HN jira: Badge min b’a jira i ka codebase bɛ bɛn cogo min na LLM ka context window kɔnɔ

aw ka kodɔnba bɛ ni mɛtɛrɛ kura ye min nafa ka bon tiɲɛ na

San tan caman kɔnɔ, baarakɛlaw bɛ u ɲɛmajɔ kode sinsinniw na, sikɔlɔmatiki gɛlɛyaw, sɛgɛsɛgɛliw dafalen kɛmɛsarada la, ani u bilali caya. Nka mɛtɛrɛ kura dɔ bɛ ka injiniyɛri jɛkuluw bɛ miiri cogo min na u ka kodebasew la, o cogoya sɛmɛntiya ni dususuma ye: context window fit — i ka kodebase bɛɛ kɛmɛsarada la, LLM bɛ se ka min sɛgɛsɛgɛ ɲininkali kelen kɔnɔ. A bɛ i n’a fɔ a nɔgɔn don nanbara la, nka nin jate in bɛ ka kɛ taamasiyɛn nafama dɔ ye min b’a jira ko i ka kulu bɛ se ka baara kɛ ni yiriwali baarakɛminɛnw ye minnu bɛ dɛmɛ sɔrɔ AI fɛ cogo min na. Wa n' i b' i ɲɛmajɔ a la , i bɛ sɔrɔko tɔnɔba to tabali kan .

Hakilina in ye jɔrɔnanko sɔrɔ kɔsa in na yiriwalikɛlaw ka jɛkuluw kɔnɔ porozɛ dɔ bɔlen kɔfɛ min bɛ badge nɔgɔman dɔ lawuli — a tɛ danfara bɔ i ka jɔli-tɛmɛsira walima datugulan lakanani dɔntaw la — min b’a jira tigitigi i ka marayɔrɔ hakɛ bɛ bɛn LLM kɔnɔko finɛtiriw kɔnɔ minnu bɛ fɔ kosɛbɛ. A ye baro nafama kabakoma dɔ lawuli kodebase jɔcogo kan, monorepos ni microservices, ani n’an ka kan k’an ka kode for AI faamuyali dilan. O nɔfɛkow bɛ boli ka taa a fɛ ka tɛmɛ yiriwalikɛlaw fanba tun bɛ min faamuya fɔlɔ.

Context Window bɛ bɛn min ma, o bɛ suman tiɲɛ na

Kanko misali belebele bɛɛ bɛ baara kɛ sigida danmadɔ kɔnɔ — a bɛ se ka sɛbɛnni hakɛ min baara siɲɛ kelen. GPT-4 Turbo bɛ token 128K ɲɔgɔn minɛ. Claude ka modeli labanw bɛ tɛmɛn 200K tokenw kan. Gemini b’a fɔ ko miliyɔn kelen ni kɔ. N'i y'i ka kodebase (kode-base) balo o modɛliw dɔ la kelen na walasa ka sɛgɛsɛgɛli kɛ, ka hakilinaw labɛn kokura, walima ka bug detection (bug detection) kɛ, modeli bɛ se ka "ye" dɔrɔn min bɛ bɛn o finɛtiri kɔnɔ. a kɔfɛ fɛn o fɛn bɛ ye , o bɛɛ tɛ ye , i n' a fɔ a tɛ yen .

Context window fit bɛ i ka kodebase hakɛ bɛɛ lajɛlen (tokenw kɔnɔ) ni modeli dilen dɔ ka context window cɛ hakɛ suman. Dakun min bɛ tokenize ka kɛ 80K tokenw ye, o bɛ se ka 100% bɛn 200K-token modɛli kɔnɔ — AI bɛ se k’i ka porozɛ bɛɛ faamuya tɛmɛsira kelen kɔnɔ. A ye 2 miliyɔn token monorepo ye wa? I bɛ kɛmɛsarada la jateden kelenw lajɛ, o kɔrɔ ye ko AI bɛ baara kɛ ni yɔrɔw ye, a tɛ ja dafalen faamu abada. O danfara in nafa ka bon kosɛbɛ AI ka kode hakilinaw, fɛn dilanni seginnkanniw, ani otomatiki refactoring.

Badge hakilina bɛ o kɛ kristali ye ka kɛ mɛtɛrɛ yelen ye, min bɛ se ka tila. Aw k’a nɔrɔ aw ka README kɔnɔ aw ka CI cogoya n’a ka coverage kɛmɛsarada la. A bɛ fɛn dɔ fɔ dɛmɛbagaw ni ladonbagaw ye min nafa ka bon tiɲɛ na : nin kodɔn in bɛ AI-terikɛ cogo di ?

Mun na nin mɛtɛrɛ in bɛ fɛn caman Changé ekipuw bɛ porogaramuw dilan cogo min na

Porozɛw dilanni latigɛw kɛra hadamadenw haminankow de ye tuma bɛɛ — kalanni, ladonni, baarakɛcogo, jɛkulu sigicogo. Context window fit bɛ mɔgɔ kura dɔ don nin barow kɔnɔ: AI pair programmer. Ni i ka kodebase bɛɛ bɛ bɛn context window kɔnɔ, AI baarakɛminɛnw bɛ se ka hakilinata di haminankow kan minnu bɛ ɲɔgɔn cɛ, ka cakɛda minnu bɛ tali kɛ ɲɔgɔn na, minnu ka nɔgɔn, ani ka fɛn dɔw jira minnu bɛ kɛ sababu ye ka sistɛmu dafalen jira. Ni a ma kɛ, i b’a ɲini AI dɔ fɛ a jɔyɔrɔba la, a k’i ka dumunikɛyɔrɔ labɛn kokura k’a sɔrɔ a bɛ sanuyaso dɔrɔn de jira a la.

o bɛ na ni kɔlɔlɔ waleyali ye , ɛntɛrinɛti ɲɛmɔgɔw bɛ ka daminɛ ka minnu ta kosɛbɛ . Ekipu minnu ka jatedenw bɛ bɛn sigida ma, olu bɛ jaabi ɲumanw fɔ minnu bɛ se ka suman ka bɔ AI kode lajɛ baarakɛminɛnw na. Bug detection rates bɛ ɲɛ ka da a kan model bɛ se ka execution siraw ladege files (dosiyɛriw) kɔnɔ. Ladilikan minnu bɛ kɛ ka fɛnw labɛn kokura, olu bɛ kɛ fɛn ye min bɛ se ka kɛ fɛn dilanni siratigɛ la sanni u ka kɛ sigida la fɛn ɲumanw ye nka u bɛ tiɲɛni kɛ diɲɛ kɔnɔ. Ɛntɛrinɛti jɛkulu dɔ min bɛ SaaS baarakɛda cɛmancɛ la, o ye sɛbɛn kɛ 40% dɔgɔyali kan AI ka kɔseginw la u ka monorepo tilalen kɔfɛ ka kɛ baarakɛminɛn misɛnninw ye, minnu bɛ bɛn sigida ma.

mɛtiri fana bɛ fanga baara dɔ dabɔ ɛntɛrinɛti wale ɲumanw kama , ekipuw ka kan ka tugu minnu kɔ cogo o cogo . Kodebasi minnu bɛ jate ɲuman sɔrɔ context window fit kan, olu ka teli ka kɛ ni modulu dancɛ saniyalen ye, kode salen ka dɔgɔ, haminankow farali ka ɲɛ, ani ka fɛnmarayɔrɔw sinsin kosɛbɛ. AI faamuyali jateminɛ bɛ laban ka kɛ proxy ye kode kɛnɛya bɛɛ lajɛlen na.

Mɔgɔ si tun ma miiri min na , fɛn dilanni na

Baro min kɛra context window fit lamini na, o ye monorepo ni polyrepo sɔsɔli lawuli kokura ni fan kura ye pewu. Monorepo lafasabagaw y’a jira kabini tuma jan ko ka fɛn bɛɛ mara fɛnmarayɔrɔ kelen kɔnɔ, o bɛ danfarako ɲɛnabɔli nɔgɔya, ka atomikɛlanw ka cɛsiriw kɛ baarakɛminɛnw bɛɛ kɔnɔ, ani ka dɔ bɔ jɛ-ka-baara dimi na. Nka n’i ka monorepo tokenize ka se token miliyɔn 5 ma ani context window min bɛ sɔrɔ a bɛɛ la ɲuman ye 200K ye, i ye codebase dɔ Dabɔ min AI baarakɛminɛn si tɛ Se k’a Faamu kosɛbɛ.

O kɔrɔ tɛ ko monorepow sara — a yɔrɔ ka jan a la . Ekipu hakilitigiw bɛ ka cɛmancɛ-yɔrɔ sɔrɔ. Fɛɛrɛ minnu bɛ ka bɔ kɛnɛ kan, olu dɔw ye :

• hakilila tigɛli : Ka baara kɛ ni .contextignore filew ye (minnu ni .gitignore bɛ tali kɛ ɲɔgɔn na) walasa ka kode dabɔlenw, feerekɛlaw ka dantigɛliw, ani kɔrɔbɔli fixtures bɔ AI sɛgɛsɛgɛli kɔnɔ
• Module-level context maps : manifeste nɔgɔmanw dilanni minnu bɛ AI baarakɛminɛnw dɛmɛ u k’a faamu ko dosiye minnu bɛ tali kɛ fɛn minnu na k’a sɔrɔ u ma fɛn bɛɛ doni
• Jɛkulu sɛbɛnni i n’a fɔ sigida : Ka fara fɛn dilanni latigɛ sɛbɛn kunkurunninw kan (ADR) minnu bɛ AI sigicogo faamuyali di k’a sɔrɔ a ma wajibiya a ka jɛɲɔgɔnyaw jateminɛ ka bɔ kode dɔrɔn na
• baarakɛminɛnw bɔli fɛɛrɛtigɛ : Ka modulu yɛrɛmahɔrɔnyalen lakikaw tila ka kɛ fɛnmarayɔrɔ danfaralenw ye k' a sɔrɔ haminanko lakikaw t' u la ni sigida jɔnjɔn ye

Hakilila jɔnjɔn ye ko ka ɲɛtaa kɛ ka ɲɛsin context window fit ma, o tɛ i ka kodebase dɔgɔyali ye — a bɛ kɛ sababu ye k’a kɛ faamuyali ye kosɛbɛ, AI baarakɛminɛnw ni hadamaden minnu bɛ baara kɛ u kɛrɛfɛ.

I yɛrɛ ka kodɔn sumani : Baarakɛda min bɛ kɛ

Sanni i ka daminɛ ka i ka sistɛmu bɛɛ labɛn kokura walasa ka badge metric dɔ nɔfɛ, a ka kan ka faamuya cogo min na i bɛ se ka context window fit suman cogo kɔrɔ la. I ka marayɔrɔ bɛɛ ka token jate raw ye daminɛ ye, nka a ye minɛn ye min bɛ kɛ ni bɔgɔbɔgɔ ye. Fɛɛrɛ min bɛ nuancɛ kosɛbɛ, o bɛ jateminɛ kɛ AI ka kan ka min ye tiɲɛ na baara suguya wɛrɛw kama.

ye

"Ɲininkali lakika tɛ n’i ka kodebase bɛɛ bɛ bɛn sigida finɛtiri dɔ ma — a ye ko kodɔn min bɛ tali kɛ baara dilen o baara dilen na, o bɛ bɛn. Kodebase min labɛnna ka ɲɛ ni dancɛ jɛlenw ye, o b’a to AI baarakɛminɛnw bɛ u mago bɛ fɛn min na tigitigi, o doni, hali n’a y’a sɔrɔ a marayɔrɔ bɛɛ ka bon kosɛbɛ."

💡 DID YOU KNOW?

Mewayz replaces 8+ business tools in one platform

CRM · Invoicing · HR · Projects · Booking · eCommerce · POS · Analytics. Free forever plan available.

Start Free →

Walisa ka sumanikɛcogo ɲuman sɔrɔ, i k’a daminɛ ni i ka baarakɛminɛn koloma kode tokenize ye — ka node_modules, vendor directories, build artifacts, ani generated files bɔ o la. Bi tokenizɛri fanba (i n’a fɔ OpenAI ka tiktoken walima Anthropic ka token jatebɔ fɛɛrɛ bɔlenw) bɛ se ka ɲɛbilasɛbɛn dɔ baara segin kɔnɔ. Aw ye o jaabi suma ni aw ka kulu bɛ baara kɛ ni misali minnu ye tiɲɛ na, olu ka contexte windows (yɔrɔ) finɛtiriw ye. Ni i ka baarakɛminɛn koloma kode bɛ bɛn context window kelen kɔnɔ ni yɔrɔ ye min bɛ se ka kɛ ka ɲɛsin ɲininkaliw ni cikanw ma, i bɛ cogo ɲuman na. Ni a tɛmɛna finɛtiri kan ni 2-5x ye, strategic chunking bɛna kɛ wajibi ye. 10x kɔfɛ, i b’a fɛ ka wari bila fɛn dilanni caman caman cili la walima RAG (retrieval-augmented generation) pibiliki kɛrɛnkɛrɛnnenw na walasa ka AI baarakɛminɛnw kɛ nafama ye.

Ekipu minnu bɛ jɔ plateformew kan i n’a fɔ Mewayz, yɔrɔ min na modulari architecture bɛ haminankow faranfasi kaban ka kɛ modulu danfaralenw ye — CRM, fatura, HR, analytics, ani 200 ni kɔ wɛrɛw — nin sumani in bɛ kɛ kɛrɛnkɛrɛnnenya la nafama ye. Module kelen-kelen bɛɛ bɛ baara Kɛ i n’a fɔ a yɛrɛ-yɛrɛ-yɔrɔ min bɛ ni ‘interfaces (interfaces) jɛlenw ye, min bɛ karti cogo la ka Kɛ context-window-friendly chunks (yɔrɔ-ko-finɛtiriw) ye. O ye fɛn dilanni misali sugu ye min bɛ nafa sara hadamadenw ka ladonni ni AI faamuyali fila bɛɛ la.

Yɛlɛmabaa jɛkulu bɛ ka sɔsɔli min kɛ tiɲɛ na

Hacker News baro min kɛra context-window badges lamini na, o ye gɛlɛya caman bɔ kɛnɛ kan minnu bɛ mɔgɔ ɲɛnajɛ yiriwalikɛlaw ka jɛkulu kɔnɔ. Fɔlɔ ye filozofi ye : yala an ka kan ka kode dilan AI tacogo kama wa ? Saniyabagaw b’a fɔ ko kode ka kan ka sɛbɛn hadamadenw de kama fɔlɔ, wa AI baarakɛminɛnw ka kan ka ladamu. Pragmatistw bɛ sɔsɔli kɛ ko ni fɛn dilanni sugandili nɔgɔman dɔ y’a to i ka kulu bɛ se ka baara kɛ 30% ni AI baarakɛminɛnw ye ni musaka zeru ye hadamaden ka kalanni na, ka ban k’a kɛ hakilina ye ka tɛmɛ ɛntɛrinɛti kan.

Jɛkafɔ filanan sinsinnen bɛ a kan ko context window fit yɛrɛ ye mɛtɛrɛ sabatilen ye min nafa ka bon ka tugu ɲɔgɔn kɔ . Context windows yiriwara kosɛbɛ — k’a ta 4K tokenw na GPT-3.5 daminɛ na ka se miliyɔn kelen ni kɔ ma Gemini 1.5 Pro kɔnɔ. Ni finɛtiriw tora ka bonya, bi ta "tɛ bɛn" bɛ kɛ sini ta "bɛn nɔgɔya la". Nka ɛntɛrinɛti injiniyɛri minnu bɛ se kosɛbɛ, olu b’a jira ko hali ni contexte windows belebelebaw bɛ yen, modeli baarakɛcogo bɛ dɔgɔya ni contexte janya ye. Modeli min bɛ baara kɛ ni 50K tokenw ye minnu sinsinnen don, minnu bɛ tali kɛ kode la, o bɛna tɛmɛ o modɛli kelen kan min bɛ baara kɛ ni 500K tokenw ye monorepo jɛnsɛnnen dɔ la, hali n’a y’a sɔrɔ u fila bɛɛ "bɛnna" fɛɛrɛko siratigɛ la. Kow kɛcogo ɲuman nafa ka bon i n’a fɔ hakɛ.

Baro sabanan , min bɛ se ka kɛ kosɛbɛ , o bɛ lamini baarakɛminɛnw de kan . Dabɔbaa b’a fɛ IDE integrations (daɲɛw) minnu bɛ context-aware (daɲɛw) dɔn minnu bɛ a jira u yɛrɛma ko file minnu ka kan ka don a kɔnɔ ni u bɛ kode ci AI dɔ ma. U b’a fɛ hakilitigiya sɔrɔ min bɛ sɔrɔ marabolo-dakun na min bɛ modulu dancɛw faamuya ni bololabaarakɛcogo tɛ. Porozɛ dafalen caman bɛ ka nin gɛlɛya tigitigi in ɲɛnabɔ sisan, ka fɛn hakɛ min jɔ "contexte compilateurs" la minnu bɛ dosiyew kulu ɲuman lajɛ AI dɛmɛni baara o baara dilen na.

Ka nin tigɛli kɛ ɲɔgɔndan nafa ye

Jagokɛlaw fɛ — a tɛ kɛ yiriwalikɛlaw ka jɛkuluw dɔrɔn ye — context window fit bɛ ni jiginɛw nɔfɛkow ye minnu ka kan ka faamuya. Baarakɛda minnu bɛ porogaramuw ci joona, ni bugw ka dɔgɔ, ni musaka dɔgɔman ye, olu bɛ se sɔrɔ u ka suguw la. AI dɛmɛni yiriwali ye fanga cayalan lakika ye, nka ni a jukɔrɔla kodebase labɛnna walasa ka nafa sɔrɔ a la dɔrɔn. Jɛkulu minnu bɛ wari bila AI-friendly codebases kɔnɔ bi, olu bɛ ka nafa camanba jɔ minnu bɛna bonya waati tɛmɛnen kɔfɛ.

Nin sariyasen in bɛ tɛmɛn porozɛw ka tɔn saniyalenw kan. Jagokɛlaw minnu bɛ baara kɛ plateformew kan i n’a fɔ Mewayz, n’o bɛ CRM, fatura, sarako, HR, fleet management, ani analytics fara ɲɔgɔn kan ka kɛ modular system kelen ye, olu bɛ nafa sɔrɔ o hakilina kelen in na operational taabolow la. Ni i ka jago kunnafoniw bɛ balo moduluw kɔnɔ minnu labɛnna ka ɲɛ, minnu sirilen don ɲɔgɔn na sanni u ka jɛnsɛn SaaS baarakɛminɛn 15 kɔnɔ minnu ni ɲɔgɔn cɛ tɛ, AI bɛ se ka hakilinata di i ka baara bɛɛ kan — ka misaliw dɔn feereli, dɛmɛ ani wariko siratigɛ la minnu tɛna ye silo sitɛmuw kɔnɔ. O sariyakolo kelen min bɛ kodebase kɛ AI-friendly ye, o bɛ jago kɛ AI-friendly ye : sigicogo jɛlen, dancɛ saniyalen, ani sigida bɛɛ lajɛlen.

taamacogo waleyali injiniyɛri ɲɛmɔgɔw bolo , o ye tilennen ye . Aw bɛ aw ka context window fit sumani daminɛ bi — hali cogo la min tɛ labɛn ye. Aw b’a fara aw ka ɛntɛrinɛti kɛnɛya dakunw kan jɔli waatiw ni sɛgɛsɛgɛli dafalen kɛrɛfɛ. Baara kɛ n’a ye i n’a fɔ donnafɛn kelen (a tɛ donnafɛn kelen ye) n’i bɛ fɛn dilanni latigɛw kɛ. Ani k’a dɔn ko kodebazi minnu bɛna nafa caman sɔrɔ AI yiriwali baarakɛminɛn nataw la, olu de bɛ labɛn walasa ka faamuya sisan yɛrɛ.

Bada ye baro daminɛ ye , a tɛ taayɔrɔ ye

README badge min bɛ "87% context fit — Claude 200K" jira, o ye ko fitinin ye. A bɛ seginkanni Kɛ ka 'sèn kelen Sɔrɔ ani k'a Minɛ i ka poroze sɛbɛnw kɔnɔ. Nka a bɛ min jira — cɛsiri min bɛ kɛ ni laɲini ye, min bɛ se ka suman, ka ɲɛsin kodebase faamuyali ma — o bɛ fɛn dɔ jira min kɔrɔ ka bon ɛntɛrinɛti jɛkulu dɔ ka fɔlɔfɔlɔ kow kan. A b’a fɔ ko : an bɛ miiri an ka kode bɛna faamuya cogo min na, a tɛna kɛ baarakɛla nata dɔrɔn fɛ, nka AI sitɛmuw fɛ minnu bɛ ka caya yiriwali baara bɛɛ kɔnɔ.

Nin taabolo in kɔlɔlɔ min nafa ka bon kosɛbɛ , o tɛ badge yɛrɛ ye . O ye barow ye a bɛ minnu lawuli fɛn dilanni lajɛw senfɛ, boli bolodacogo, ani fɛɛrɛ juruw barow senfɛ. Ni "context window fit" kɛra i ka ɛntɛrinɛti daɲɛw kɔnɔko dɔ ye, i b’a daminɛ ka latigɛw kɛ minnu bɛ kɛ ka bɛn an ye fɛn o fɛn dɔn porozɛw dilanni ɲuman kan san tan caman kɔnɔ : module misɛnninw, minnu sinsinnen don ni interfaces jɛlenw ye ani coupling fitinin. AI ka jyɛnnatigɛ ma o sariyakolow Dabɔ. Nka a bɛ dalilu kura di ekipuw ma, min bɛ se ka jate, walasa u ka laban ka tugu u kɔ.

Ɲininkali minnu bɛ kɛ tuma caman na

context window fit ye mun ye ani mun na a nafa ka bon ?

Context window fit bɛ i ka kodebase kɛmɛsarada la hakɛ suman LLM bɛ se ka baara kɛ ni ɲininkali kelen ye. Kɛmɛsarada la min ka ca, o kɔrɔ ye ko AI baarakɛminɛnw bɛ se k’i ka porozɛ caman faamuya siɲɛ kelen, o bɛ na ni kode ladilikan ɲumanw ye, ka fɛnw kɛcogo kura kɛ ka ɲɛ, ani ka hakilijagabɔ dɔgɔya. Ni AI dɛmɛni yiriwali bɛ kɛ sariya ye, o mɛtɛrɛ in bɛ nɔ bila tigitigi i ka jɛkulu bɛ se ka nafa sɔrɔ cogo min na ni baarakɛminɛnw ye i n’a fɔ Copilot, Cursor ani Claude.

ne bɛ se ka n ka kodebase ka kɔnɔko finɛtiri fitinin lajɛ cogo di ?

I bɛ se ka baara kɛ ni badge dafalen baarakɛminɛn ye min bɛ tila Hacker News kan walasa ka taamasiyɛn yelen dɔ dilan i ka marayɔrɔ kama. A b’i ka codebase token jate bɛɛ jate ani k’a suma ni LLM context windows ye minnu bɛ fɔ kosɛbɛ. Badge bɛ kɛmɛsarada la jate jira i bɛ se ka min don i ka README kɔnɔ, ka dɛmɛbagaw ni bololabaarakɛlaw bila ka ja teliya la, i ka porozɛ labɛnnen don cogo min na AI la.

fɛɛrɛ jumɛnw bɛ kodɔnba dɔ ka context window fit score ɲɛ ?

Ka sinsin modulu architecture kan, haminankow farali jɛlen kan, ani ka kode salenw ban. Monorepo minnu labɛnna ka ɲɛ ni dancɛw ye minnu bɛ bɛn hakili ma, olu b’a to LLMw ka modulu nafamaw baara u yɛrɛma. Kode fila-fila dɔgɔyali, ka dosiyew kɛ ka surunya, ani ka dependency jiriw saniyalenw mara, o bɛɛ bɛ dɛmɛ don. Plateformes i n’a fɔ Mewayz bɛ nin sariyakolo in jira — ka modulu 207 pake ka kɛ jago OS nɔgɔman ye min dabɔra ladonni ni baarakɛcogo ɲuman kama.

yala kodebase fitinin kɔrɔ ye tuma bɛɛ AI bɛnkan ɲuman ye wa ?

A man kan ka kɛ . Kodebase fitinin min bɛ ni dependencies tangled ani documentation jugu ye, o bɛ se ka gɛlɛya LLMw ma ka hakilinata kɛ ka tɛmɛ a belebeleba kan, min labɛnna ka ɲɛ. Min nafa ka bon, o ye ko sigida min bɛ bɛn o ma, o bɛ bɛn finɛtiri kɔnɔ cogo min na. Abstraction saniyalenw, tɔgɔdacogo basigilenw, ani modular design b’a to AI baarakɛminɛnw ka baara kɛ ka ɲɛ hali n’u tɛ se ka kode layini bɛɛ minɛ siɲɛ kelen.