inputs stringlengths 13 578k ⌀ | targets stringlengths 0 6.59k ⌀ | task stringclasses 484
values |
|---|---|---|
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: A small page-boy was waiting outside her own door when she returned to it.
B: The page-boy was small, and had good intentions. | neutral. | glue/mnli |
With no explanation, label A→B with either "entailment" or "not_entailment".
A: What profession was Nathan Alterman?
B: Nathan Alterman, the Israeli poet, was born in Warsaw, as was Moshe Vilenski, the Israeli composer, lyricist, and pianist, who studied music at the Warsaw Conservatory. | entailment. | glue/qnli |
With no explanation, label A→B with either "entailment" or "not_entailment".
A: In 1979, the leaders signed the Egypt-Israel peace treaty on the White House lawn. Both President Begin and Sadat received the Nobel Peace Prize for their work. The two nations have enjoyed peaceful relations to this day.
B: The Israel-Egyp... | entailment. | glue/rte |
With no explanation, label A→B with either "not_entailment" or "entailment".
A: George got free tickets to the play, but he gave them to Eric, because he was not particularly eager to see it.
B: Eric was not particularly eager to see it. | not_entailment. | glue/wnli |
With no explanation, label A→B with either "not_equivalent" or "equivalent".
A: The committee was appointed by Defense Secretary Donald Rumsfeld under orders from Congress .
B: The public hearing is the second for the panel created by Defense Secretary Donald Rumsfeld under pressure from Congress . | not_equivalent. | glue/mrpc |
With no explanation, label A→B with either "not_duplicate" or "duplicate".
A: Can someone share your experiences about real time puppet issues and how did you solve them?
B: Can someone share your experiences about real time docker issues and how did you solve them? | not_duplicate. | glue/qqp |
With no explanation, label the following with either "False" or "True".
do cows have to be pregnant to get milk | False. | super_glue/boolq |
With no explanation, label A→B with either "entailment", "contradiction" or "neutral".
A: Modify the arachnids, said the researchers. Change their bodies and conditions, and you could get fibres like glass, still monofilament, but with logarithmic progressions of possibilities of strength and flexibility, and the abili... | contradiction. | super_glue/cb |
With no explanation, label A→B with either "False" or "True".
A: In many parts of the world, trees lose their leaves in autumn. The leaves turn color. Then fall from the trees to the ground. As the leaves are falling, they have kinetic energy. While they are still attached to the trees they also have energy. When they ... | False. | super_glue/multirc |
With no explanation, label A→B with either "False" or "True".
A: wish : It was his last wish. :
B: wish : They should respect the wishes of the people. : | True. | super_glue/wic |
With no explanation, label A→B with either "entailment" or "not_entailment".
A: The customer asked to speak with the manager because she would be able to fix the billing error.
B: The customer would be able to fix the billing error. | not_entailment. | super_glue/axg |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: Miniatürk is a miniature park situated at the north-eastern shore of Golden Horn in Istanbul, Turkey. It was opened May 2, 2003. Miniatürk covers a total area of 60000 m2 . It is one of the world's largest miniature parks with its... | neutral. | anli/a1 |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: Bouck's Island is a farm near Breakabeen, New York within the town of Fulton, Schoharie County, New York near Fultonham, New York. Bouck's Island was the home of former New York governor William C. Bouck. Congressman Joseph Bouck ... | neutral. | anli/a2 |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: Could we then go back to Justice O'Connor's question? To make that very specific, if we agree with you, does that mean that we would, in principle, have to hold the 1976 extension unconstitutional? I mean, in 1976, Congress extend... | entailment. | anli/a3 |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: John picked up the football there. John discarded the football. John took the football there. Mary went back to the kitchen. Sandra moved to the garden. John went to the hallway. John got the apple there. John got the milk there. Mary went to t... | entailed. | babi_nli/two-supporting-facts |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: Daniel and John moved to the office. After that they went back to the bathroom. Daniel and Mary went back to the garden. After that they journeyed to the bathroom. Mary and John journeyed to the hallway. Then they went to the bedroom. Daniel an... | entailed. | babi_nli/compound-coreference |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: Julius is a lion. Greg is a swan. Bernhard is a frog. Lily is a rhino. Bernhard is gray. Julius is gray. Lily is white. Brian is a rhino. Greg is white.
B: Brian is gray. | not-entailed. | babi_nli/basic-induction |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: Fred got the football there. Fred travelled to the garden. Fred passed the football to Jeff. Mary went to the kitchen. Jeff passed the football to Fred. Fred handed the football to Jeff. Mary grabbed the milk there. Jeff journeyed to the kitche... | entailed. | babi_nli/three-arg-relations |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: The chocolate fits inside the container. The suitcase fits inside the container. The chest fits inside the container. The chest fits inside the box. The container is bigger than the suitcase. The box of chocolates fits inside the chest. The box... | not-entailed. | babi_nli/size-reasoning |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: John travelled to the bedroom. Mary picked up the apple. Daniel went to the kitchen. John went to the bathroom. Mary travelled to the office. Sandra travelled to the bedroom. Mary dropped the apple there. Mary got the apple there. Sandra went t... | entailed. | babi_nli/three-supporting-facts |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: This morning Julie travelled to the kitchen. Julie travelled to the cinema yesterday. Mary went to the kitchen yesterday. Yesterday Bill moved to the kitchen. Mary travelled to the bedroom this morning. This afternoon Julie journeyed to the bed... | entailed. | babi_nli/time-reasoning |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: Mary went back to the bedroom. John travelled to the hallway. Mary journeyed to the office. Daniel journeyed to the bathroom. John moved to the office. John moved to the kitchen. Mary travelled to the hallway. Daniel got the football there. Dan... | entailed. | babi_nli/lists-sets |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: Daniel travelled to the hallway. Sandra travelled to the bathroom. Sandra travelled to the hallway. John travelled to the bathroom. Sandra went back to the bedroom. Daniel travelled to the bedroom. Sandra went back to the bathroom. Mary moved t... | entailed. | babi_nli/counting |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: John moved to the garden. After that he journeyed to the bathroom.
B: John is in the garden. | not-entailed. | babi_nli/basic-coreference |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: Sandra travelled to the office. Sandra moved to the garden.
B: Sandra is in the office. | not-entailed. | babi_nli/single-supporting-fact |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: The hallway is west of the bathroom. The bathroom is west of the kitchen.
B: The kitchen east of is bathroom. | entailed. | babi_nli/two-arg-relations |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: The red square is to the left of the pink rectangle. The red square is to the right of the red sphere.
B: The red sphere is to the left of the pink rectangle. | entailed. | babi_nli/positional-reasoning |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: The bathroom is east of the garden. The kitchen is west of the bedroom. The bathroom is north of the kitchen. The hallway is south of the office. The kitchen is east of the office.
B: You go from the bathroom to the office by heading s,w. | entailed. | babi_nli/path-finding |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: Sandra travelled to the bathroom. Daniel is in the office. Sandra travelled to the kitchen. Daniel went to the hallway. John is in the office. Mary travelled to the garden.
B: Daniel is in the hallway. | entailed. | babi_nli/simple-negation |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: Mary and Daniel moved to the kitchen. Mary and Daniel went to the office.
B: Mary is in the kitchen. | not-entailed. | babi_nli/conjunction |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: Cats are afraid of mice. Gertrude is a cat. Sheep are afraid of mice. Emily is a cat. Wolves are afraid of cats. Jessica is a wolf. Winona is a sheep. Mice are afraid of wolves.
B: Jessica is afraid of cat. | entailed. | babi_nli/basic-deduction |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: John went to the kitchen. Sandra travelled to the hallway. John went to the bedroom. Sandra travelled to the bedroom. Mary moved to the office. Sandra travelled to the hallway. Mary grabbed the football there. Daniel journeyed to the bedroom.
B... | entailed. | babi_nli/yes-no-questions |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: Fred is in the bedroom. Julie journeyed to the kitchen. Fred is either in the school or the park. Mary travelled to the kitchen.
B: Mary is in the kitchen. | entailed. | babi_nli/indefinite-knowledge |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: A man and woman are talking
B: A man and a woman are having a conversation | entailment. | sick/label |
With no explanation, label A→B with either "A_contradicts_B", "A_entails_B" or "A_neutral_B".
A: A man and woman are talking
B: A man and a woman are having a conversation | A_entails_B. | sick/entailment_AB |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: A group of students are walking through the campus.
B: A group of people are walking together | entailment. | snli |
With no explanation, label A→B with either "entailment" or "neutral".
A: The soil was a well-drained sandy loam.
B: This statement best describes sandy soils: sandy soils allow water to drain quickly. | neutral. | scitail/snli_format |
With no explanation, label A→B with either "entailment" or "non-entailment".
A: The tourist that shouted saw the scientists .
B: The tourist saw the scientists . | entailment. | hans |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: But I think you'll find that the city's been gradually moving in that direction for some time.
B: The city has not been moving in that direction for some time. | contradiction. | WANLI |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: He makes some good observations on a few of the pic 's .
B: The making happened | entailed. | recast/recast_factuality |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: Alleyah heard that rare commodities are worth more than good
B: Alleyah did not hear a pun | entailed. | recast/recast_puns |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: When asked about the product, Enrique said, 'I advise EVERYONE DO NOT BE FOOLED!'.
B: Enrique liked the product . | not-entailed. | recast/recast_sentiment |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: Some of the latest secret U.S. cables released by WikiLeaks say Saudi Arabia proposed deploying an Arab military force backed by the U.N. , U.S. and NATO to crush Hezbollah forces in Lebanon .
B: WikiLeaks is a geographical/political entity | not-entailed. | recast/recast_ner |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: Gregory journeyed the gink.
B: Someone or something changed physically . | not-entailed. | recast/recast_verbcorner |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: a particular person did n't come around to do a particular thing .
B: that person did that thing . | not-entailed. | recast/recast_megaveridicality |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: The grocery cart hit against the wall .
B: The hitting happened in a forceful manner . | entailed. | recast/recast_verbnet |
With no explanation, label A→B with either "valid" or "invalid".
A: Overly dramatic couple pose for a picture where an "angry "man "chokes" a woman who sticks out her tongue.
B: It is likely that a man is mad at the woman | valid. | probability_words_nli/usnli |
With no explanation, label A→B with either "valid" or "invalid".
A: There is a very good chance that Mary moved to the garden. It is highly likely that Bernhard is green. It is likely that Sandra left the milk.
B: There is little chance that 'Mary moved to the garden and Sandra left the milk'. | valid. | probability_words_nli/reasoning_1hop |
With no explanation, label A→B with either "valid" or "invalid".
A: It is almost certain that Antoine is thirsty. It is impossible that Lily is a lion. It is probably the case that Jason is tired. There is a very good chance that if either 'Lily is a lion' or 'Jason is tired' but not both then Daniel put down the milk.... | invalid. | probability_words_nli/reasoning_2hop |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: Not often do we see her lose her cool like that.
B: We do not see her lose her cool like him often. | neutral. | nan-nli/joey234--nan-nli |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: Invasion literature remains a part of popular literature to this day.
B: Invasion literature ( or the invasion novel ) is a literary genre most notable between 1871 and the First World War ( 1914 ) but still practised to this day ... | neutral. | nli_fever |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: The auctioneer begins the bidding on the first item of camping gear.
B: The auctioneer begins the bidding on the sixth item of camping gear. | contradiction. | breaking_nli |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: Time travel romance stories may or may not have a happily ever after ending.
B: Time travel romance stories may have a happily ever after ending. | entailment. | conj_nli |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: Each resident of the North American continent can travel freely within Europe. Every Canadian resident is a resident of the North American continent.
B: Each Canadian resident can travel freely within Europe. | entailment. | fracas |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: i am a collector of memorabilia from the 1950 1959 .
B: i am a collector of memorabilia from the 1953 1959 . | contradiction. | dialogue_nli |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: A young girl making a wide mouthed expression while swinging. A little girl shouts for joy as she swings on the swing. A toddler in a pink shirt in a bucket swing. A toddler girl in a swing is laughing.
B: A kid swings. | entailment. | mpe |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: Top-seeded Eyles now meets titleholder Peter Nicol of Scotland who overcame Simon Parke of England 15-7 15-12 15-12 .
B: Peter Nicol is a location | not-entailed. | dnc |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: in 1997 , under a reported multi - million dollar deal , india agreed to construct a `` sophisticated chemical plant at qazvim , on the outskirts of tehran . ''
B: in 1997 , under a reported multi - million dollar deal , india concerted to cons... | not-entailed. | recast_white/fnplus |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: After troubled Heritage Media proposed acquiring POP Radio in a stock swap , POP Radio 's shares tumbled 4 to 14 3\/4 .
B: 4 existed during the tumbling. | not-entailed. | recast_white/sprl |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: The ruling class treats the citizens poorly because they are superior .
B: The ruling class treats the citizens poorly because the citizens are superior . | not-entailed. | recast_white/dpr |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: But he places most blame on how contemporary children are reared .
B: People put the most blame on how contemporary children are reared . | neutral. | robust_nli/IS_CS |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: A photo of four people in a kitchen .
B: A photo of four people in a building . | contradiction. | robust_nli/LI_LI |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: As if this earth in fast thick pants were breathing .
B: It 's like the earth was sighing because we made it sad and true is true . | neutral. | robust_nli/ST_WO |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: No , they did n't , no , just that one .
B: There was only one person the whole time . | neutral. | robust_nli/PI_SP |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: A patient is being worked on by doctors and nurses .
B: A man is at the doctor . | neutral. | robust_nli/PI_CD |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: Model yields an estimate of the percentage change in a household 's demand for postage as a result of owning a computer
B: There is a change in the use of postage depending on whether a househokd owns a computer . | entailment. | robust_nli/ST_SE |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: As if this earth in fast thick pants were breathing .
B: It 's like the earth was sighing because we made it sad and false is not true . | neutral. | robust_nli/ST_NE |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: As if this earth in fast thick pants were breathing and true is true and true is true and true is true and true is true and true is true .
B: It 's like the earth was sighing because we made it sad . | neutral. | robust_nli/ST_LM |
With no explanation, label A→B with either "non-entailment" or "entailment".
A: The manager admired the tourist in front of the author .
B: The author admired the manager . | non-entailment. | robust_nli_is_sd |
With no explanation, label A→B with either "non-contradiction" or "contradiction".
A: She never told you the story then , eh ?
B: How old were you when she shared this story with you ? | contradiction. | robust_nli_li_ts |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: A small child with a shaved head is riding a bouncy sea-horse in the park.
B: A child with long hair. | contradiction. | gen_debiased_nli/snli_seq_z |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: A man lecturing a group of people in a medium sized room.
B: There are people listening to a man speak. | entailment. | gen_debiased_nli/snli_z_aug |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: Figure skater does a trick on a pink BMX bmx bicycle.
B: A rollerblader does a trick. | contradiction. | gen_debiased_nli/snli_par_z |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: So why hasn't Slate outed the leakers on Starr's staff and elsewhere?
B: So, why did Slate out Group X? | contradiction. | gen_debiased_nli/mnli_par_z |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: She was smart enough to object initially to the choice of Ginsburg as her attorney.
B: The rejecting the initial offer, she knew she could get a better attorney. | neutral. | gen_debiased_nli/mnli_z_aug |
With no explanation, label A→B with either "entailment", "neutral" or "contradiction".
A: Numerous studies show that there is no association between music and suicide.
B: There are many studies that prove there is no link between music and suicides. | entailment. | gen_debiased_nli/mnli_seq_z |
With no explanation, label A→B with either "not-entailed" or "entailed".
A: A dog looking at the camera in snow .
B: A dog looking at the camera in red snow . | not-entailed. | add_one_rte |
With no explanation, label A→B with either "_entailment", "_neutral" or "_contradiction".
A: If Ruth does get a job, it's okay.
B: Ruth is unemployed. | _entailment. | imppres/presupposition_change_of_state/presupposition |
With no explanation, label A→B with either "_entailment", "_neutral" or "_contradiction".
A: If it is Andrea that does work with the Clintons, it's okay
B: Someone does work with the Clintons | _entailment. | imppres/presupposition_cleft_existence/presupposition |
With no explanation, label A→B with either "_entailment", "_neutral" or "_contradiction".
A: If it is this waitress who is observing Christina, it's okay
B: Exactly one person is observing Christina. | _entailment. | imppres/presupposition_cleft_uniqueness/presupposition |
With no explanation, label A→B with either "_entailment", "_neutral" or "_contradiction".
A: If both schools that preferred Carolyn to telephone do confer, it's okay.
B: There are exactly two schools that preferred Carolyn to telephone | _entailment. | imppres/presupposition_both_presupposition/presupposition |
With no explanation, label A→B with either "_entailment", "_neutral" or "_contradiction".
A: If the six drivers who go to the river were intending to shout, it's okay.
B: There are exactly six drivers who go to the river. | _entailment. | imppres/presupposition_all_n_presupposition/presupposition |
With no explanation, label A→B with either "_entailment", "_neutral" or "_contradiction".
A: If Suzanne's car does appear, it's okay.
B: Suzanne has a car. | _entailment. | imppres/presupposition_possessed_definites_existence/presupposition |
With no explanation, label A→B with either "_entailment", "_neutral" or "_contradiction".
A: If Katherine's window that embarrasses the committees did shock Ronald, it's okay.
B: Katherine has exactly one window that embarrasses the committees. | _entailment. | imppres/presupposition_possessed_definites_uniqueness/presupposition |
With no explanation, label A→B with either "_entailment", "_neutral" or "_contradiction".
A: If Marcus reveals when Galileo smiles, it's okay.
B: Galileo smiles | _entailment. | imppres/presupposition_question_presupposition/presupposition |
With no explanation, label A→B with either "_entailment", "_neutral" or "_contradiction".
A: If Tracy has only laughed, it's okay.
B: Tracy has laughed. | _entailment. | imppres/presupposition_only_presupposition/presupposition |
With no explanation, label A→B with either "pragmatic_entailment", "pragmatic_neutral" or "pragmatic_contradiction".
A: Neither the patients nor these guests learned who likes to leave.
B: The patients and these guests didn't both learn who likes to leave. | pragmatic_contradiction. | imppres/implicature_connectives/prag |
With no explanation, label A→B with either "pragmatic_entailment", "pragmatic_neutral" or "pragmatic_contradiction".
A: That casserole is not big.
B: That casserole is not enourmous. | pragmatic_contradiction. | imppres/implicature_gradable_adjective/prag |
With no explanation, label A→B with either "pragmatic_entailment", "pragmatic_neutral" or "pragmatic_contradiction".
A: The waiter did not try to talk to Florence.
B: The waiter did not manage to talk to Florence. | pragmatic_contradiction. | imppres/implicature_gradable_verb/prag |
With no explanation, label A→B with either "pragmatic_entailment", "pragmatic_neutral" or "pragmatic_contradiction".
A: Women couldn't hide.
B: Women didn't need to hide. | pragmatic_contradiction. | imppres/implicature_modals/prag |
With no explanation, label A→B with either "pragmatic_entailment", "pragmatic_neutral" or "pragmatic_contradiction".
A: Ten children don't think these cashiers will date.
B: One hundred children don't think these cashiers will date. | pragmatic_neutral. | imppres/implicature_numerals_10_100/prag |
With no explanation, label A→B with either "pragmatic_entailment", "pragmatic_neutral" or "pragmatic_contradiction".
A: Meredith didn't sound like two banks.
B: Meredith didn't sound like three banks. | pragmatic_neutral. | imppres/implicature_numerals_2_3/prag |
With no explanation, label A→B with either "pragmatic_entailment", "pragmatic_neutral" or "pragmatic_contradiction".
A: No drivers would research the governments.
B: Not all drivers would research the governments. | pragmatic_contradiction. | imppres/implicature_quantifiers/prag |
With no explanation, label A→B with either "logical_entailment", "logical_neutral" or "logical_contradiction".
A: The waiter did not try to talk to Florence.
B: The waiter did not manage to talk to Florence. | logical_entailment. | imppres/implicature_gradable_verb/log |
With no explanation, label A→B with either "logical_entailment", "logical_neutral" or "logical_contradiction".
A: Women couldn't hide.
B: Women didn't need to hide. | logical_entailment. | imppres/implicature_modals/log |
With no explanation, label A→B with either "logical_entailment", "logical_neutral" or "logical_contradiction".
A: Neither the patients nor these guests learned who likes to leave.
B: The patients and these guests didn't both learn who likes to leave. | logical_entailment. | imppres/implicature_connectives/log |
With no explanation, label A→B with either "logical_entailment", "logical_neutral" or "logical_contradiction".
A: No drivers would research the governments.
B: Not all drivers would research the governments. | logical_entailment. | imppres/implicature_quantifiers/log |
With no explanation, label A→B with either "logical_entailment", "logical_neutral" or "logical_contradiction".
A: Meredith didn't sound like two banks.
B: Meredith didn't sound like three banks. | logical_entailment. | imppres/implicature_numerals_2_3/log |
With no explanation, label A→B with either "logical_entailment", "logical_neutral" or "logical_contradiction".
A: That casserole is not big.
B: That casserole is not enourmous. | logical_entailment. | imppres/implicature_gradable_adjective/log |
With no explanation, label A→B with either "logical_entailment", "logical_neutral" or "logical_contradiction".
A: Ten children don't think these cashiers will date.
B: One hundred children don't think these cashiers will date. | logical_entailment. | imppres/implicature_numerals_10_100/log |
End of preview. Expand in Data Studio
YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
3 GIẢ THIẾT
- Máy học như người học, dạy người thế nào thì dạy máy như thế.
- Làm thành "giáo trình" từ dễ tới khó như dạy người
- Làm dữ liệu dễ hiểu và hay thì máy dễ học và cho kết quả hay.
- Bỏ thêm noise vào thì dễ, khử noise rất khó.
Câu truyện Lọ Lem, trộn gạo và đỗ rất dễ, nhặt riêng 2 loại rất mệt
Noise là gì? (thường là những thứ không mong muốn hoặc không biết)
- Là những thứ mình không biết nó tồn tại, không định vị được (học thừa)
- Không biết sự ảnh hưởng của nó (có thể gây ảnh hưởng xấu)
- Là những thứ không mong muốn làm ảnh hưởng xấu tới mục tiêu học (mất công khử nhiễu)
- Là sự không cân bằng của dữ liệu làm tốn thời gian học (học thừa)
Để khử noise triệt để ngay từ đầu cần:
- Xác định mục tiêu học rõ ràng (sự ko rõ ràng cũng là noise)
- Chọn lọc nguồn và làm dữ liệu thật cẩn thận
- Cân bằng dữ liệu (dedup, phân cụm / tỉa cụm, lựa chọn số thành viên mỗi cụm ...)
- Lên giáo trình học hợp lý (nền tảng học trước, skills học sau ...)
Với một mô hình đã được huấn luyện
- Có cách khởi tạo tham số của một thành phần mới của model hợp lý để giảm nhiễu, vì khi mở rộng tham số với khởi tạo ngẫu nhiên là ta đang inject white noise vào model (white noise là thuật ngữ trong âm thanh)
- Khi inject cái mới vào model, cần inject từ từ để không gây sốc
- Nếu làm được 1. và 2. thì không cần mô hình lớn để đạt được kỹ năng muốn học
- Học từ dễ tới khó khiến model "dễ" vào hơn (kể cả model nhỏ)
- Dồn tất cả số tham số (nơ-ron) vào việc học pattern muốn học (không học thừa)
Nếu chưa trở thành nghệ nhân hãy là thợ thủ công lành nghề
AI là phần mềm (software), và software là một nghề thủ công (hand-craft), người viết phần mềm giỏi nhất là những nghệ nhân (rất hiếm), người viết phần mềm giỏi là các người thợ lành nghề.
Mội người thợ lành nghề cần sử dụng một bộ các công cụ và nguyên liệu đầu vào để tạo nên sản phẩm (với người thợ đạt đẳng cấp nghệ nhân thì gọi đó là tác phẩm). Toán học hay bất kỳ những công cụ nào (dù là lý thuyết hay thực hành, dù là hữu hình hay trừu tượng) cũng chỉ là một công cụ trong bộ công cụ. Cần sử dụng hợp lý, đúng lúc đúng chỗ và không một công cụ nào quyết định được giá trị của sản phẩm.
Quá trình tạo nên sản phẩm là một quá trình kiên nhẫn, tỉ mỉ, lâu dài, chấp nhận làm đi làm lại một việc cho tới khi master các kỹ năng để hoàn thành việc đó.
=> Chế tạo AI / LLM là một nghề thủ công!
Làm để vượt qua sợ và tham, lấy sự thỏa mãn lớn khi giải được bài toán khó làm phần thưởng
...
Datasets
https://docs.google.com/spreadsheets/d/1tkRZ8T7LZ0ojDUzNFWMKcrPiQJ8Glh-GIFfB7Bksb60
https://github.com/waikato-datamining/llm-dataset-converter (( tool ))
zinz: Dân văn vở đi làm thư ký hoặc đi làm giải trí (nhập vai, đối thoại)
https://docs.google.com/document/d/1nLeNLTYkfw_aADf5CI7_goFTV4zTkWtwiJ-etfqrz2k/edit
- Dữ liệu song ngữ Anh - Việt, Việt - Anh
- Dữ liệu hội thoại Việt / Anh
- Dữ liệu nhập vai Việt / Anh
Dữ liệu song ngữ tốt nhất là các sách / ấn phẩm đã được dịch từ tiếng Anh hoặc tiếng Trung sang tiếng Việt
(( b1 )) Tìm và soạn các tập dữ liệu song ngữ AV, VA để chuyển giao tri thức từ Anh sang Việt
=> Bot đi du học mới về nước, và bắt đầu học tiếng Việt
(( b2 )) Soạn tập dữ liệu tiếng Việt sạch, đúng chính tả để học tiếng Việt, và các kiến thức cơ bản có trong tiếng Việt.
=> Bot học tiếng Việt phổ thông, trình độ tốt nghiệp cấp 3
(( b3 )) Soạn tập dữ liệu sách, epub, văn học, có kiến thức về văn vở tốt.
=> Bot tiếp tục học nâng cao về chuyên ngành văn học, khoa văn, khoa báo chí ...
(( b4 )) Soạn tập dữ liệu truyện để học kỹ năng hội thoại, nhập vai. Có thể lồng ghép nhiều hơn kiến thức giải trí, vui vẻ hài hước ở bước này.
=> Bot tốt nghiệp khoa văn vở và chuyển sang làm ngành giải trí
(( b5 )) Sưu tầm các dữ liệu nhập vai, instruct chatbot tiếng Anh và tạo tập dữ liệu tương đương cho tiếng Việt để finetune thành chatbot. Tập dữ liệu instruct song ngữ cũng giúp kích hoạt lại 1 lần nữa việc chuyển giao tri thức giữa 2 ngôn ngữ.
=> Bot có năng lực nhập vai và trở thành những chuyên gia trong ngành giải trí
Note: Có thể dùng 1 tập instruct song ngữ chất lượng cao nhất để kích hoạt thay cho data nhập vai song ngữ.
(( b4 nâng cao)) Build dataset cẩn thận và sáng tạo như docs/textbooks.md để mở rộng năng lực.
- Lấy mẫu các đoạn hội thoại để phân tích nhìn ra điểm mạnh yếu của các mẩu hội thoại
- Lọc và cân bằng lại các mẩu hội thoại
- Dùng bot xây dựng được ở b3 hoặc chatgpt để viết lại hội thoại
- Dùng bot b3, bloomz 176b, hoặc chatgpt để sinh tiếp các hội thoại chưa có trong dataset
Note: các bot có thể dùng để làm tốt dữ liệu bao gồm:
- bot tự train (b3 for example)
- bot open source (bloomz 176b for example)
- bot close source (chatgpt, google bard, claude ...)
- Downloads last month
- 3,560