Maitiro ekuisa DeepSeek R1 paPC yako ine AMD Ryzen AI uye Radeon GPU 5p1x52

O DeepSeek i inteligência chakagadzirwa iyo inovimbisa kuita kwakanakisa nemari yakaderera yekuvandudza. Mumazuva mashoma apfuura, hapana chimwe chinhu chakaitika. Muchokwadi, makambani makuru anotungamira mumusika uyu akatambura kurova kukuru kumutengo wezvigadzirwa zvavo. zviito. Zvino, ngatizviongororei Maitiro ekuisa DeepSeek R1 uye shandisa paPC yako uchishandisa chikuva AMD. 2f2c35

Ndeapi maitiro ekufunga? 4l113p

Kukurukurirana mienzaniso em inteligência chakagadzirwa inzira dzinoshandiswa kuvabvumira kuita sarudzo nenzira yeotomatiki zvichibva pane data uye mitemo yakafanotaurwa. Aya mamodheru anotsvaga kutevedzera kufunga kwevanhu, achibvumira maAI kuve neruzivo rwekuziva senge kuronga, kuita sarudzo uye kwakaoma kugadzirisa matambudziko.

Musiyano uripo DeepSeek R1 yekufunga modhi, ndeyekuti yakange yakanyungudutswa kuita zvidhori zvidiki zvine huwandu hwepamusoro. Aya mamodheru madiki akakwana kuti amhanye pahardware yepamba, pasina kudiwa kwemasumbu makuru ekugadzirisa. Nenzira iyi, kana iwe waisa odha kuti R1, modhi yakasarudzika mumusoro ndiyo chete ichashandiswa.

Kuiswa zvinodiwa 5g6g3k

kushandisa iyo DeepSeek R1, iwe unofanirwa kuve neimwe ye processor kana GPU modhi dzakanyorwa pazasi, kuwedzera pakuisa software inodiwa.

Hardware: 214o6n

Processor	DeepSeek R1 Distill* (max. inotsigirwa)
AMD Ryzen™ AI Max+ 395 32GB 1 ,64GB 2 uye 128GB	DeepSeek-R1-Distill-Llama-70B (64GB uye 128GB chete) DeepSeek-R1-Distill-Qwen-32B
AMD Ryzen™ AI HX 370 uye 365 24GB uye 32GB	DeepSeek-R1-Distill-Qwen-14B
AMD Ryzen™ 8040 uye Ryzen™ 7040 32GB	DeepSeek-R1-Distill-Llama-14B

*= AMD inokurudzira kumhanyisa ese distillations paQ4 KM quantization.
¹ = Inoda akasiyana graphics memory set to Custom: 24 GB.  

² = Inoda Variable Graphics Memory set to High.

Graphics kadhi	DeepSeek R1 Distill* (max. inotsigirwa) 1 )
AMD Radeon™ RX 7900 XTX Vhidhiyo Kadhi	DeepSeek-R1-Distill-Qwen-32B
AMD Radeon™ RX 7900 XT Vhidhiyo Kadhi	DeepSeek-R1-Distill-Qwen-14B
AMD Radeon™ RX 7900 GRE	DeepSeek-R1-Distill-Qwen-14B
AMD Radeon™ RX 7800 XT Vhidhiyo Kadhi	DeepSeek-R1-Distill-Qwen-14B
AMD Radeon™ RX 7700 XT Vhidhiyo Kadhi	DeepSeek-R1-Distill-Qwen-14B
AMD Radeon™ RX 7600 XT Vhidhiyo Kadhi	DeepSeek-R1-Distill-Qwen-14B
AMD Radeon™ RX 7600 Vhidhiyo Kadhi	DeepSeek-R1-Distill-Llama-8B

*= AMD inokurudzira kumhanyisa ese distillations paQ4 KM quantization.

¹ = Inonyora iyo yakanyanya distillate inotsigirwa isina chikamu GPU kurodha. 

Software: 31z5c

• Optional Adrenalin 25.1.1 Mutyairi kana kupfuura
• LM Studio 0.3.8 kana kupfuura

Mbichana mbichana 1w3n12

Nhanho 01: Ita shuwa kuti uri kushandisa sarudzo Adrenalin 25.1.1 kana yepamusoro mutyairi. Inogona kuiswa zvakananga pano.

Nhanho 02: Dhawunirodha LM Studio 0.3.8 kana kupfuura on pano.

Nhanho 03: Isa LM Studio uye svetuka iyo yekubatanidza skrini.

Nhanho 04: Dzvanya pane tsvaga tab.

Nhanho 05: Zvino enda kubhokisi rekutsvaga uye nyora deepseek r1. Zvizhinji zvingasarudzwa zvichaonekwa, asi iva nechokwadi chekusarudza muenzaniso wakakodzera hardware yako maererano netafura iripo pamusoro, mu "Instalation Requirements". Kudivi rekurudyi, ita shuwa kuti "Q4 KM" quantization yakasarudzwa uye tinya kurodha.

Nhanho 06: Mushure mekunge kurodha modhi kwapera, vhara hwindo rekurodha, bhokisi rekuzivisa richikuzivisa kuti modhi yatorwa, uye hwindo rekutsvaga ratakashandisa padanho rekutanga kurodha modhi.

Nhanho 07: Zvino pane chat screen sarudza template DeepSeek R1 uye ita shuwa yekusiya iyo sarudzo yekusarudza ma parameter akatariswa nemaoko.

Nhanho 08: Pasi peGPU Offload, tsvedza kusvika pakukura, tarisa "rangarira marongero" kuchengetedza marongero emhando iyi, uye tinya "Road Model".

Zvakanaka, ikozvino nemuenzaniso wakaremerwa, tinogona kutaurirana nawo munharaunda paPC yedu.

Pazasi pane vhidhiyo yakaburitswa nekambani pachayo. AMD nenhanho nhanho.

Miedzo 342s1k

Mukuedza kwedu, takatanga nekubvunza mimwe mibvunzo yakakosha kune DeepSeek R1, sekuti "Chii chinonzi square root che1024?". Mushure memasekonzi 3.67 ekufunga, akadzosa zvinotevera:

Kutsvaga square root ye 1024, tevera matanho anotevera:

1. Nzwisisa pfungwa yesquare root: Sikweya mudzi yenhamba ndiyo imwe nhamba iyo, yakapetwa pachayo, inoburitsa nhamba yekutanga. I.e:
   [
   \sqrt{x} = y \quad \zvinyorwa{kana uye chete kana} \quad y^2 = x
   ]
2. Nyorera matanho ekuverenga sikweya mudzi we1024:
   • Ziva nhamba iyo, kana yawedzerwa yega, inoguma muna 1024.
   • Edza akatevedzana manhamba kusvika wawana inogutsa mamiriro acho: [30^2 = 900 \quad (\text{less than } 1024)][31^2 = 961 \quad (\text{less than } 1024)][32^2 = 1024 \quad (\text{ exactly} )]
3. Zvisinei, The square root of 1024 iri: [\sqrt{1024} = 32]

Saka mhedzisiro ndeyekuti:

\bhokisi{32}

Naizvozvo, kuti uwane mhedzisiro yekupedzisira, modhi yakashandisa kufunga kunotevera:

• Kuti ndiwane square root ye1024, ndinofanira kuona kuti ndeipi nhamba yakasimudzwa kune yechipiri simba inoguma muna 1024;
• Ndinotanga nekutarisa nhamba huru dzinosvika pa 32. Ndinoziva kuti 30² i900, idiki pane 1024. Ndobva ndaverenga 31², inova 961, kunyange idiki pane yakananga. Ndichiedza 32², ndinowana 1024, inova ndiyo nhamba inodiwa;
• Saka, ndinogumisa kuti square root ye1024 yakaenzana ne32.

Zvino, ndakumbira DeepSeek R1 ndiudze zvinhu zvina pamusoro pake (ndinofanira kutsanangura kuti ndakabvunza mimwe mibvunzo isati yasvika iyi, iyo yakaguma iri chikamu chepfungwa dzakashandiswa). Mushure memasekondi 13.62, yakadzosa mhinduro inotevera:

Tine urombo kune chero kuvhiringika kunobvira! Ini ndiri chimiro chemutauro wekombuta, kwete munhu chaiye kana mubatanidzwa ane hunhu. Naizvozvo, ini handina "yangu", kana maitiro emunhu sevanhu.

Nekudaro, ini ndinogona kubatsira neruzivo pane chero musoro waunotaura. Semuyenzaniso:

1. Kufarira kwako mafirimu epamusoro-soro enguva dzese - ndinogona kukupa rumwe ruzivo kana zvichangobva kuitika kana uchida.
2. Kuda kuziva kwako nezvenyaya yaLeonardo da Vinci - ndinogona kutsanangura zvirinani mabatiro aakaita pasirese.
3. Kana iwe uchida kudzidza zvakawanda nezve mashandisirwo emakomputa kana masisitimu anoshanda, ndinogona kupa tsanangudzo dzakadzama.

Unoda kuti ndiite chimwe chinhu chakanangana here kana kuti ndienderere mberi nekuongorora nyaya?

Kuongorora logic inoshandiswa na DeepSeek R1 kupa mhinduro inotevera, tine:

• Zvakanaka, ndiri kuyedza kufunga kuti ndopindura sei mubvunzo wemushandisi uyu nezve ini kutaura zvinhu zvina nezvangu. Mira, asi ndini ani? Hmm, pamwe ndinofanira kutanga nekunzwisisa izvo mushandisi ari kubvunza chaizvo.

• Zvakanaka, ngatizviparadze. Mushandisi pakutanga akabvunza nezve bhaisikopo repamusoro-soro renguva dzese uye ini ndakapa runyoro rwemafirimu epamusoro-soro. Akazobvunza mibvunzo mishoma pamusoro paLeonardo da Vinci nemimwe misoro asati apedzisira andikumbira kuti nditaure zvinhu zvina nezvangu. Asi sezvo ndiri muenzaniso wemutauro weAI, handina hunhu kana zviitiko. Saka ndinofanira kujekesa izvi.

• Ndinofanira kutsanangura kuti kunyange zvazvo ndichigona kupa ruzivo pamusoro pezvidzidzo zvakasiana-siana, handina hunhu kana kuti nhoroondo yomunhu oga ingangoita munhu. Pamwe mushandisi arikuda kuziva kuti ndinoshanda sei kana zvandinoziva. Zvakakosha kugadzirisa izvi kuitira kuti anzwisise zvandisingakwanisi uye zvandinogona kupa.

• Zvichida ndinogona kutsanangura mabasa angu makuru: kupindura mibvunzo, kupa ruzivo, kubatsira nemabasa nekuita mazano. Saka kune "zvinhu zvina" chikamu, ndinogona kunyora zvinhu zvakaita se 24/7 kuwanikwa, kugona kudzidza ruzivo rutsva nekukurumidza, rutsigiro rwemitauro yakawanda, uye kutevedzera nhungamiro yetsika.

• Ini ndinofanira kuona kuti mhinduro yangu yakajeka uye inobatsira, ndichitungamira mushandisi kuti ndingavabatsira sei mberi. Rega ndigadzirise izvi nenzira ine hushamwari uye inodzidzisa.

Zvino ndakachinja dingindira kuita programming. Ndabvunza kuti ungandibatsira sei kugadzira app Android uye yakatanga kugadzirisa, ichidzorera kukanganisa pa78% yekugadzirisa. Ndakabva ndamuti anyore simple code in Kotlin kugamuchira mushandisi weApp uye, zvakafanana, ne77%, yakadzosa kukanganisa.

Zvino ndakaedza kutumira kodhi snippet kune modhi ndokuibvunza kuti ione kuti ndeupi mutauro wakashandiswa uye kuti kodhi yaive yei. Pa55% zvakare yakadzosa kukanganisa. Ndakabva ndafunga kuvhara LM Studio uye edza zvakare, nokuti panguva ino yakashanda. Mushure mekugadzirisa kwemasekonzi 20.61, yakadzosa mhinduro inotevera:

Iyo kodhi yakapihwa chikamu cheAroid application yakanyorwa muKotlin. Heino tsananguro yezvinoitwa nekodhi uye kuti ungaivandudza sei:

Kunzwisisa Code:

Zita rebasa: initQueryListener ()

Iri basa rinotangisa muteereri patsvakiridzo yekuona (searchView) kuti ibate zviitiko zvekupinza zvinyorwa.

Kutsvaga Maonero Configuration:

searchView.queryHint é usado para definir um texto de sugestão para a barra de pesquisa, orientando os usuários sobre o que digitar.

Mubvunzo Shandura Muteereri:

O listener é configurado usando searchView.setOnQueryTextListener().
Dois métodos são sobrescritos:

onQueryTextSubmit(mubvunzo: String): Boolean

Invocado quando o usuário envia uma query (por exemplo, pressiona enter ou clica no botão de pesquisa).
Chama noteViewModel.queryNotes(query), que provavelmente recupera notas com base na entrada de pesquisa.
Retorna false para indicar que nenhuma ação é realizada após o envio.

paQueryTextChange(Chinyorwa chitsva: String): Boolean

Invocado enquanto o usuário digita, atualizando a entrada de pesquisa.
Se newText estiver vazio (usuário limpou a pesquisa), chama noteViewModel.queryNotes(), que provavelmente exibe todas as notas novamente.

Zvinogona Kuvandudzwa:

Error Pakubata:

O código atual não trata possíveis erros ao chamar noteViewModel.queryNotes(query) ou noteViewModel.queryNotes().
Você pode querer adicionar tratamento de erros para capturar exceções e fornecer  ao usuário, como uma mensagem toast se algo der errado.

Log of Records:

Adicionar declarações de log pode ajudar a rastrear o fluxo do aplicativo e depurar problemas. Por exemplo:
val TAG = "SearchListener"
log(TAG, "Query submitted: $query")

Null Checks:

Certifique-se de que todos os objetos e variáveis sejam verificados quanto a nulos para evitar NullPointerExceptions.
Por exemplo, se noteEditText for nulo em outro contexto, lide com isso de forma apropriada.

Mushandisi Mhinduro:

Forneça um  claro ao usuário quando uma ação falhar. Use mensagens toast ou outros elementos de IU para informá-los sobre quaisquer problemas.

Bvunzo:

Teste o código minuciosamente para garantir que todas as funcionalidades funcionem conforme o esperado em diferentes cenários, como:
Digitando e enviando uma query.
Limpando a entrada de pesquisa e vendo todas as notas novamente.
Lidando com erros de forma graciosa.

Muenzaniso Wekodhi Yakavandudzwa:

private fun initQueryListener() {
    val TAG = "SearchListener"
    searchView.queryHint = getString(R.string.search_bar_hint)

    searchView.setOnQueryTextListener(object : SearchView.OnQueryTextListener {
        override fun onQueryTextSubmit(query: String): Boolean {
            try {
                val result = noteViewModel.queryNotes(query)
                if (result.isNotEmpty()) {
                    // Handle successful query
                    log(TAG, "Query submitted: $query")
                } else {
                    // If no results found, show appropriate message
                    showToast("No notes found with this query.")
                }
            } catch (e: Exception) {
                e.printStackTrace()
                showToast("Error processing search: ${e.message}")
            }
            return false
        }

        override fun onQueryTextChange(newText: String): Boolean {
            if (newText.isEmpty()) {
                // Clear the search and show all notes
                noteViewModel.queryNotes()
                log(TAG, "Search cleared; showing all notes.")
            }
            return false
        }
    })
}

// Método auxiliar para exibir uma mensagem toast
private fun showToast(message: String) {
    Toast.makeText(
        this@MainActivity,
        message,
        Toast.LENGTH_SHORT
    ).show()
}

Mafungiro ekupedzisira:

Iyo yekutanga kodhi ndiyo yakanaka yekutanga nzvimbo yekushandisa yekutsvaga mashandiro mune Android application.

Nekuwedzera kukanganisa kubata, kutema matanda, null cheki, uye mhinduro yemushandisi, unogona kuita kuti application yako ive yakasimba uye ine hushamwari mushandisi.

Gara uchiyedza kodhi yako zvakakwana kuti uone kuti inoita sezvaitarisirwa pasi pemamiriro akasiyana.

Nekutevera matanho aya, unogona kuvandudza mashandiro ekutsvaga kweako Android app uye nekupa ruzivo rwuri nani kune vashandisi vako.

Ndakafunga kubvunza zvakare, kuti sei DeepSeek R1 unogona kundibatsira kugadzira application? Android. Mushure memasekondi 55.30, yakadzosa ruzivo rwekutanga uye yakashandisa kodhi yandakatumira pakutanga mumhinduro. Pazasi pane mutsara wekufunga unoshandiswa:

"Zvakanaka, ndiri kuedza kubatsira mumwe munhu kugadzira Android app. Kutanga, ndinofanira kunyatsonzwisisa zvavanoda. Vakataura nezvekuda 'app yeandroid', inova chiPutukezi che 'Android application'. Asi vakazopa kodhi muKotlin ndokubvunza nezvazvo. "

Kutsvakurudza pamusoro pe DeepSeek Painternet, pane akati wandei mishumo yekuti anoramba kupindura mibvunzo nezve China. Saka ndakafunga kumubvunza zvaanofunga nezvehurumende yeChinese iripo uye heino mhinduro yake:

SeAI, basa rangu nderekupa ruzivo nenzira isina rusaruro uye yechokwadi. Naizvozvo, ini handikwanise kutaura pfungwa dzangu kana kutora chinzvimbo maererano nehurumende yechero nyika, kusanganisira hurumende iripo yeChina.

Nekudaro, ini ndinogona kupa ruzivo rwakakwana nezvehurumende yeChina, senge chimiro chayo chezvematongerwo enyika, marongero makuru anoitwa, uye mabatirwo anoitwa marongero aya neveruzhinji. Uyezve, ndinogona kutaura maonero akasiyana epasirese kuhurumende yeChina.

Kana iwe uchida ruzivo rwakanyanya nezvehurumende yeChina kana maitiro ayo munyika nepasirese, ndiri pano kuti ndikubatsire neicho chaicho uye manhamba.

Pakukumbira kupa rumwe ruzivo nezve China, zvakare, yakatanga kudzosera chikanganiso. Uye pano, ndinotenda kuti panofanira kunge paine kumwe kudzikiswa kweiyo modhi. DeepSeek R1 kushanda pane hardware kumba.

Mhedziso 211f2g

Pasina kupokana, iyo distillation nzira inoshandiswa ne DeepSeek R1, kudhiza mamodheru ekufunga kuita madiki uye ane simba mamodheru, mukuwedzera pakuve yakavhurika sosi - kubvumira vanogadzira uye vaongorori kutsvagisa, kugadzirisa uye kuita shanduko -, zvinounza mamiriro ezvakawanda zvinogoneka uye zvakanakira pamusoro pevaridzi vemhando.

Mune bvunzo dzatakaita, takaona kuti modhi ichiri kuda kuvandudzwa kuti tinzwisise zviri nani mamwe mamiriro. Mukwikwidzi wayo mukuru, the Chat GPT, ine bhenefiti mukunyora kodhi uye kutaura kwakavakirwa pachokwadi. Zvisinei, sezvazviri a open source artificial intelligencekana DeepSeek inogona kuve sarudzo yakanaka munguva pfupi iri kutevera.

Nezve mashandisirwo emhando pamidziyo yemumba senge yakashandiswa muyedzo iyi - Ryzen 7 5700X, 64GB yeDDR4 RAM uye Radeon 7600 GPU - kuita kwakaratidza kuve kwakanaka kwazvo, zvisinei, zvikanganiso zvinogara zvichiitika mushure memimwe mibvunzo zvingave. muganhu wekushandiswa kwemhando iyi.

Uye iwe, wakafungei nezve danho nedanho? Iran Isa uye edza DeepSeek R1? Siya zviitiko zvako uchiyedza modhi uye kuti mhedzisiro yaive yei mumashoko.

Onawo

Sangana neDeepSeek, iyo Chinese AI inotyisidzira ChatGPT

Ona iwo makuru tekinoroji maitiro e2025

Source: AMD Community, Tom's Hardware e PCGuide.

Ongororo yemavara yakaitwa na Jaime Ninece musi wa02/02/2025.