Fashewar AI ta bayyana ƙaddamar da sabon sigar dakin karatu kyauta «SpaCy»Wanne yana da aiwatarwa na tsarin sarrafa harshe na asali (NLP). A aikace, ana iya amfani da aikin don gina abubuwan ban mamaki, bots, masu raba rubutu, da kuma tsarin tattaunawa iri daban daban wadanda suke tantance ma'anar jimloli.
Laburare an tsara shi don samar da API mai ɗorewa Ba shi da alaƙa da algorithms da aka yi amfani da shi kuma a shirye yake don amfani a cikin samfuran gaske. Laburare yana amfani da sabbin ci gaba a cikin NLP da kuma mafi ƙarancin algorithms samuwa don aiwatar da bayanai.
Idan algorithm mai inganci ya bayyana, ana wuce da laburaren zuwa gare shi, amma wannan canjin baya shafar API ko aikace-aikace.
Wani fasalin spaCy Har ila yau, gine-ginen da aka tsara don aiwatar da cikakkun takardu, ba tare da yin fifiko ba a cikin magabata waɗanda suka raba daftarin aiki zuwa jimloli. Ana ba da samfuran cikin sifofi biyu: don iyakar aiki da kuma iyakar daidaito.
Babban fasalin spaCy:
- Taimako don kusan harsuna 60
- An riga an koyar da samfuran horo don yare da aikace-aikace daban-daban.
- Koyon Multitask ta hanyar amfani da tiransifoma da aka horar a baya kamar BERT (Bidirectional Encoder Renderings of Transformers).
- Taimako don abubuwanda aka riga aka horar dasu da kalmomin shiga.
- Babban aiki.
- Shirye-shiryen amfani da samfurin tsarin horo kan aiki.
- Alamar motsa harshe
- Abubuwan da aka shirya-shirye suna nan don haɗa mahaɗan da aka ambata, sa alama sassan magana, rarrabe rubutu, nazarin abubuwan dogaro, rarraba jumloli, sanya alamun sassan magana, nazarin halittar jiki, ci gaba, da dai sauransu.
- Tallafi don faɗaɗa ayyuka tare da abubuwan haɗin al'ada da halayen.
- Tallafa don ƙirƙirar samfuranku bisa PyTorch, TensorFlow da sauran tsarin.
- Ginannen kayan aikin da aka sanya don Neman Maɗaukakiyar Maɗaukaki da Nunawa na Ganawa (NER, Ganewar amedungiyar Suna)
- Saukakken tsari na marufi da tura samfurai da gudanar da aiki.
- Babban daidaito.
Laburare an rubuta shi a cikin Python tare da abubuwa a cikin Cython, fadada Python wanda ke ba da izinin aiki kai tsaye a cikin yaren C.
Lambar aikin an rarraba shi a ƙarƙashin lasisin MIT. An shirya samfurin yare don harsuna 58.
Game da sabon sigar spaCy 3.0
Siffar spaCy 3.0 ta fito waje don aiwatar da samfurin iyalai sake horarwa don harsuna 18 kuma An horar da bututun mai 59 gaba ɗaya, gami da sabbin bututun mai 5 na wuta
Ana ba da samfurin a cikin nau'i uku (16 MB, 41 MB - vectors dubu 20 da 491 MB - 500 dubu vectors) kuma an inganta aiki a ƙarƙashin nauyin CPU kuma ya hada da tok2vec, morphologizer, parser, senter, ner, attribute_ruler, da lemmatizer.
Mun kasance muna aiki akan spaCy v3.0 fiye da shekara guda, kuma kusan shekaru biyu idan ka ƙidaya duk aikin da aka yi akan Thinc. Babban burinmu tare da ƙaddamarwar shine a sauƙaƙe kawo irinku samfura a cikin SPACY, musamman ƙirar zamani irin ta masu kawo canji. Kuna iya rubuta samfuran da ke ciyar da abubuwan spaCy zuwa cikin tsari kamar PyTorch ko TensorFlow, ta amfani da sabon tsarin mu na ban mamaki don bayyana duk saitunan ku. Kuma tunda ayyukan NLP na zamani galibi suna ɗauke da matakai da yawa, akwai sabon tsarin aikin aiki don taimaka muku kiyaye aikinku cikin tsari.
Sauran muhimman sabbin abubuwa wanda ya fice daga sabon sigar:
- Sabuwar hanyar aiki don samfuran horo.
- Sabon tsarin daidaitawa.
- Taimako don samfurin mai bututun mai canzawa, wanda ya dace da ilmantarwa da yawa.
- Ikon haɗawa da samfuran ku ta amfani da tsarin ilmantarwa na injina daban-daban, kamar su PyTorch, TensorFlow, da MXNet.
- Tallafin aikin don gudanar da duk matakai na gudanawar aiki, daga aikin sarrafawa zuwa aiwatar da ƙira.
- Taimako don haɗuwa tare da Controlaukar Bayanin Bayanai (DVC), Streamlit, Weights & Biases da Ray ɗin kunshe.
- Sabbin abubuwan da aka gina: SentenceRecognizer, Morphologizer, Lemmatizer,
- AttributeRuler da gidan wuta.
- Sabuwar API don ƙirƙirar abubuwan haɗinku.
A ƙarshe, idan kuna sha'awar ƙarin sani game da shi na wannan sabon sigar ko game da spaCy, zaku iya bincika bayanan A cikin mahaɗin mai zuwa.