -2.39, -2.58, -1.34, -1.92, -1.69, -0.84, -1.25, -1.75, -1.42, -1.47, -1.51, -0.8, -1.89, -1.56, -1.63, 0.44, -1.87, -2.55, -2.09, -1.76, -1.33, -0.88, -1.63, -2.39, -2.63, -0.99, 2.83, -1.11, -1.19, -1.33, 4.24, -1.51, ], ), ActivationRecord( tokens=[ "esc", "aping", " the", " studio", " ,", " pic", "col", "i", " is", " warm", "ly", " affecting", " and", " so", " is", " this", " ad", "roit", "ly", " minimalist", " movie", " .", ], activations=[ -0.69, 4.12, 1.83, -2.28, -0.28, -0.79, -2.2, -2.03, -1.77, -1.71, -2.44, 1.6, -1, -0.38, -1.93, -2.09, -1.63, -1.94, -1.82, -1.64, -1.32, -1.92, ], ), ], first_revealed_activation_indices=[10, 3], explanation="present tense verbs ending in 'ing'", ), Example( activation_records=[ ActivationRecord( tokens=[ "as", " sac", "char", "ine", " movies", " go", " ,", " this", " is", " likely", " to", " cause", " massive", " cardiac", " arrest", " if", " taken", " in", " large", " doses", " .", ], activations=[ -0.14, -1.37, -0.68, -2.27, -1.46, -1.11, -0.9, -2.48, -2.07, -3.49, -2.16, -1.79, -0.23, -0.04, 4.46, -1.02, -2.26, -2.95, -1.49, -1.46, -0.6, ], ), ActivationRecord( tokens=[ "shot", " perhaps", " '", "art", "istically", "'", " with", " handheld", " cameras", " and", " apparently", " no", " movie", " lights", " by", " jo", "aquin", " b", "aca", "-", "as", "ay", " ,", " the", " low", "-", "budget", " production", " swings", " annoy", "ingly", " between", " vert", "igo", " and", " opacity", " .", ], activations=[ -0.09, -3.53, -0.72, -2.36, -1.05, -1.12, -2.49, -2.14, -1.98, -1.59, -2.62, -2, -2.73, -2.87, -3.23, -1.11, -2.23, -0.97, -2.28, -2.37, -1.5, -2.81, -1.73, -3.14, -2.61, -1.7, -3.08, -4, -0.71, -2.48, -1.39, -1.96, -1.09, 4.37, -0.74, -0.5, -0.62, ], ), ], first_revealed_activation_indices=[5, 20], explanation="words related to physical medical conditions", ), Example( activation_records=[ # The sense of togetherness in our town is strong. ActivationRecord( tokens=[ "the", " sense", " of", " together", "ness", " in", " our", " town", " is", " strong", " .", ], activations=[ 0, 0, 0, 1, 2, 0, 0.23, 0.5, 0, 0, 0, ], ), ActivationRecord( tokens=[ "a", " buoy", "ant", " romantic", " comedy", " about", " friendship", " ,", " love", " ,", " and", " the", " truth", " that", " we", "'re", " all", " in", " this", " together", " .", ], activations=[ -0.15, -2.33, -1.4, -2.17, -2.53, -0.85, 0.23, -1.89, 0.09, -0.47, -0.5, -0.58, -0.87, 0.22, 0.58, 1.34, 0.98, 2.21, 2.84, 1.7, -0.89, ], ), ], first_revealed_activation_indices=[0, 10], explanation="phrases related to community", ), ] COLANGV2_EXAMPLES = [ Example( activation_records=[ ActivationRecord( tokens=[ "The", " editors", " of", " Bi", "opol", "ym", "ers", " are", " delighted", " to", " present", " the", " ", "201", "8", " Murray", " Goodman", " Memorial", " Prize", " to", " Professor", " David", " N", ".", " Ber", "atan", " in", " recognition", " of", " his", " seminal", " contributions", " to", " bi", "oph", "ysics", " and", " their", " impact", " on", " our", " understanding", " of", " charge", " transport", " in", " biom", "olecules", ".\n\n", "In", "aug", "ur", "ated", " in", " ", "200", "7", " in", " honor", " of", " the", " Bi", "opol", "ym", "ers", " Found", "ing", " Editor", ",", " the", " prize", " is", " awarded", " for", " outstanding", " accomplishments", ], activations=[ 0, 0.01, 0.01, 0, 0, 0, -0.01, 0, -0.01, 0, 0, 0, 0, 0, 0.04, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3.39, 0.12, 0, -0.01, 0, 0, 0, 0, -0, 0, -0, 0, 0, -0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, -0, 0, 0, -0.01, 0, 0.41, 0, 0, 0, -0.01, 0, 0, 0, 0, 0, ], ), # We sometimes exceed the max context size when this is included :( # We can uncomment this if we start using an 8k context size. # ActivationRecord( # tokens=[ # " We", # " are", # " proud", # " of", # " our", # " national", # " achievements", # " in", # " mastering", # " all", # " aspects", # " of", # " the", # " fuel", # " cycle", # ".", # " The", # " current", # " international", # " interest", # " in", # " closing", # " the", # " fuel", # " cycle", # " is", # " a", # " vind", # "ication", # " of", # " Dr", # ".", # " B", # "hab", # "ha", # "’s", # " pioneering", # " vision", # " and", # " genius", # ], # activations=[ # -0, # -0, # 0, # -0, # -0, # 0, # 0, # 0, # -0, # 0, # 0, # -0, # 0, # -0.01, # 0, # 0, # -0, # -0, # 0, # 0, # 0, # -0, # -0, # -0.01, # 0, # 0, # -0, # 0, # 0, # 0, # 0, # 0, # -0, # 0, # 0, # 0, # 2.15, # 0, # 0, # 0.03, # ], # ), ], first_revealed_activation_indices=[7], # , 19], explanation="language related to something being groundbreaking", ), Example( activation_records=[ ActivationRecord( tokens=[ '{"', "widget", "Class", '":"', "Variant", "Matrix", "Widget", '","', "back", "order", "Message", '":"', "Back", "ordered", '","', "back", "order", "Message", "Single", "Variant", '":"', "This", " item", " is", " back", "ordered", '.","', "ordered", "Selection", '":', "true", ',"', "product", "Variant", "Id", '":', "0", ',"', "variant", "Id", "Field", '":"', "product", "196", "39", "_V", "ariant", "Id", '","', "back", "order", "To", "Message", "Single", "Variant", '":"', "This", " item", " is", " back", "ordered", " and", " is", " expected", " by", " {", "0", "}.", '","', "low", "Price", '":', "999", "9", ".", "0", ',"', "attribute", "Indexes", '":[', '],"', "productId", '":', "196", "39", ',"', "price", "V", "ariance", '":', "true", ',"', ], activations=[ -0.03, 0, 0, 0, 4.2, 0, 0, 0, 0, -0, 0, -0, 0, 0, 0, 0, -0.01, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, -0, 0, 0, 0, 0, 0, 0, -0.03, 0, 0, 0, 0, -0.02, 0, 0, 0, 0, 0, 0, 0, 0, 0, -0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, -0, -0, 0, 0, 0, 0.01, -0.01, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, -0.02, 0, 0, 0, 0, 0, 1.24, 0, 0, 0, ], ), ActivationRecord( tokens=[ "A", " regular", " look", " at", " the", " ups", " and", " downs", " of", " variant", " covers", " in", " the", " comics", " industry", "…\n\n", "Here", " are", " the", " Lego", " variant", " sketch", " covers", " by", " Leon", "el", " Cast", "ell", "ani", " for", " a", " variety", " of", " Marvel", " titles", ",", ], activations=[ 0, -0.01, 0, 0, 0, 0, 0, 0, 0, 6.52, 0, 0, 0, -0, 0, 0, 0, 0, 0, 0, 1.62, 0, 0, 0, 0, 0, 0, 0, 0, 0, -0, 0, 0, 0, -0, 0, ], ), ], first_revealed_activation_indices=[2, 8], explanation="the word “variant” and other words with the same ”vari” root", ), ] COLANGV2_SINGLE_TOKEN_EXAMPLE = Example( activation_records=[ ActivationRecord( tokens=[ "B", "10", " ", "111", " MON", "DAY", ",", " F", "EB", "RU", "ARY", " ", "11", ",", " ", "201", "9", " DON", "ATE", "fake higher scoring token", # See below. ], activations=[ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.37, # This fake activation makes the previous token's activation normalize to 8, which # might help address overconfidence in "10" activations for the one-token-at-a-time # scoring prompt. This value and the associated token don't actually appear anywhere # in the prompt. 0.45, ], ), ], first_revealed_activation_indices=[], token_index_to_score=18, explanation="instances of the token 'ate' as part of another word", ) @dataclass class AttentionSimulationExample(FastDataclass): token_pair_example_index: int token_pair_coordinates: tuple[int, int] label: int @dataclass class AttentionTokenPairExample(FastDataclass): tokens: list[str] token_pair_coordinates: list[tuple[int, int]] @dataclass class AttentionHeadFewShotExample(FastDataclass): token_pair_examples: list[AttentionTokenPairExample] explanation: str simulation_examples: list[AttentionSimulationExample] | None = None ATTENTION_HEAD_FEW_SHOT_EXAMPLES: list[AttentionHeadFewShotExample] = [ # gpt2-xl, layer 1, head 1 AttentionHeadFewShotExample( token_pair_examples=[ AttentionTokenPairExample( tokens=[ " dreams", " of", " a", " future", " like", " her", " biggest", " idol", ",", " who", " was", " also", " born", " visually", " impaired", ".", "\n", "\n", '"', "My", " ultimate", " dream", " would", " be", " to", " sing", " at", " Carol", "s", " [", "by", " Candle", "light", "]", " and", " to", " become", " a", " famous", " musician", " like", " Andrea", " Bo", "cell", "i", " ...", " and", " to", " show", " people", " that", " if", " you", " have", " a", " disability", " it", " doesn", "'t", " matter", ',"', " she", " said", ".", ], # 45 = "attended from", 33 = "attended to" token_pair_coordinates=[(45, 33)], ), AttentionTokenPairExample( tokens=[ "omes", " Ever", " Sequ", "enced", "]", "\n", "\n", "One", " mystery", " of", " cat", " development", " is", " how", " cats", " have", " come", " to", " have", " such", " varied", " coats", ",", " from", " solid", " colours", " to", ' "', "mac", "ke", "rel", '"', " tab", "by", " patterns", " of", " thin", " vertical", " stripes", ".", " The", " researchers", " were", " particularly", " interested", " in", " what", " turns", " the", " mac", "ke", "rel", " pattern", " into", " a", ' "', "bl", "ot", "ched", '"', " tab", "by", " pattern", ",", ], token_pair_coordinates=[(5, 4)], ), AttentionTokenPairExample( tokens=[ ",", " 6", ",", " 8", ",", " 4", "]", "':", "rb", ".", "sort", ".", "slice", "(", "1", ",", "2", ");", " #", " More", " advanced", ",", " this", " is", " Ruby", "'s", " map", " and", " each", "_", "with", "_", "index", " #", " This", " shows", " the", " :", "rb", " post", "fix", "-", "operator", " sugar", " instead", " of", " EV", "AL", ' "[', "1", ",", "2", ",", "3", ",", "4", "]", '":', "rb", " .", "map", "(", "->", " $", ], token_pair_coordinates=[(7, 6)], ), AttentionTokenPairExample( tokens=[ " him", " a", " W", "N", " [", "white", " nationalist", "]", " until", " there", " is", " an", " indication", " as", " such", "...", " The", " fact", " that", " he", " targeted", " a", " church", " gives", " me", " an", " ink", "ling", " that", " it", " was", " religion", "-", "related", ',"', " wrote", " White", "Virgin", "ian", ".", "\n", "\n", '"', "Yep", ",", " bad", " news", " for", " gun", " rights", " advocates", " as", " well", ',"', " wrote", " math", "the", "ory", "l", "over", "2008", ".", ' "', "Another", ], token_pair_coordinates=[(15, 7)], ), AttentionTokenPairExample( tokens=[ "23", "]", "\n", "\n", "While", " preparing", " to", " take", " the", " fight", " to", " Prim", "ord", "us", ",", " B", "alth", "azar", " learned", " about", " T", "aim", "i", "'s", " machine", " and", " how", " it", " could", " supposedly", " kill", " two", " Elder", " Dragons", " with", " a", " single", " blow", ",", " which", " p", "iqu", "ed", " his", " interest", ".", " This", " piece", " of", " news", ",", " as", " well", " as", " Mar", "j", "ory", "'s", " sudden", " departure", " from", " his", " side", " which", ], token_pair_coordinates=[(3, 1)], ), ], explanation="attends to the latest closing square bracket from arbitrary subsequent tokens", ), # gpt2-xl, layer 2, head 8 AttentionHeadFewShotExample( simulation_examples=[ AttentionSimulationExample( token_pair_example_index=0, token_pair_coordinates=(63, 15), label=0, ), AttentionSimulationExample( token_pair_example_index=0, token_pair_coordinates=(50, 15), label=1, ), ], token_pair_examples=[ AttentionTokenPairExample( tokens=[ " he", " said", ".", ' "', "Coming", " off", " winning", " the", " year", " before", ",", " I", " love", " playing", " links", " golf", ",", " and", " I", " love", " playing", " the", " week", " before", " a", " major", ".", " It", " was", " tough", " to", " miss", " it", ".", " I", "'m", " just", " glad", " to", " be", " back", '."', "\n", "\n", "F", "owler", " out", "played", " his", " partners", " Rory", " Mc", "Il", "roy", " (", "74", ")", " and", " Hen", "rik", " St", "enson", " (", "72", ], token_pair_coordinates=[(50, 15)], ), AttentionTokenPairExample( tokens=[ " Club", ":", "\n", "\n", "1", ".", " World", " renowned", " golf", " course", "\n", "\n", "2", ".", " Vern", " Mor", "com", " designed", " golf", " course", "\n", "\n", "3", ".", " Great", " family", " holiday", " destination", "\n", "\n", "4", ".", " Play", " amid", " our", " resident", " Eastern", " Grey", " k", "ang", "aroo", " population", "\n", "\n", "5", ".", " Terr", "ific", " friendly", " staff", "\n", "\n", "6", ".", " Natural", " pictures", "que", " bush", " setting", "\n", "\n", "7", ".", " L", ], token_pair_coordinates=[(9, 8)], ), AttentionTokenPairExample( tokens=[ "615", " rpm", " on", " average", ").", " As", " a", " result", ",", " each", " player", " was", " hitting", " longer", " drives", " on", " their", " best", " shots", ",", " while", " achieving", " a", " stra", "ighter", " ball", " flight", " that", " was", " less", " affected", " by", " wind", ".", "\n", "\n", "Every", " Golf", "WR", "X", " Member", " gained", " yard", "age", " with", " a", " new", " Taylor", "Made", " driver", ";", " the", " largest", " distance", " gain", " was", " an", " impressive", " +", "10", ".", "1", " yards", ",", ], token_pair_coordinates=[(47, 37)], ), AttentionTokenPairExample( tokens=[ " of", " being", "?", " Well", ",", " having", " perfected", " the", " art", " of", " swimming", ",", " Phelps", " has", " moved", " on", " to", " another", " cherished", " summer", " past", "ime", " –", " golf", ".", " Here", " he", " is", " participating", " in", " the", " Dun", "hill", " Links", " Championship", " at", " Kings", "b", "arn", "s", " in", " Scotland", " today", ".", " The", " greens", " over", " there", " are", " really", " big", ",", " so", " the", " opportunity", " for", " 50", "-", "yard", " put", "ts", " exist", ".", " Of", ], token_pair_coordinates=[(45, 23)], ), AttentionTokenPairExample( tokens=[ "OTUS", " is", " getting", " to", " see", " aliens", ".", "\n", "\n", "RELATED", ":", " Barack", " Obama", " joins", " second", " D", ".", "C", ".-", "area", " golf", " club", "\n", "\n", '"', "He", " goes", ",", " '", "they", "'re", " freaking", " crazy", " looking", ".'", " And", " then", " he", " walks", " up", ",", " makes", " his", " put", "t", ",", " turns", " back", ",", " walks", " off", " the", " green", ",", " leaves", " it", " at", " that", " and", " gives", " me", " a", " wink", ',"', ], token_pair_coordinates=[(52, 20)], ), ], explanation='attends to the token "golf" from golf-related tokens', ), # gpt2-xl, layer 1, head 10 AttentionHeadFewShotExample( simulation_examples=[ AttentionSimulationExample( token_pair_example_index=0, token_pair_coordinates=(37, 36), label=0, ), AttentionSimulationExample( token_pair_example_index=0, token_pair_coordinates=(14, 12), label=1, ), ], token_pair_examples=[ AttentionTokenPairExample( tokens=[ " security", " by", " requiring", " the", " user", " to", " enter", " a", " numeric", " code", " sent", " to", " his", " or", " her", " cellphone", " in", " addition", " to", " a", " password", ".", " A", " lot", " of", " websites", " have", " offered", " this", " feature", " for", " years", ",", " but", " Int", "uit", " just", " made", " it", " widely", " available", " earlier", " this", " year", ".", "\n", "\n", '"', "When", " you", " give", " your", " most", " sensitive", " data", " and", " that", " of", " your", " family", " to", " a", " company", ",", ], token_pair_coordinates=[(14, 12)], ), AttentionTokenPairExample( tokens=[ " 3", " months", ",", " they", " separated", " the", " men", " and", " women", " here", ".", " I", " don", "'t", " know", " where", " they", " took", " the", " men", " and", " the", " children", ",", " but", " they", " took", " us", " women", " to", " Syria", ".", " They", " kept", " us", " in", " an", " underground", " prison", ".", " My", " only", " wish", " is", " that", " my", " children", " and", " husband", " escape", " ISIS", ".", " They", " brought", " us", " here", " from", " Raqqa", ",", " my", " sisters", " from", " the", " PKK", ], token_pair_coordinates=[(8, 6)], ), AttentionTokenPairExample( tokens=[ " an", " emphasis", " on", " the", " pursuit", " of", " power", " despite", " interpersonal", " costs", '."', "\n", "\n", "The", " study", ",", " which", " involved", " over", " 600", " young", " men", " and", " women", ",", " makes", " a", " strong", " case", " for", " assessing", " such", " traits", " as", ' "', "r", "uth", "less", " ambition", ',"', ' "', "dis", "comfort", " with", " leadership", '"', " and", ' "', "hub", "rist", "ic", " pride", '"', " to", " understand", " psychopath", "ologies", ".", "\n", "\n", "The", " researchers", " looked", " at", ], token_pair_coordinates=[(23, 21)], ), AttentionTokenPairExample( tokens=[ " 4", " hours", ".", " These", " results", ",", " differ", " between", " men", " and", " women", ",", " however", ".", " We", " can", " see", " that", " although", " both", " groups", " have", " a", " large", " cluster", " of", " people", " at", " exactly", " 40", " hours", " per", " week", ",", " there", " are", " more", " men", " reporting", " hours", " above", " 40", ",", " whereas", " there", " are", " more", " women", " reporting", " hours", " below", " 40", ".", " Result", " 3", ":", " Male", " Hours", " Work", "ed", " [", "Info", "]", " Owner", ], token_pair_coordinates=[(10, 8)], ), AttentionTokenPairExample( tokens=[ " they", " were", " perceived", " as", " more", " emotional", ",", " which", " made", " participants", " more", " confident", " in", " their", " own", " opinion", '."', "\n", "\n", "Ms", " Sal", "erno", " said", " both", " men", " and", " women", " reacted", " in", " the", " same", " way", " to", " women", " expressing", " themselves", " angrily", ".", "\n", "\n", '"', "Particip", "ants", " confidence", " in", " their", " own", " verdict", " dropped", " significantly", " after", " male", " hold", "outs", " expressed", " anger", ',"', " the", " paper", "'s", " findings", " stated", ".", "\n", ], token_pair_coordinates=[(26, 24)], ), ], explanation="attends to male-related tokens from paired female-related tokens", ), # gpt2-xl, layer 1, head 3 AttentionHeadFewShotExample( token_pair_examples=[ AttentionTokenPairExample( tokens=[ "\n", "********************************", "************", "***", "\n", "\n", "V", "iet", "namese", " Ministry", " of", " Foreign", " Affairs", " spokesperson", " Le", " Hai", " Bin", "h", " is", " seen", " in", " this", " file", " photo", ".", " .", " Tu", "oi", " Tre", "\n", "\n", "The", " Ministry", " of", " Foreign", " Affairs", " has", " ordered", " a", " thorough", " investigation", " into", " a", " case", " in", " which", " a", " Vietnamese", " fisherman", " was", " shot", " dead", " on", " his", " boat", " in", " Vietnam", "'s", " Tru", "ong", " Sa", " (", "Spr", "at", ], token_pair_coordinates=[(1, 1)], ), AttentionTokenPairExample( tokens=[ " J", "okin", "en", " tells", " a", " much", " different", " story", ".", " He", " almost", " sounded", " like", " a", " pitch", "man", ".", "\n", "\n", '"', "All", " the", " staff", ",", " team", " service", " guys", ",", " all", " the", " trainers", ",", " they", "'re", " unbelievable", " guys", ',"', " said", " J", "okin", "en", ".", ' "', "It", "'s", " not", " just", " the", " players", ",", " it", "'s", " the", " staff", " around", " the", " team", ".", " I", " feel", " really", " bad", " for", " them", ], token_pair_coordinates=[(1, 1)], ), AttentionTokenPairExample( tokens=[ " a", " Pv", "E", " game", ",", " we", " probably", " would", " use", " it", " but", " based", " on", " the", " tests", " we", "'ve", " run", " on", " it", ",", " that", " wouldn", "'t", " be", " our", " first", " choice", " for", " a", " live", " R", "v", "R", " game", ".", " Now", ",", " could", " we", " use", " it", " for", " prototyp", "ing", "?", " Yep", ",", " we", " are", " already", " doing", " that", ".", " Second", ",", " as", " to", " other", " engines", " there", " are", " both", " financial", ], token_pair_coordinates=[(1, 1)], ), AttentionTokenPairExample( tokens=[ "-", "tun", "er", " is", " also", " custom", "isable", " for", " hassle", "-", "free", " experimentation", ".", " The", " St", "rix", " X", "399", "-", "E", " Gaming", " takes", " up", " to", " three", " double", "-", "wide", " cards", " in", " SLI", " or", " Cross", "Fire", "X", ".", " Primary", " graphics", " slots", " are", " protected", " by", " Safe", "Slot", " from", " damages", " that", " heavy", " GPU", " cool", "ers", " can", " potentially", " cause", ".", "\n", "\n", "Personal", "ised", " RGB", " lighting", " is", " made", " possible", ], token_pair_coordinates=[(1, 1)], ), AttentionTokenPairExample( tokens=[ " to", " abolish", " such", " a", " complex", "?", " Are", " there", " ways", " ve", "gans", " can", " eat", " more", " sustain", "ably", "?", " What", " are", " some", " of", " the", " health", " challenges", " for", " new", " ve", "gans", ",", " and", " how", " can", " we", " raise", " awareness", " of", " these", " issues", " so", " that", ",", " for", " instance", ",", " medical", " professionals", " are", " more", " supportive", " of", " vegan", "ism", "?", "\n", "\n", "Moreover", ",", " it", " is", " essential", " that", " ve", "gans", " differentiate", ], token_pair_coordinates=[(1, 1)], ), ], explanation="attends from the second token in the sequence to the second token in the sequence", ), ]