API MiMo Kore-ārai Tirohanga
Koinā anake te tauira o tā mātou rārangi e hono ana i te pono taumaha 8-pī ki te māramatanga whakaahua — ā, he utu ōrite tonu ki ngā tauira kuputuhi-anake o te utu papa: $0.50 tāuru / $2.00 whakaputa ia miriona tohu.
Whakarāpopoto
MiMo-V2.5-W8A16 ko te mema whai-tirohanga o te takirua MiMo i runga i te Kūwaha Shannon AI, he uri nō te raina MiMo a Xiaomi — he tauira reo whakaaroaro-tuatahi kua whakapiritia ki tētahi pou tirohanga — ā, e whakahaerea ana i te W8A16, tā mātou taumata quantization pono-rawa, i runga i tā mātou ake tautau GPU. Horopaki 262,144-tohu, 16,384 tohu whakaputa nui rawa, te karanga taputapu, te schema JSON pūmau, te streaming, me tētahi pae reasoning_effort. Kāore he paparanga whakakāhore, kāore he tātari whakaputa. Ka tautau ki a MiniMax-M3 mō te tirohanga utu-iti rawa o te rārangi, ā ka toa i a ia mō te pono.
He utu-iti ngā pito ara reo-maha nā te mea kua tapahia tētahi mea, ā, ko te mea ka tapahia te nuinga o te wā ko te wāhanga kāore e taea e koe te kite: kua kōpakina ngā taumaha ki te toru, whā pī rānei, ā, ka whakautu tonu te tauira mā te whakapono nui mō tētahi tūtohi kāore ia e tino kaha ki te pānui. Ko MiMo-V2.5 te tauira whakahē i tā mātou rārangi — kei te utu papa me te taumata pono-rawa i te wā kotahi, nā te mea i hangaia tōna tūāpapa hei mea iti, kiato, kaua hei mea nui rawa, korekore.
01Nō hea a MiMo: he tauira whakaaroaro tuatahi, he tauira tirohanga tuarua
Ko te raina MiMo te kaupapa rangahau taumaha-tuwhera a Xiaomi, ā, he tautohenga kaumātua tōna tīmatanga: ka whakatakotoria te āheinga whakaaroaro i te wā whakangungu-tōmua, kaua e akiakihia whai muri mā te RL. He mahi hangarau-raraunga te nuinga — i hanga anō e Xiaomi te tango me te tātari kuputuhi kia piki ake te kiato o ngā tauira whakaaroaro i te kohinga, i hanga i ngā raraunga whakaaroaro hangai i te rahi, ā, i whakangungua-tōmua a MiMo-7B-Base ki te tata 25 taera tohu i raro i tētahi whakaranu e toru wāhanga, me te Multiple-Token Prediction hei whāinga āwhina e mahi tahi ana i te wā whakatau hei kaiwhakawetewete matapae (te reiti whakaaetanga tata ki te 90%, e ai ki a Xiaomi).
He mea whāiti-a-hinengaro te whakangungu-whai-muri: 130,000 rapanga pāngarau me te waehere, ko ngā utu tika-ā-ture anake, kāore he tauira utu kua akona, i runga i te whakaaro he mata hackable tētahi tauira utu-ā-neural. Kāore e taea te kōrero-māware i ngā utu whakamanahia. Ka whakaputa a MiMo-7B-RL i ngā tau kāore e ōrite ki tōna rahi — e pūrongo ana a Xiaomi i te 68.2% i te AIME 2024, te 95.8% i te MATH-500, te 57.8% i te LiveCodeBench v5 i te pass@1, e ōrite ana ki a o1-mini a OpenAI. Koirā te āhua o te tauira e karangahia ana e koe: kiato, kua whakangungua-hōhonu, kua whakatikaina kia mahi whakaaro puta i tētahi rapanga, kaua ki te maumahara noa i tētahi whakautu — ā, ka ora tonu taua tū āhua ki roto i ngā uri reo-maha.
02I pēhea a MiMo i ako ai ki te titiro: te hoahanga MiMo-VL
Ko MiMo-VL te peka reo-tirohanga, ā, he mea whakaako tōna hangahanga. E toru ngā wāhanga i whakahiatohia e Xiaomi: he kaiwhakawaehere ViT taumata-whakaahua-taketake e pupuri ana i te taipitopito menemene kaua e akiaki ana i ia whakaahua puta i tētahi ripanga taumata-iti pūmau, he kaiwhakaputa MLP kua tīmatahia matapōkere ki roto i te ahanoa huna o te tauira reo, ā — te mea nui rawa — ko te tuarā whakaaroaro MiMo hei tauira reo. Ka riro i te pūnaha tirohanga tētahi kaiwhakawetewete kua pai kē ki te whakaaro maha-hipanga, kaua ki tētahi e pai noa ana ki te tuhi kupu-tāpiri.
I rere te whakangungu i ngā wāhanga whakangungu-tōmua e whā — te whakamahana kaiwhakaputa, te whakahāngai reo-tirohanga, te whakangungu-tōmua reo-maha whānui, te SFT horopaki-roa — puta noa i te tata 2.4 taera tohu, kātahi ka whai i te Mixed On-policy Reinforcement Learning (MORL), e kōpaki ana i te tika o te whakamahara, te pono o te whakaū-ā-kite, te tika o te whakaaroaro, me te hiahia tangata/AI ki roto i te ara aro-kaupapa kotahi. E kī pono ana te pūrongo he mea nekeneke te whakatikatika i ēnā katoa i te wā kotahi — he anga pono ake i tā te nuinga o ngā pepa reo-maha e whakaputa ana.
He kaha ngā hua kua whakaputaina mō te tatauranga tawhā. E pūrongo ana a Xiaomi i te toa a MiMo-VL-7B-RL i a Qwen2.5-VL-7B i te 35 o ngā 40 mahi, te 59.4 i te OlympiadBench (kei runga ake i ngā tauira tae atu ki te 78B), me te 56.1 i te OSWorld-G mō te whakaū-ā-kite GUI, kei mua i ngā kaihautū atanga i hangaia mō taua take anake. Ko te whakahoutanga 2508 i muri mai i whakaeke i te MMMU ki tua atu i te 70 mō te wā tuatahi i te raupapa (70.6), i eke ki te 94.4 i te ChartQA, te 92.5 i te ScreenSpot-v2 me te 70.8 i te VideoMME, ā, i tāpiritia tētahi rīpene whakahaere-whakaaroaro kia taea ai te whakaweto i ngā aho whakaaroaro ia tono.
He kupu mō ngā kerēme putanga
Ko ia tau kei runga ake nei he tau nā Xiaomi ake i whakaputaina mō tētahi tohu-oti kua tukuna motuhake, kua hono atu ki te poraka puna kōrero i raro iho. He wāhi i muri mai i taua whakapapa tā mātou urunga rārangi. Kei te whakaahua mātou i te nōhea mai te tauira me te mahi i hangaia ai te whānau — kaua e kī ana ka whakawhiti tika ērā tau ki te hanga e karangahia ana e koe. Paearutia tāu ake kawenga mahi; koirā te take o te utu papa.
03He aha te tikanga o W8A16, ā he aha ai e nui ake ana mō te tirohanga i te kuputuhi
Ka kawe ia tauira o te kūwaha i tētahi taunga quantization. Ehara i te whakapaipai noa — e whakaatu ana ki a koe he aha i hokona kia taea ai te whakahaere i te tauira.
| Taunga | He aha tēnei | He aha i whakamahia ai |
|---|---|---|
| 3BIT-REAP | He taumaha 3-pī me te REAP, he tikanga tapahi-mātanga e muku ana i ngā mātanga onoono-whakarite-ara mai i tētahi tauira MoE i mua i te quantizing | Te huarahi anake e uru ai, e whakahaerea ai ngā tūāpapa nunui rawa i tētahi tere whaitake |
| W4A16-AUTOROUND-REAP | He taumaha 4-pī, hohenga 16-pī, kua quantized mā AutoRound — te porowhita kua akona e whakahoki mai ana i te tika i tua atu i te porowhita-ki-te-tata-rawa māori — me te tapahi REAP | Te taumata waenga mō ngā tūāpapa nunui kāore e hiahia te kōpaki kaha rawa |
| W8A16 | He taumaha 8-pī, hohenga 16-pī — te taumata pono-rawa e whakahaeretia ana e mātou | E whakamahia ana i te wā he iti noa te tūāpapa nō reira kāore e hiahiatia te kōpaki kaha |
Ka whiwhi a MiMo-V2.5 i te rārangi tuatoru mā tōna kiato: kāore e hiahia tētahi tūāpapa mātotoru o tēnei akomanga kia parea ki te toru pī kia uru, nō reira ka taea e koe te pupuri i te waru, ā, ka noho tata tonu ngā taumaha e tauira ai koe ki ngā mea i whakangungua ai. He nui ake tōna take mō te mahi whakaahua, mō tētahi take hanganga. I roto i tētahi kaiwhakawetewete kuputuhi-anake, ko tētahi whakauru e kotiti ana i raro i te quantization ka whakahoropakitia anō e ngā tekau paparanga aro o muri mai — he maha ngā whāinga hei whakatika. Kāore he mea pēnā mō tētahi tū tirohanga: ka rere te mōhiohio ataata mā roto i te kaiwhakawaehere kotahi me te kaiwhakaputa whāiti kotahi i mua i tōna taenga ki te kaiwhakawetewete, ā, ka whakaritea ngā mea katoa o muri mai e te mea i puta mai i taua kōhī. Ko te tohu menemene — te 3 me te 8 i tētahi mataki-iarere-iti, te pika tika e mutu ai tētahi pae i runga i te aho tūtohi, te rohe o tētahi huānga UI e akiakihia ana koe kia pāwhiri — e noho ana i roto i te taipitopito taumaha rahinga-iti e mukua tuatahitia ana e te quantization kaha.
Kāore tētahi tauira tirohanga 3-pī e hapa ana kia rangona; ka hapa puku noa, ka hanga i tētahi whakaahuatanga rerenga-pai o tētahi tuhinga me ngā mati e rua kua whakawhitia i roto i te tapeke. Ko te taumaha tika ake te inihua utu-iti rawa mō taua hapa, ā, i konei he kore-utu — ka utua e koe te utu papa ahakoa pēhea.
04Te utu papa, i roto i te horopaki
E whā ngā tauira o tā mātou rārangi tekau mā rua e noho ana i te utu papa $0.50 tāuru / $2.00 whakaputa ia miriona tohu. Kotahi anake o rātou e whakaae ana ki ngā whakaahua. Anei te tūnga o ngā tauira whai-tirohanga i te utu whakaputa, te nama e whakahaere ana i te nuinga o ngā nama utu tūturu:
Utu whakaputa i te US$ ia 1M tohu, ngā reiti rārangi a te Kūwaha Shannon AI, Hepetema 2026. He whakarite te roa o te pae ki te tauira tirohanga utu-nui rawa o te rārangi.
Hei whakataurite ki a Kimi-K3 — te tauira tirohanga taumata teitei rawa, ā, he mea kaha tūturu ake — he tata 9.6× te utu-iti ake o MiMo-V2.5 mō te whakaputa, ā, he 7.7× te utu-iti ake mō te tāuru. Ehara tēnā i te kerēme ōritenga; he tautohenga mō te āhua o te kawenga mahi. I te kotahi rau mano whakaahua ia marama, kāore e taea te utu i te tauira taumata teitei rawa, engari ka taea te tauira utu papa, ā, ko te pātai pono ko te whakaeke rānei o te tauira utu papa i tō paepae tika — kaua ko te toa rānei i tētahi paearu.
05MiMo-V2.5 rāua ko MiniMax-M3: ngā tauira tirohanga utu-iti e rua, he whakataurite pono
Ka tautau tika ēnei tauira e rua i te utu, i te horopaki, i te mata pito ara. Ehara rāua i te mea whakawhiti-noa: he tauira te taumata quantization mō ngā tāpaetanga rerekē e rua mō te aha e whai take ai tētahi tauira tirohanga.
| MiMo-V2.5 · W8A16 | MiniMax-M3 · 3BIT-REAP | |
|---|---|---|
| Tāuru / whakaputa ia 1M | $0.50 / $2.00 | $0.50 / $2.00 |
| Te pono taumaha | 8-pī | 3-pī + te tapahi mātanga |
| Te rahi o te tūāpapa | Kiato, mātotoru | Nui, MoE korekore |
| Tirohanga | Āe | Āe |
| Horopaki | 262,144 | 262,144 |
| Schema · taputapu · streaming | Āe | Āe |
Kōwhiria a MiMo-V2.5 ina koinei te wāhanga uaua te whakaahua: te OCR mātotoru i runga i ngā mataki hapa, te tango tau mai i ngā tūtohi me ngā ripanga, te whakaū-ā-kite hopukanga mata me UI, te pānui kuputuhi-iti, te wehewehe i ngā whakaahua hua tata-ōrite — ngā mea katoa e noho ana he koha te mati kua whakawhitia, kaua he hapa āhua noa. Kei te hiahia koe i te ara pono-rawa mai i ngā pika ki ngā tohu, ā, ko ngā taumaha 8-pī te huarahi ka whiwhi ai koe i tēnā i tēnei utu.
Kōwhiria a MiniMax-M3 ina koinei te wāhanga māmā te whakaahua. He tūāpapa nui, korekore, kua kōpakina ki te 3-pī, e mau tonu ana i te mātauranga ao me te whānuitanga whakaaroaro nui ake i tā tētahi tauira kiato, mātotoru, ahakoa i muri i te tapahi. Mēnā ko te mahi he "titiro ki tēnei whakaahua, whakaaroaro roa mō tāna e tohu ana," he whakaahua rānei tētahi tāuru i roto i tētahi kōrerorero kuputuhi-te-nuinga, ko te tūāpapa nui ake te taputapu pai ake — ehara te mātai-ā-kite i tukua atu e koe i te here mau.
06MiMo-V2.5 rāua ko MiMo-V2.5-Pro: te tuakana/teina i hoko i ōna karu mō te hōhonu
Koinei te whakataurite e hanga ana i te nuinga o ngā tiikiti tautoko, nā reira me kī mārama. Ka tiritahi a MiMo-V2.5-Pro-W8A16 rāua ko MiMo-V2.5-W8A16 i te utu, i te matapihi horopaki, i te taumata pono, me te ingoa whānau. Kāore a Pro e whakaae ki ngā whakaahua. Ehara i te huinga-nui-ake. He tohatoha rerekē o te pūtea ōrite.
Ko te whakahaere i tētahi tauira reo-maha te tikanga me utu mō tētahi pou tirohanga me tētahi kaiwhakaputa hei tāpiri ki te kaiwhakawetewete — i tētahi whāinga utu pūmau, he kaha kāore i whakapaua ki te tauira reo ake. Ka whakapaua katoatia e Pro ki te kuputuhi: he tūāpapa nui ake me te wāhi maha ake mō te tuhi roa, te whakaaroaro whānui, me te waehere. Ka whakapaua e MiMo-V2.5 tētahi wāhanga ki te tirohanga.
Nō reira ko te ture whakatau mō ō tāuru, kaua mō ō wawata. Mēnā ka taea e tētahi tono te mau i tētahi hopukanga mata, tētahi whārangi mataki, tētahi tūtohi, tētahi hoahoa, tētahi whakaahua pūhoto rānei, me hiahia koe i a MiMo-V2.5 — ka whakakāhoretia e Pro te poraka ihirangi. Mēnā he kuputuhi ia tono, ā, i toro atu koe ki te tauira tirohanga nā te māhitanga noa, kei te whakarere koe i te kounga tūāpapa mō te kore: tangohia a MiMo-V2.5-Pro i te reiti ōrite.
He tauira e tika ana kia whāngaihia: whakaritea mā te momo ihirangi. Ko ngā mea katoa he wāhanga whakaahua tōna ka haere ki a MiMo-V2.5-W8A16, ko ērā atu katoa ki a MiMo-V2.5-Pro-W8A16. He whānau ōrite, he utu ōrite, he pūtea horopaki ōrite — he ngāwari noa te kaiwhakarite ara, kāore hoki he tauira utu hei whakarite.
07He aha tino te tikanga o te "māramatanga whakaahua kore-ārai"
Kāore he paparanga whakakāhore kei mua i tēnei tauira, kāore hoki he tātari ihirangi kei runga i tāna whakaputa. He whakatau hua tēnā me tētahi hua ake mō te mahi tirohanga: ka whakaahuatia e te tauira te mea kei roto i te papa, kaua e whakakāhoretia i runga i te kaupapa.
He mea nui tēnei nā te mea kāore ngā API tirohanga kua tātaritia e noho hanganga-pai ana ki tētahi katoa akomanga mahi whaimana. Me titiro tētahi awa whakahaere-ihirangi ki te rauemi e whakahaeretia ana e ia, ā, me kī he aha tēnā; ehara tētahi kaiwhakarōpū e whakakāhore ana ki te tirotiro i te mea e whakarōpūtia ana e ia i te kaiwhakarōpū. E mau ana tēnei puta noa i tētahi rārangi roa o ngā mahi tūturu:
- Ngā taputapu whakawhirinaki me te haumaru — te tapa tika i ngā whakaahua kāore tētahi API auau e whakahaere, ā, koirā tonu ngā whakaahua e tino hiahiatia ana kia tapaina.
- Ngā whakaahua rongoā, taunekeneke, aitua hoki — ngā whakaahua haumanu, whara, wāhi aitua e whakakāhoretia ana e ngā tātari whānui hei mea kino ahakoa ko wai e pātai ana.
- Te rangahau haumarutanga — te pānui i ngā whārangi hī-tāware, ngā atanga kino, me ngā mata kua hopukina, kāore te tauira e karo i te take.
- Te whakarōpū ihirangi pakeke me te whakaraupapa pūranga — ngā awa e noho ana hei tautuhi i te rauemi, kāore e taea te tuku atu i te titiro, tae atu ki ngā kohinga hītori e kīia ana e ngā tātari he kore-e-taea-te-tapa.
Ko te tango i te paparanga whakakāhore he tango i tētahi ārai hangarau, kaua he ārai ture. E whakahaerea ana a Shannon AI e Shannon Lab LLC i New Mexico, USA, ā, e whakahaerea ana te whakamahinga o te API e tā mātou Kaupapa Here Whakamahi Haepapa. He āheinga te whakaputa kore-tātari, kaua he pukapuka whakaae.
08Te karanga i a ia: ngā pito ara, te tāuru whakaahua, me tōna wāhi e uru ai
Ko te tauira he MiMo-V2.5-W8A16, ka tae atu mā ngā reo tono e toru e pā ana ki ngā taumaha ōrite, nō reira ka pupuri tonu koe i tēhea SDK kua whai kē koe:
/v1/chat/completions— rite ki OpenAI. Ka uru ngā whakaahua hei wāhanga ihirangiimage_url, hei URL https, hei URI raraunga base64 rānei./v1/messages— rite ki Anthropic. Ka uru ngā whakaahua hei poraka ihirangiimage./v1/responses— te reo momo-Responses.
Ka mahi te streaming me te karanga taputapu i ngā mea e toru, nō reira ka taea e te tauira te titiro ki tētahi hopukanga mata, ka karanga i tētahi mahinga mō tāna i kite ai i te hurihanga kotahi. Kei te tuhinga API ngā schema tono me te whakautu katoa, ā, kei reira hoki tētahi papa tākaro ka taea e koe te whakapiri whakaahua ki roto i mua i te tuhi waehere.
Ka tautoko a MiMo-V2.5 i te json_schema me te response_format pūmau — te rerekētanga i waenga i tētahi tauira-whakaatu me tētahi awa mahi. Hoatu he rīti me tētahi schema he māra kua tāritarihia mō te kaihoko, te rā, ngā tūemi, me te tapeke, ā, ka riro mai he ahanoa ka taea e tō waehere te whakamana, kaua he parāweta me regex e koe. E rua o ngā tauira tekau mā rua o tēnei rārangi kāore e tautoko katoa ana i te putanga hanganga; ehara tēnei, ā, koinā te take ka mahi ai hei ratonga tango-raraunga nui-rawa.
He 262,144 tohu te matapihi horopaki; kua tapatapahia te whakaputa hangaia ki te 16,384 tohu ia tono. Ka pau ngā tohu tāuru e ngā whakaahua e ai ki te taumata whakaahua, nō reira ka kai ngā mataki whārangi taumata-teitei i te horopaki tere ake i tētahi tatauranga kupu ōrite o te kuputuhi. Whārangitia ngā tuhinga roroa kaua e uta i tētahi mataki 300-whārangi i te tono kotahi, ā, whakaarohia te tapa whakaputa hei here tūturu mō te tango-raraunga whakapau-katoa — wehewehea ngā ripanga mātotoru mā te awhe rārangi kaua ka riro he whakautu kua tapahia.
He mana tūturu tō te pae reasoning_effort i konei, e ai ki te whakapapa o te whānau. Whakahekea mō te whakarōpū nui-rawa, i reira he tapa te whakautu ā, he utu pū noa ngā tohu whakaaroaro; whakapikitia mō te whakamāori tūtohi, te pāngarau-ā-kite, me te whakaaroaro hoahoa, i reira me whakakotahi te tauira i ētahi mea maha i kitea e ia i mua i tāna whakautu.
Tōna wāhi e uru ai, me tōna wāhi kāore e uru ai
Ngā mea e uru pai ana. Te tango-raraunga tuhinga me te rīti nui-rawa; te mahi tata-ki-te-OCR i runga i ngā mataki hapa; te pānui tūtohi, papatohu, ripanga hoki; te māramatanga hopukanga mata me te whakaū-ā-kite UI; te tapa rārangi me te QA ā-kite; ngā awa whakahaere-ihirangi — ngā mahi reo-maha katoa e here ana te utu ia-karanga, ā, ko te pono taipitopito te mea kāore koe e whakaae kia tukuna.
Ngā mea kāore e uru pai ana. Te whakaaroaro reo-maha uaua-rawa e hiahia ana i te tauira kaha rawa ahakoa te utu — koirā te mahi a Kimi-K3, ā, he tata tekau whakarea tōna utu mō tētahi take. Te whakaputa kuputuhi-anake, i reira ka hoatu e MiMo-V2.5-Pro tētahi tūāpapa pai ake i te reiti ōrite. Te mahi e whakawhirinaki ana ki te mātauranga ao ihirangi-nui, i reira ka toa tētahi tūāpapa nui, korekore, i tētahi tauira kiato, mātotoru. Ā, ngā mea katoa e hiahia ana i te horopaki nui ake i te 262K, kāore he tauira i konei e whakaratoa ana i tēnā.
Ko ia tau kua whakaputaina i runga ake nei nā Xiaomi ake, mō tētahi tohu-oti kua whakaingoatia i te raupapa MiMo me MiMo-VL. Pānuitia te rauemi taketake, kaua tā mātou whakarāpopototanga.
09Ngā pātai auau
He aha te API tirohanga MiMo kore-ārai?
Ko te tauira MiMo-V2.5 W8A16 i runga i te Kūwaha Shannon AI: he tauira whai-tirohanga i te whakapapa MiMo a Xiaomi, e whakahaerea ana i runga i tā mātou ake tautau GPU i te pono taumaha 8-pī me te horopaki 262,144-tohu, kāore he paparanga whakakāhore, kāore hoki he tātari whakaputa. Ka riro i a ia ngā whakaahua me te kuputuhi mā /v1/chat/completions, /v1/messages me /v1/responses i te $0.50 ia miriona tohu tāuru me te $2.00 ia miriona tohu whakaputa.
Ko tēhea te tauira kore-ārai utu-iti rawa e tino kite ana i ngā whakaahua?
E rua ngā tauira e tautau ana i te utu papa $0.50 tāuru / $2.00 whakaputa ia miriona tohu: MiMo-V2.5 W8A16 me MiniMax-M3 3BIT-REAP. Ko MiMo-V2.5 te mea pono-rawa o ngā mea e rua i ngā taumaha 8-pī, te taunoa haumaru rawa mō te mahi whakaahua e whai wāhi ana ki te taipitopito. Ka whakahaere a MiniMax-M3 i tētahi tūāpapa nui ake i kōpakina ki te 3-pī, e whai ana i te whānuitanga o te mātauranga ao i tua atu i te pono ia-pika.
He aha te tikanga o W8A16, ā he aha ai e nui ana mō te tirohanga?
Ko te tikanga o W8A16 he taumaha 8-pī me ngā hohenga 16-pī, te taumata pono-rawa o te rārangi Shannon, e whakamahia ana i te wā he iti noa te tūāpapa nō reira kāore e hiahiatia te kōpaki kaha. He mea nui mō te tirohanga nā te mea ka kawea e tētahi pou tirohanga me tāna kaiwhakaputa te taipitopito menemene mā tētahi ara whāiti, ā, ko ngā hapa i tāpirihia i reira kāore e taea te whakatika i muri mai, kaua i te āhua o tētahi tāuru kuputuhi āhua pōuriuri e taea tonutia te whakatika.
He aha te rerekētanga o MiMo-V2.5 i a MiMo-V2.5-Pro?
He utu ōrite, he horopaki 262,144-tohu ōrite, he taumata W8A16 ōrite, engari ehara i te tauira ōrite. Ka whakaae a MiMo-V2.5 ki ngā whakaahua; kāore a MiMo-V2.5-Pro, e whakapau ana i taua pūtea whakahaere ki tētahi tūāpapa kuputuhi nui ake. Whakamahia a MiMo-V2.5 ina uru ana ki ngā tāuru he hopukanga mata, he tuhinga, he tūtohi, he whakaahua pūhoto rānei; whakamahia a Pro mō te mahi kuputuhi-anake i te reiti ōrite.
Ka tautoko te tauira tirohanga MiMo kore-ārai i te schema JSON me te karanga taputapu?
Āe. Ka tautoko a MiMo-V2.5 W8A16 i te putanga hanganga json_schema pūmau, te response_format, te karanga taputapu, te streaming me tētahi mana whakahaere reasoning_effort. Koirā te mea e whai take ai hei tuara tango-raraunga: hoatu he whakaahua ki a ia, ka riro mai he ahanoa kua whakamanahia, kaua he tuhinga me wetewete e koe.
Tukua he whakaahua ki a ia
Tirohanga 8-pī, horopaki 262K, kore paparanga whakakāhore — i te utu o tētahi tauira kuputuhi-anake.
Pānuitia ngā Tuhinga API Tirohia Ngā Tauira 12 KatoaId tauira: MiMo-V2.5-W8A16 · e whakahaerea ana e tā mātou Kaupapa Here Whakamahi Haepapa
Puna kōrero: Pūrongo hangarau MiMo (arXiv 2505.07608) · Pūrongo Hangarau MiMo-VL (arXiv 2506.03569) · XiaomiMiMo/MiMo-VL · MiMo-VL-7B-RL-2508. He tau nā Xiaomi i pūrongotia ngā tatauranga paearu katoa mō ngā tohu-oti kua whakaingoatia i runga ake nei, ā, e whakaahua ana i te whakapapa whānau, kaua i te hanga tika kei tēnei pito ara. He uara rārangi nā te Kūwaha Shannon AI ngā utu, te horopaki, me ngā haki āheinga, Hepetema 2026.