UStan Gerard kanye nethimba le-Quiver Bran benze indlela entsha yokuhlola amandla okucubungula nesivinini samamodeli e-AI. Le ndlela ivumela ama-Large Language Models (LLMs) ukuthi alawule abalingisi kugeyimu yevidiyo ye-Street Fighter 3. Aqhudelana ukuze anqume ukuthi iyiphi imodeli ye-AI eshesha kakhulu futhi esabela kakhulu. I-LLM Coliseum, njengoba ibizwe kanjalo, kulapho amamodeli e-AI angenza khona izinqumo zesikhathi sangempela ngesikhathi sokudlala, futhi ukusebenza kwawo kuthonywa izici ezifana nesivinini sokwenza izinqumo.
Amamodeli e-AI afana ne-Mistral Small kanye ne-Mistral Medium athathe indawo ephambili enkundleni ye-LLM Coliseum, ebonisa ukukhalipha kwawo kwamasu nekhono lokwenza izinqumo endaweni esheshayo ye-Street Fighter 3. ayavivinywa futhi ukusebenza kwawo kuhlolwa kusetshenziswa amamethrikhi athuthukile afana nalawa. njenge-matrix yezinga lokuwina kanye nesistimu yokukala ye-ELO. Njengamanje, i-OpenAI's GPT-3.5 Turbo ihamba phambili, ibonisa amakhono ayo aphakeme nokuzivumelanisa nezimo enkundleni yempi ebonakalayo.
Izimpi ze-Street Fighter AI
Impumelelo yalawa mamodeli e-AI incike ekhonweni lawo lokwenza izinqumo ezihlukanisa isibili nokusebenzisa amasu ahlakaniphile ekuphenduleni amandla egeyimu ashintshashintsha njalo. I-Street Fighter 3 idinga ukukhalipha okusheshayo, isikhathi esinembile, kanye nekhono lokulindela kanye nokuphikisa ukunyakaza kwesitha. . Ama-AI aqhudelana ku-“LLM Coliseum” afinyelele inselele, ebonisa ikhono elimangalisayo lokuhlaziya izimo zegeyimu, ukubikezela izenzo zesitha, nokuklama amasu okulwa asebenzayo uma undiza.
I-LLM Coliseum Open-Source Framework
Esinye sezici ezijabulisa kakhulu zephrojekthi ye-LLM Coliseum imvelo yayo yomthombo ovulekile, ekhuthaza abathanda nabacwaningi ukuthi babambe iqhaza ngenkuthalo kuguquko lwe-AI. Le phrojekthi ihlinzeka ngesifundo esibanzi esiqondisa abasebenzisi kunqubo yokusetha isofthiwe edingekayo kanye nokuthola okhiye be-API abadingekayo ukuze benze amamodeli e-AI aphile endaweni ye-Street Fighter 3.
Ngokwenza uhlaka lufinyeleleke kubabukeli abaningi, i-“LLM Coliseum” ihlose ukukhuthaza ukubambisana, ukusungula izinto ezintsha kanye nokushintshisana ngemibono ngaphakathi kwe-AI nemiphakathi yamageyimu. Ababambiqhaza banethuba lokuhlola ubunkimbinkimbi bokuthatha izinqumo, ukuhlolwa kwe-AI. ngamamodeli ahlukene kanye nokucushwa, futhi afake isandla ekuthuthukisweni okuqhubekayo nasekuthuthukisweni kwephrojekthi.
Ukuhlaziywa Kokusebenza kanye Nezinga Lokuncintisana
I-“LLM Coliseum” igcizelela ukuhlaziya ukusebenza kanye namazinga okuncintisana. I-Win Rate Matrix iyithuluzi elibalulekile lokuhlola ukusebenza kahle kokuqhathanisa kwamamodeli ahlukene e-AI, enikeza ukuqonda ngamandla awo, ubuthakathaka, nokusebenza kwawo kukonke kuma-matchups ahlukahlukene. Le ndlela eqhutshwa idatha inikeza ukuqonda okungcono kwezinto ezinomthelela empumelelweni yemidlalo elawulwa yi-AI.
Ukwengeza, uhlelo lwezinga le-ELO lethula ingxenye yokuncintisana kanye nezikhundla phakathi kwamamodeli e-AI. Ngokunikeza izilinganiso ngokusekelwe emandleni abo okufunda, ukuzivumelanisa nezimo, nokuphuma enqobile empini, uhlelo lwe-ELO lusungula izinga eligqamisa izimbangi ze-AI ezesabekayo. Kuyathakazelisa ukuthi amamodeli amancane e-AI avame ukudlula ozakwabo abakhulu kulesi simo, agqamisa indima ebalulekile yejubane nokusebenza kahle kubuhlakani bemidlalo ye-AI.
Ukunwetshwa kanye namathuba esikhathi esizayo
I-LLM Coliseum iyaqhubeka nokuvela futhi yanda, imelela amandla amakhulu ekusasa le-AI kwezemidlalo. Imephu yomgwaqo yephrojekthi ibona kusengaphambili ukuhlanganiswa kwemidlalo eyengeziwe nabalingisi, okuzonweba ububanzi bamageyimu aqhutshwa yi-AI futhi kuvule izindlela ezintsha zokuhlola nokuhlola.
Ngaphezu kwalokho, intuthuko eyenziwe ku-LLM Coliseum inemiphumela efinyelela kude ngale kwemidlalo. Izifundo ezitholwe kumamodeli e-AI enza izinqumo zesikhathi sangempela futhi enze ukuhlela kwamasu zingasetshenziswa emikhakheni eyahlukene, njengamarobhothi, izinhlelo ezizimele, nezimo eziyinkimbinkimbi zokuxazulula izinkinga. Le phrojekthi iwubufakazi bokuthuthuka ngokushesha kobuchwepheshe be-AI namandla ayo okuguqula izimboni ngaphandle kokuzijabulisa.
I-AI Battle Arena iphusha imingcele yalokho okungenzeka nge-AI kwezemidlalo futhi imema abathanda, abacwaningi, nabathuthukisi ukuthi bajoyine i-adventure futhi basize ekulolongeni ikusasa lalo mkhakha ojabulisayo. Ngokusethwa kwayo okusebenziseka kalula kanye nesifundo esibanzi, iphrojekthi ivumela abantu ukuba bangene emhlabeni wemidlalo ye-AI, bathole ulwazi olubalulekile, futhi ekugcineni bashiye uphawu lwabo endaweni ehlala ishintsha ye-AI. ubuhlakani bokwenziwa. Ukuze uthole ulwazi olwengeziwe, vakashela inqolobane esemthethweni ye-GitHub.
Funda kabanzi Umhlahlandlela:
- Izindaba zakamuva ze-AI kuleli sonto: I-OpenAI ChatGPT-5, i-Anthropic, i-Elon nokuningi.
- Ungakucwengisisa kanjani ukwenza izinqumo kwe-AI ku-Semantic Router?
- I-Google Gemini Pro ikhuphukela endaweni yesibili ezilinganisweni zokusebenza kwe-chatbot
- Yakha ikhonsoli yokulingisa ye-Nintendo Raspberry Pi 5
- Engeza izinhlamvu ze-NPC ku-AI kumageyimu ukuze uthuthukise ukucwiliswa, ukuba ngokoqobo, nokuningi
- 4 Izindawo lapho ubuhlakani bokwenziwa (AI) buzothuthukela khona ngo-2024
