Mafi Kyawun Masu Ba da LLM Open-Source 2026 — BYOI & ShareAI’s Hybrid Route

TL;DR — Akwai hanyoyi guda uku masu amfani don gudanar da LLMs na buɗe tushen a yau:

(1) Gudanarwa (ba tare da uwar garke ba; biyan kuɗi bisa ga miliyan kalmomi; babu kayan aiki da za a kula),

(2) Masaukin LLM na Buɗe Tushen (kai tsaye gudanar da samfurin da kake so), kuma

(3) BYOI haɗe da hanyar sadarwa mai zaman kansa (gudanarwa akan kayan aikin ka na farko, sannan canza kai tsaye zuwa ƙarfin hanyar sadarwa kamar RabaAI). Wannan jagorar yana kwatanta zaɓuɓɓuka masu jagoranci (Hugging Face, Together, Replicate, Groq, AWS Bedrock, io.net), yana bayyana yadda BYOI ke aiki a ShareAI (tare da sauyawa fifiko akan Na'urata bisa maɓalli), kuma yana bayar da alamu, lambar, da tunani kan farashi don taimaka maka jigilar aiki da kwarin gwiwa.

Don cikakken bayyani na kasuwa, duba labarin yanayin Eden AI: Mafi kyawun Masu Ba da Masaukin LLM na Buɗe Tushen.

Teburin abun ciki

• Tashin Masaukin LLM na Buɗe Tushen
• Menene “masaukin LLM na buɗe tushen” ke nufi
• Me yasa za a gudanar da LLMs na buɗe tushen?
• Hanyoyi uku don gudanar da LLMs
  • 4.1 Gudanarwa (ba tare da uwar garke ba; biyan kuɗi bisa miliyan alamu)
  • 4.2 Bude-Tushen LLM Hosting (kai tsaye)
  • 4.3 BYOI + hanyar sadarwa mai rarrabuwa (ShareAI haɗin gwiwa)
• ShareAI a cikin dakika 30
• Yadda BYOI tare da ShareAI ke aiki (muhimmanci ga na'urarka + madaidaicin madadin)
• Matiriksin kwatance mai sauri (masu samarwa a taƙaice)
• Bayanan masu samarwa (karatun gajere)
• Inda ShareAI ya dace da wasu (shirin yanke shawara)
• Ayyuka, jinkiri & amintuwa (tsarin ƙira)
• Gudanarwa, bin doka & wurin bayanai
• Tsarin farashi: gudanarwa vs kai tsaye vs BYOI + rarrabuwa
• Mataki-mataki: farawa
• Yankunan lambar
• Misalai na ainihi
• Tambayoyi akai-akai (SEO mai dogon lokaci)
• Tunani na ƙarshe

Tashin Masaukin LLM na Buɗe Tushen

Samfura masu nauyi na buɗe kamar Llama 3, Mistral/Mixtral, Gemma, da Falcon sun sauya yanayin daga “API ɗaya mai rufe ya dace da kowa” zuwa zaɓuɓɓuka masu yawa. Ka yanke shawara inda gudanawar fassara (GPUs ɗinka, ƙarshen da aka sarrafa, ko ƙarfin da aka rarraba), kuma ka zaɓi ma'amala tsakanin iko, sirri, jinkiri, da farashi. Wannan littafin yana taimaka maka zaɓar hanya madaidaiciya — kuma yana nuna yadda RabaAI yana ba ka damar haɗa hanyoyi ba tare da sauya SDK ba.

Yayin karantawa, ka riƙe ShareAI Kasuwar samfura a buɗe don kwatanta zaɓuɓɓukan samfura, jinkiri na yau da kullum, da farashi tsakanin masu samarwa.

Menene “masaukin LLM na buɗe tushen” ke nufi

• Nauyi na buɗe: sigogin samfura suna wallafa ƙarƙashin lasisi na musamman, don haka za ka iya gudanar da su a gida, a kan-prem, ko a cikin girgije.
• Gudanar da kansa: ka gudanar da uwar garken fassara da lokacin gudu (misali, vLLM/TGI), ka zaɓi kayan aiki, kuma ka kula da tsari, faɗaɗa, da lura.
• Gudanar da masaukin samfura na buɗe: mai samarwa yana gudanar da kayan aiki kuma yana ba da API mai shirye don samfura masu nauyi na buɗe.
• Ƙarfin da aka rarraba: hanyar sadarwar nodes tana ba da GPUs; manufar hanyar ka tana yanke shawarar inda buƙatun ke zuwa da yadda sauyawa ke faruwa.

Me yasa za a gudanar da LLMs na buɗe tushen?

• Iya gyarawa: daidaita akan bayanan yanki, haɗa adaftan, da kulle nau'ikan don tabbatar da maimaituwa.
• Farashi: sarrafa TCO tare da nau'in GPU, batching, caching, da wurin; guji farashin masu tsada na wasu APIs masu rufe.
• Sirri & wurin zama: gudanar da aiki a cikin gida/na yanki don cika manufofi da buƙatun bin doka.
• Jinkiri na wurin zama: sanya fassarar kusa da masu amfani/bayanan; amfani da hanyoyin yanki don rage p95.
• Iya lura: tare da masu samar da kansu ko masu lura da yanayin aiki, za ka iya ganin yawan aiki, zurfin jerin aiki, da jinkiri daga farko zuwa ƙarshe.

Hanyoyi uku don gudanar da LLMs

4.1 Gudanarwa (ba tare da uwar garke ba; biyan kuɗi bisa miliyan alamu)

Menene shi: ka sayi fassarar a matsayin sabis. Babu direbobi da za a girka, babu ƙungiyoyi da za a kula. Ka kafa wani ƙarshen hanya kuma ka kira shi daga aikinka.

Fa'idodi: mafi sauri don samun daraja; SRE da autoscaling suna sarrafawa a gare ka.

Ciniki: farashin kowane alama, ƙuntatawar mai samarwa/API, da iyakancewar sarrafa kayan aiki/telemetry.

Zaɓuɓɓuka na yau da kullum: Hugging Face Inference Endpoints, Together AI, Replicate, Groq (don jinkiri mai ƙanƙanta sosai), da AWS Bedrock. Yawancin ƙungiyoyi suna farawa nan don jigilar sauri, sannan su haɗa BYOI don sarrafawa da tsadar da za a iya hasashe.

4.2 Bude-Tushen LLM Hosting (kai tsaye)

Menene shi: ka kafa kuma ka gudanar da samfurin — akan workstation (misali, 4090), akan sabobin gida, ko girgije naka. Ka mallaki scaling, lura da yanayin aiki, da aiki.

Fa'idodi: cikakken iko akan nauyi/lokaci/telemetry; kyakkyawan tabbacin sirri/mazauni.

Ciniki: kai ne ke daukar nauyin scalability, SRE, shirye-shiryen iya aiki, da daidaita farashi. Cunkoson zirga-zirga na iya zama mai wahala ba tare da buffers ba.

4.3 BYOI + hanyar sadarwa mai rarrabuwa (ShareAI haɗin gwiwa)

Menene shi: haɗin gwiwa ta hanyar ƙira. Kai Ka Zo Da Kayan Aikinka (BYOI) kuma ka ba shi fifiko na farko don fassarar. Lokacin da nodinka ke aiki ko a kashe, zirga-zirga ta kasa ta atomatik zuwa cibiyar sadarwa mai zaman kanta da/ko amintattun masu ba da kulawa — ba tare da sake rubuta abokin ciniki ba.

Fa'idodi: iko da sirri lokacin da kake son su; juriya da sassauci lokacin da kake bukatar su. Babu lokacin rashin aiki: idan ka zaɓi shiga, GPUs ɗinka na iya samun kuɗi lokacin da ba ka amfani da su (Ladabi, Musanya, ko Manufa). Babu kulle mai siyarwa guda ɗaya.

Ciniki: sauƙin saitin manufofi (fifiko, yankuna, iyaka) da sanin yanayin node (kan layi, iya aiki, iyaka).

ShareAI a cikin dakika 30

• API ɗaya, masu ba da kulawa da yawa: duba Kasuwar samfura da sauyawa ba tare da sake rubutawa ba.
• BYOI na farko: saita manufar don na'urorinka su fara karɓar zirga-zirga.
• Komawa ta atomatik: yawaita zuwa cibiyar sadarwa ta ShareAI mai rarrabawa da/ko masu ba da kulawa da ka yarda.
• Tattalin arziki mai adalci: mafi yawan kowanne dalar na zuwa ga masu ba da aikin.
• Samu daga lokacin da ba a amfani: zaɓi shiga kuma bayar da ƙarfin GPU da ba a amfani; zaɓi Kyauta (kuɗi), Musanya (kiredit), ko Manufa (gudunmawa).
• Fara da sauri: gwada a cikin Filin wasa, sannan ƙirƙiri mabuɗi a cikin Kwamitin sarrafawa.. Duba API Farawa.

Yadda BYOI tare da ShareAI ke aiki (muhimmanci ga na'urarka + madaidaicin madadin)

A cikin ShareAI ka sarrafa fifikon hanya ta kowane mabuɗin API ta amfani da fifiko akan Na'urata sauya. Wannan saitin yana yanke shawarar ko buƙatun za su gwada na'urorinka da aka haɗa da farko ko na hanyar sadarwar al'umma farko — amma kawai lokacin da samfurin da aka nema yana samuwa a wurare biyu.

Tsallake zuwa: Fahimci sauyawa · Abin da yake sarrafawa · KASHE (tsoho) · KUNNA (na gida-farko) · Inda za a canza shi · Tsarin amfani · Jerin duba mai sauri

Fahimci sauyawa (kowane mabuɗin API)

Zaɓin yana adana don kowane mabuɗin API. Aikace-aikace daban-daban/muhalli na iya riƙe halayen hanya daban-daban — misali, mabuɗin samarwa da aka saita zuwa al'umma-farko da mabuɗin gwaji da aka saita zuwa na'ura-farko.

Abin da wannan saitin yake sarrafawa

Lokacin da samfurin yana samuwa akan duka na'urarka(na'urori) da hanyar sadarwar al'umma, zaɓin yana zaɓar wane rukuni ShareAI zai tambaya da farko. Idan samfurin yana samuwa a rukuni ɗaya kawai, wannan rukunin za a yi amfani da shi ba tare da la'akari da zaɓin ba.

Lokacin da aka kashe (tsoho)

• ShareAI yana ƙoƙarin rarraba buƙatar zuwa na'urar al'umma mai raba samfurin da aka nema.
• Idan babu na'urar al'umma da ke samuwa don wannan samfurin, ShareAI zai yi ƙoƙarin na'urarka da aka haɗa.

Mai kyau don: sauke lissafi da rage amfani akan injinka na gida.

Lokacin da aka kunna (na gida-da-farko)

• ShareAI yana duba da farko idan akwai wani daga cikin na'urorinka (akan layi kuma yana raba samfurin da aka nema) zai iya sarrafa buƙatar.
• Idan babu wanda ya cancanta, ShareAI zai koma zuwa na'urar al'umma.

Mai kyau don: daidaiton aiki, wurin gida, da sirri lokacin da kuka fi son buƙatun su kasance akan kayan aikin ku idan zai yiwu.

Inda za a canza shi

Bude Dashboard Maɓallin API. Canjawa fifiko akan Na'urata kusa da lakabin maɓalli. Daidaita kowane lokaci ta maɓalli.

Tsarin amfani da aka ba da shawara

• Yanayin sauke aiki (KASHE): Fi son al'umma da farko; na'urarku tana amfani ne kawai idan babu damar al'umma da ake da ita don wannan samfurin.
• Yanayin farko na gida (KUNNE): Fi son na'urarku da farko; ShareAI zai koma zuwa al'umma ne kawai idan na'urarku(na'urorinku) ba za su iya ɗaukar aikin ba.

Jerin duba mai sauri

• Tabbatar da samfurin an raba akan duka na'urarka(na'urori) da al'umma; in ba haka ba zaɓin ba zai yi aiki ba.
• Saita zaɓin akan madaidaicin maɓallin API aikace-aikacenka yana amfani da shi (maɓallan na iya samun zaɓuɓɓuka daban-daban).
• Aika buƙatar gwaji kuma tabbatar da hanyar (na'ura vs al'umma) ta dace da yanayin da ka zaɓa.

Matiriksin kwatance mai sauri (masu samarwa a taƙaice)

Mai bayarwa / Hanya	Mafi dacewa ga	Kundin nauyi mai buɗewa	Daidaitawa	Bayanin jinkiri	Hanyar farashi	Yanki / akan-prem	Komawa baya / gazawa	Daidaituwar BYOI	Lura
AWS Bedrock (Gudanarwa)	Daidaiton kamfani & tsarin AWS	Tsararren saiti (buɗe + mallakar)	Ee (ta SageMaker)	Mai ƙarfi; mai dogaro da yankin	Per buƙata/token	Yanki da yawa	Ee (ta app)	Amincewar dawowa	Ƙarfin IAM, manufofi
Ƙarshen Tunanin Hugging Face (Gudanarwa)	OSS mai dacewa da masu haɓaka tare da nauyin al'umma	Babba ta Hub	Masu daidaitawa & kwantena na musamman	Kyau; autoscaling	Ta hanyar ƙarshen maɓalli/amfani	Yanki da yawa	Eh	Na farko ko madadin	Kwantena na musamman
Tare AI (Gudanarwa)	Ma'auni & aiki akan nauyi a buɗe	Babban kundin adireshi	Eh	Gudun aiki mai gasa	Alamomin amfani	Yanki da yawa	Eh	Kyakkyawan yawaita	Zaɓuɓɓukan horo
Maimaitawa (Gudanarwa)	Samfuri mai sauri & ML na gani	Fadi (hoto/bidiyo/nassoshi)	Ƙayyadadden	Mai kyau don gwaje-gwaje	Biyan kuɗi yayin amfani	Yankunan girgije	Eh	Matakin gwaji	Kwantena na Cog
Groq (Gudanarwa)	Fassarori mai ƙarancin jinkiri sosai	Saita da aka tsara	Ba babban mayar da hankali ba	Ƙarancin p95 sosai	Amfani	Yankunan girgije	Eh	Matakin jinkiri	Kwayoyin al'ada
io.net (Rarraba)	Tanadin GPU mai motsi	Ya bambanta	Babu	Ya bambanta	Amfani	Duniya	Babu	Haɗa kamar yadda ake buƙata	Tasirin cibiyar sadarwa
RabaAI (BYOI + Cibiyar sadarwa)	Sarrafawa + juriya + samun kuɗi	Kasuwa tsakanin masu samarwa	Ee (ta hanyar abokan hulɗa)	Gasa; mai dogaro da manufofi	Amfani (+ zaɓin samun kuɗi)	Hanyar yankin	Na asali	BYOI na farko	Haɗaɗɗen API

Bayanan masu samarwa (karatun gajere)

AWS Bedrock (Gudanarwa)

Mafi dacewa ga: bin ka'idojin kamfanoni, haɗin IAM, sarrafa yankin. Ƙarfi: matsayi na tsaro, jerin samfuran da aka tsara (buɗe + mallakar). Ciniki: kayan aikin da suka fi dacewa da AWS; farashi/gudanarwa suna buƙatar tsari mai kyau. Haɗa tare da ShareAI: adana Bedrock a matsayin madadin da aka ambata don ayyukan da aka tsara yayin gudanar da zirga-zirgar yau da kullum akan ƙananan ku.

Hugging Face Inference Endpoints (Gudanarwa)

Mafi dacewa ga: sauƙin amfani ga masu haɓakawa OSS hosting wanda al'ummar Hub ke tallafawa. Ƙarfi: babban jerin samfura, kwantena na musamman, masu daidaitawa. Ciniki: farashin ƙarshen hanya/fitowa; kula da kwantena don bukatun musamman. Haɗa tare da ShareAI: saita HF a matsayin na farko don takamaiman samfura kuma kunna ShareAI don tabbatar da UX mai laushi yayin ƙaruwa.

Tare AI (Gudanarwa)

Mafi dacewa ga: aiki a sikelin a duk samfuran nauyi-bude. Ƙarfi: ƙarfin gasa, zaɓuɓɓukan horo/daidaitawa, yankuna da yawa. Ciniki: dacewar samfur/ayyuka ya bambanta; gwada farko. Haɗa tare da ShareAI: gudanar da BYOI tushe kuma ƙaru zuwa Tare don daidaitaccen p95.

Maimaitawa (Gudanarwa)

Mafi dacewa ga: saurin ƙirƙira, hanyoyin hotuna/bidiyo, da sauƙin tura. Ƙarfi: kwantena Cog, babban kundin bayanai bayan rubutu. Ciniki: ba koyaushe mafi arha ba don samarwa mai ɗorewa. Haɗa tare da ShareAI: ci gaba da Maimaitawa don gwaje-gwaje da samfuran musamman; tura samarwa ta BYOI tare da ShareAI a matsayin madadin.

Groq (Gudanarwa, kwakwalwan al'ada)

Mafi dacewa ga: fassarar ultra-low-latency inda p95 ke da mahimmanci (aikace-aikacen lokaci-na-gaskiya). Ƙarfi: tsari mai tabbas; kyakkyawan ƙarfin aiki a batch-1. Ciniki: zaɓaɓɓen zaɓin samfur. Haɗa tare da ShareAI: ƙara Groq a matsayin matakin jinkiri a cikin manufar ShareAI ɗinku don ƙwarewar da ba su wuce daƙiƙa ɗaya ba yayin hauhawar.

io.net (Tsarin Rarraba)

Mafi dacewa ga: tanadin GPU mai motsi ta hanyar hanyar sadarwar al'umma. Ƙarfi: faɗin ƙarfin aiki. Ciniki: aiki mai canzawa; manufofi da sa ido suna da mahimmanci. Haɗa tare da ShareAI: haɗa madadin rarraba tare da tushe na BYOI ɗinku don sassauci tare da matakan kariya.

Inda ShareAI ya dace da wasu (shirin yanke shawara)

RabaAI yana zaune a tsakiya a matsayin “mafi kyau daga duniyoyi biyu” mataki. Kuna iya:

• Gudanar da kayan aikin ku na farko (muhimmancin BYOI).
• Fashewa zuwa hanyar sadarwar rarraba ta atomatik lokacin da kuke buƙatar sassauci.
• Zaɓin hanya zuwa takamaiman wuraren sarrafa ƙarshen don jinkiri, farashi, ko dalilan bin doka.

Gudanar da shawara: idan sarrafa bayanai yana da tsauri, saita fifikon BYOI kuma ka takaita dawowa zuwa yankuna/mabukaci da aka amince da su. Idan jinkiri ya fi muhimmanci, ƙara matakin jinkiri mai ƙasa (misali, Groq). Idan ayyuka suna da tsanani, kiyaye BYOI mai sauƙi kuma bari hanyar sadarwar ShareAI ta kama kololuwa.

Gwada lafiya a cikin Filin wasa kafin haɗa manufofi zuwa samarwa.

Ayyuka, jinkiri & amintuwa (tsarin ƙira)

• Tattara & ajiya: sake amfani da ajiyar KV inda zai yiwu; ajiye tambayoyi masu yawan maimaituwa; kwarara sakamako idan yana inganta UX.
• Fassarawa mai tsammani: inda aka tallafa, zai iya rage jinkirin ƙarshe.
• Yanki da yawa: sanya nododin BYOI kusa da masu amfani; ƙara dawowar yanki; gwada dawowa akai-akai.
• Iya lura: bi sawun tokens/sec, zurfin layi, p95, da abubuwan dawowa; inganta iyakokin manufofi.
• SLOs/SLAs: BYOI mai tushe + dawowar hanyar sadarwa na iya cimma manufofi ba tare da yawan tanadi ba.

Gudanarwa, bin doka & wurin bayanai

Gudanar da kansa yana ba ka damar kiyaye bayanai a wurin da ka zaɓa (a kan-prem ko a cikin yanki). Tare da ShareAI, yi amfani da hanyoyin yanki da jerin izini don haka dawowa kawai yana faruwa zuwa yankuna/mabukaci da aka amince da su. Kiyaye rajistan dubawa da sawu a ƙofar ka; rubuta lokacin da dawowa ya faru da kuma zuwa wace hanya.

Takardu na tunani da bayanin aiwatarwa suna zaune a Takardun ShareAI.

Tsarin farashi: gudanarwa vs kai tsaye vs BYOI + rarrabuwa

Yi tunani a cikin CAPEX vs OPEX da amfani:

• Gudanarwa shine tsantsar OPEX: kuna biyan amfani kuma kuna samun elasticity ba tare da SRE ba. Yi tsammanin biyan karin kuɗi a kowane alama don sauƙi.
• Mai masaukin kansa yana haɗa CAPEX/haya, wutar lantarki, da lokacin aiki. Yana yin fice lokacin da amfani ya kasance mai tsinkaya ko babba, ko lokacin da iko ya zama dole.
• BYOI + ShareAI yana daidaita tushe kuma yana barin dawowa don kama kololuwa. Muhimmanci, zaku iya samun kuɗi lokacin da na'urorinku za su kasance ba tare da aiki ba — rage TCO.

Kwatanta samfura da farashin hanya na yau da kullum a cikin Kasuwar samfura, kuma ku kalli Saki ciyarwa don sabbin zaɓuɓɓuka da ragin farashi.

Mataki-mataki: farawa

Zaɓi A — Gudanarwa (ba tare da uwar garke ba)

• Zaɓi mai bayarwa (HF/Together/Replicate/Groq/Bedrock/ShareAI).
• Sanya wani ƙarshen don samfurinku.
• Kira shi daga aikace-aikacenka; ƙara sake gwadawa; saka idanu p95 da kurakurai.

Zaɓi B — Buɗaɗɗen-Tushen LLM Hosting (kai-tsaye)

• Zaɓi lokacin gudu (misali, vLLM/TGI) da kayan aiki.
• Sanya a cikin kwantena; ƙara ma'auni/masu fitarwa; saita autoscaling inda zai yiwu.
• Sanya tare da ƙofa; yi la'akari da ƙaramin sarrafa dawowa don inganta jinkirin ƙarshen.

Zaɓi C — BYOI tare da ShareAI (hade)

• Sanya wakilin kuma rajista node(s) ɗinka.
• Saita fifiko akan Na'urata bisa maɓalli don dacewa da niyyarka (KASHE = al'umma-farko; KUNNA = na'ura-farko).
• Ƙara dawowa: hanyar sadarwar ShareAI + masu samarwa da aka ambata; saita yankuna/kason.
• Kunna lada (na zaɓi) don haka kayan aikin ka yana samun kuɗi lokacin da ba a amfani da shi.
• Gwada a cikin Filin wasa, sannan aika.

Yankunan lambar

1) Sauƙaƙan samar da rubutu ta hanyar ShareAI API (curl)

curl -X POST "https://api.shareai.now/v1/chat/completions" \"

2) Kira iri ɗaya (JavaScript fetch)

const res = await fetch("https://api.shareai.now/v1/chat/completions", {;

Misalai na ainihi

Mai gina kansa (nvidia rtx 4090 guda ɗaya, masu amfani na duniya)

BYOI yana sarrafa zirga-zirgar rana; hanyar sadarwar ShareAI tana kama ƙaruwa na yamma. Jinkirin rana yana kusa da ~900 ms; ƙaruwa ~1.3 s ba tare da 5xx yayin kololuwa ba. Awanni marasa aiki suna samar da Lada don rage kuɗin wata-wata.

Hukumar kirkira (ayyuka masu ƙaruwa)

BYOI don matakin gwaji; Replicate don samfuran hoto/bidiyo; ShareAI madadin don ƙaruwa na rubutu. Ƙananan haɗarin ƙare lokaci, ƙarin p95 mai ƙarfi, kashe kuɗi mai tsari ta hanyar ƙididdiga. Editoci suna duba hanyoyin a cikin Filin wasa kafin fara samarwa.

Kamfani (bin doka + yankuna)

BYOI a kan-prem EU + BYOI US; madadin an iyakance zuwa yankuna/mastoci da aka amince da su. Yana gamsar da zama, yana kiyaye p95 mai tsari, kuma yana ba da cikakken bayanin duk wani sauyin madadin.

Tambayoyi akai-akai

Tunani na ƙarshe

Gudanarwa yana ba ka sauƙin sabarless da sikeli nan take. Mai masaukin kansa yana ba ka iko da sirri. BYOI + ShareAI yana ba ka duka biyun: kayan aikinka na farko, sauyawa ta atomatik lokacin da kake buƙata, kuma kudaden shiga lokacin da ba ka yi ba. Lokacin da ka yi shakka, fara da ɗaya node, saita fifikon kowane maɓalli don dacewa da nufinka, kunna ShareAI fallback, kuma ka ci gaba da gwaji tare da ainihin zirga-zirga.

Bincika samfura, farashi, da hanyoyi a cikin Kasuwar samfura, duba Saki don sabuntawa, kuma duba Takardu don haɗa wannan cikin samarwa. Tuni mai amfani? Shiga / Yi rijista.