HejuruUbuyobozi bukurikira
Tokenisation na Byte Pair Encoding
Tekiniki
Ururimi AI
Ihuza rya byite (Byte-pair encoding) rishyira hamwe ibimenyetso binini binyuze mu guhuza inshuro nyinshi ibice bibiri bikurikirana bikunze kuboneka mu gihe cyo kwiga.
Umva iki gitaboAI-byavuzwe · ~3 min
Ibyo byavuye mu kwiga n'uburemere bwo guhuza biyobora uburyo bwo gushaka ibice by'amagambo (subword encoding), mu gihe imiterere ya byite (byte-level variants) ikoresha ururimi rw'ibice bya byite aho gukoresha isoko ry'inyuguti gusa.
BPE was developed for data compression and adapted to word segmentation in the translation research by Sennrich, Haddow, and Birch, published in 2016. In a common subword-training procedure, text is represented through initial symbols such as characters. Adjacent pairs are counted over the training representation, and a frequent pair is merged into a larger symbol. Repeating the procedure learns reusable pieces. More frequent sequences can gain larger units, while an unfamiliar word may remain split into smaller ones. The precise behavior depends on the training data and implementation; not every common word must become one token. Training produces a vocabulary and an ordered collection of merge rules. GPT-2's published encoder applies eligible pairs according to learned ranks rather than recounting the new input to invent a new vocabulary. It first represents UTF-8 bytes through a reversible byte-to-symbol mapping. The byte alphabet covers 256 values, permitting valid UTF-8 input to be represented through its constituent bytes when larger merged pieces are unavailable. This avoids the character-alphabet coverage limitation under that configuration, but encoding success does not establish that the model knows the text's meaning. Vocabulary accounting can be explicit. In an illustrative configuration with 256 base byte tokens, two distinct additional merge tokens, and one separate special token, the total is 259. Decoding has another important order: GPT-2's implementation joins decoded token pieces, reconstructs the complete byte sequence, then decodes that sequence as UTF-8. Individual pieces need not each correspond to an independently complete Unicode character.
Kwiga uburyo bwo gushaka amagambo (BPE) mu bushakashatsi bw'igereranya bwa Sennrich n'abandi.
Kureba imbonerahamwe ya byite (byte mapping) yashyizwe ahagaruka ya GPT-2 n'uburyo bwo guhuza bishingiye ku rwego.
Kubara umubare w'amagambo (vocabulary size) hashingiwe ku isoko ryavuzwe, ibihuwe bitandukanye, n'umwanya w'ibimenyetso byihariye.
Kugarura imiterere ya byite (byte sequence) mbere yo gusubiza inyuma (decoding) inyandiko yayo ya UTF-8.
Imiterere ya BPE ishobora gugereranywa binyuze mu burebure bw'imiterere, ubwihariwe, n'ingaruka ku modere yigeze, aho kwitwa ko ari yo nziza ku rurimi cyangwa ubwoko bw'inyandiko bwose. Ihindurwa ry'ururimi rw'ibice bya byite, ibikorwa mbere y'igereranya, amagambo, cyangwa amategeko yo guhuza bihindura uburyo bwo gushaka nyabwo. Igereranya risubirwaho rigomba kubika ayo makuru kandi rigaragaze aho hahinduwe (normalization) mbere y'uko gushaka bikorwa. Ibindi buryo bwo gushaka amagambo bigenera amahitamo y'uburyo butandukanye, mu gihe ubwihariwe bwa byite buracyatandukanye n'ubumvure bw'ibisobanuro. Kureba ibice byavuye mu kwiga n'uburyo bwo kugarura ni byo byiza mu gusobanukirwa imikorere y'ibimenyetso aho kwitirirwa BPE yonyine ku masezerano yose ya modere.
Ibaruwa y'amakuru y'ubuntu
Inkuru eshatu zizewe za AI buri gitondo cy'iminsi y'akazi, zanditse mu rurimi rworoshye. Ubuntu iteka ryose, nta kwamamaza.
Imeri imwe buri munsi y'akazi. Kwiyimura mu buryo bumwe. Ntitwigera tugurisha cyangwa ngo dusangire aderesi yawe.
Isuzume
Ibisubizo ako kanya kuri buri kibazo, n'icyemezo gishobora gusangirwa gifite ID yagenzurwa umaze gutsinda isomo.
Fasha uburezi bwa AI bw'ubuntu. AI Understanding ni umuryango utegamiye kuri leta wa 501(c)(3) — nta kwamamaza, nta kwishyura, iteka ryose. Tanga impano
Ihuza rya byite (Byte-pair encoding) rishyira hamwe ibimenyetso binini binyuze mu guhuza inshuro nyinshi ibice bibiri bikurikirana bikunze kuboneka mu gihe cyo kwiga. Ibyo byavuye mu kwiga n'uburemere bwo guhuza biyobora uburyo bwo gushaka ibice by'amagambo (subword encoding), mu gihe imiterere ya byite (byte-level variants) ikoresha ururimi rw'ibice bya byite aho gukoresha isoko ry'inyuguti gusa.
Inzira yo kwiga irabara ibice bibiri bikurikirana kandi iriga guhuza; gushaka (encoding) bikorwa nyuma bikoresheje amategeko yabitswe.
BPE itangirira ku bice bito (inyuguti cyangwa byite mbisi) kandi ikura ibimenyetso binini binyuze mu guhuza.
Ururimi rwuzuye rw'ibice bya byite rutanga igisubizo cy'ibanze; ibi ntibikora ku amagambo yose ashingiye ku nyuguti gusa.
BPE yatangijwe nkubuhanga bwo gukusanya amakuru muri 1994 nyuma iza guhuzwa na subword tokenisation muri 2016.
Imiterere yayo ihitamo ibice byemewe binyuze mu rwego rwa BPE ryabitswe aho kongera kwiga ubwinshi bw'inyandiko mu gihe cyo gushaka.
Komeza wige
Abandi bayobozi batoranijwe kuriyi ngingo
HejuruUbuyobozi bukurikira
Tokenisation na Byte Pair Encoding
Tekiniki