Laola mohlala oa LLM le tokenizer | B4N1

Laola mohlala oa LLM le tokenizer

6 months ago · b4n1 Tiempo de lectura: 6 min

Ho kenya tšebetsong lipatlisiso tsa liketsahalo tse ikemetseng ka li-DevOps Agents le LLMs

Selelekela

Sebakeng sa kajeno sa dijithale se potlakileng, lihlopha tsa DevOps li tobane le phephetso e ntseng e eketseha: ho tsebahatsa le ho rarolla liketsahalo kapele kamoo ho ka khonehang. Mekhoa ea ho etsa lipatlisiso ka letsoho e ka nka nako mme ea sekamela liphosong tsa batho, e lebisang ho theohe nako e telele le ho fokotsa khotsofalo ea bareki. Ho rarolla bothata bona, re ntse re hlahloba mateano a matekoane a DevOps, lipatlisiso tsa liketsahalo tse tsamaisoang ke AI, le karolo ea Large Language Models (LLMs) ho ntlafatsa lisebelisoa tsa DevOps tse kang AWS DevOps Agent le Copado Agentia.

Patlisiso ya Hona Jwale ya Boemo ba Diketsahalo

Patlisiso ea liketsahalo ke karolo ea bohlokoa ea DevOps, e kenyelletsang boitsebahatso, tlhahlobo, le tharollo ea litaba tse amang lits'ebetso tsa software. Mekhoa ea setso e itšetlehile ka mekhoa ea matsoho, e kang tlhahlobo ea log, ho latela liphoso, le puisano le ba amehang. Le ha mekhoa ena e sebetsa hantle, e ka lieha le ho ba le liphoso tse lebisang ho:

  1. **Nako e eketsehileng ea nako **: Mekhoa ea ho etsa lipatlisiso ka letsoho e ka liehisa ho rarolla ketsahalo, e leng ho bakang ho theoha nako e telele le ho fokotsa khotsofalo ea bareki.
  2. Phoso ea botho: Mekhoa ea matsoho e ka ba le liphoso tsa batho, e leng se ka lebisang ho hlahlojoeng ho fosahetseng le ho rarolla liketsahalo.
  3. Mathata a Scalability: Ha boholo le ho rarahana ha litsamaiso li ntse li hola, mekhoa ea ho etsa lipatlisiso ka letsoho e ba thata le ho feta ho sekala.

Seabo sa Baemeli ba DevOps Patlisisong ea Diketsahalo

Baemeli ba DevOps, joalo ka AWS DevOps Agent le Copado Agentia, ba bapala karolo ea bohlokoa ho iketsetsa lipatlisiso tsa liketsahalo. Baemeli bana ba ka:

  1. **Bokella le ho sekaseka lintlha **: Baemeli ba DevOps ba ka bokella le ho sekaseka lintlha mehloling e fapaneng, ba fana ka lintlha tsa bohlokoa mabapi le lisosa tsa liketsahalo.
  2. Ho latela liphoso ka boits'oaro: Baemeli ba ka iketsetsa ho latela liphoso, ba fokotsa nako le boiteko bo hlokahalang bakeng sa lipatlisiso tsa letsoho.
  3. ** Ntlafatsa puisano**: Baemeli ba ka tsamaisa puisano lipakeng tsa lihlopha, bankakarolo le litsamaiso, ho netefatsa hore litaba tsa liketsahalo li arolelanoa ka katleho.

Seabo sa LLM Patlisisong ea Diketsahalo

Mehlala e Meholo ea Lipuo (LLMs) e fetotse tšebetso ea puo ea tlhaho (NLP), e nolofalletsang mechini ho utloisisa le ho hlahisa mongolo o kang oa motho. Boemong ba lipatlisiso tsa liketsahalo, LLMs e ka:

  1. Hlahloba litlaleho tsa diketsahalo: LLMs e ka sekaseka litlaleho tsa diketsahalo, ho supa lintlha tsa bohlokoa le mekhoa e ka supang sesosa sa ketsahalo.
  2. Hlahisa kakaretso ea liketsahalo: LLMs li ka hlahisa kakaretso e khuts'oane ea litlaleho tsa liketsahalo, tsa thusa puisano le ho etsa liqeto.
  3. Fana ka litlhahiso tse hlokomelang maemo: LLMs e ka fana ka litlhahiso tse hlokomelang maemo bakeng sa tharollo ea liketsahalo, ho nahanoa ka tsamaiso e itseng, tikoloho, le ba amehang ba amehang.

Ho kenya tšebetsong lipatlisiso tsa liketsahalo tse ikemetseng ka li-DevOps Agents le LLMs

Ho kenya tšebetsong lipatlisiso tse itirisang tsa liketsahalo le baemeli ba DevOps le LLM, re tla latela mohato ka mohato:

  1. Ho kopanya baemeli ba DevOps: Kopanya baemeli ba DevOps, ba kang AWS DevOps Agent le Copado Agentia, le tsamaiso ea hau ea liketsahalo.
  2. **Bokella le ho sekaseka lintlha **: Lokisa baemeli ba DevOps ho bokella le ho sekaseka lintlha ho tsoa mehloling e fapaneng, ho fana ka lintlha tsa bohlokoa mabapi le lisosa tsa liketsahalo.
  3. Koetlisa mofuta oa LLM: Koetlisa mofuta oa LLM ka pokello ea lintlha tsa litlaleho tsa liketsahalo, litlaleho, le lintlha tse ling tse amehang.
  4. Ho kopanya LLM le baemeli ba DevOps: Ho kopanya mofuta oa LLM le baemeli ba DevOps, ho nolofalletsa LLM ho sekaseka litlaleho tsa liketsahalo le ho fana ka litlhahiso bakeng sa tharollo ea diketsahalo.
  5. Fana ka tharollo: Sebelisa tharollo ea boithuto ba liketsahalo tse itirisang, ho etsa bonnete ba hore e ea senyeha, e ka tšeptjoa ebile e sireletsehile.

Mohlala oa Khoutu: Ho kopanya LLM le Baemeli ba DevOps

Mona ke mohlala oa snippet ea khoutu ho Python, e bonts'ang mokhoa oa ho kopanya mofuta oa LLM le moemeli oa DevOps:

ho rema lifate kantle ho naha
ho tsoa ho li-transformer li kenya AutoModelForSequenceClassification, AutoTokenizer
ho tsoa ho Aws_devops_agent import Agent


mohlala = AutoModelForSequenceClassification.from_pretrained("distilbert-base-uncased")
tokenizer = AutoTokenizer.from_pretrained("distilbert-base-uncased")

# Qala moemeli oa DevOps
moemeli = Moemeli()

# Hlalosa tšebetso ea ho sekaseka litlaleho tsa liketsahalo u sebelisa mohlala oa LLM
def analyze_incident_report(tlaleho):
    # Tsoela pele ho tlaleho ea ketsahalo
    inputs = tokenizer.encode_plus(tlaleho, 
                                    add_special_tokens=Ke 'nete, 
                                    max_length=512, 
                                    return_attention_mask=Ke nnete, 
                                    return_tensors='pt')
    
    # Sekaseka tlaleho ea ketsahalo u sebelisa mohlala oa LLM
    dipoelo = moetso(dintho['input_ids'], attention_mask=inputs['attention_mask'])
    
    # Fumana lengolo la sehlopha se boletsoeng esale pele
    predicted_class = torch.argmax(outputs.logits)
    
    kgutlisetsa_sehlopha_se boletsoeng esale pele

# Hlalosa ts'ebetso ho fana ka litlhahiso bakeng sa tharollo ea liketsahalo
def provide_suggestions(predicted_class):
    # Sebelisa lengolo la sehlopha se boletsoeng esale pele ho fana ka litlhahiso bakeng sa tharollo ea liketsahalo
    haeba ho lebeletsoe_sehlopha == 0:
        khutla "Sheba li-logs tsa sistimi bakeng sa liphoso"
    elif predicted_class == 1:
        khutlisa "Netefatsa khokahano ea marang-rang"
    tse ling:
        khutlisa "Ikopanye le sehlopha sa ntlafatso bakeng sa thuso"

# Kopanya mofuta oa LLM le moemeli oa DevOps
def incident_investigation():
    # Bokella le ho sekaseka lintlha tse tsoang ho moemeli oa DevOps
    logs = agent.get_logs()
    
    # Sekaseka mapolanka u sebelisa mohlala oa LLM
    predicted_class = analyze_incident_report(logs)
    
    # Fana ka litlhahiso tsa ho rarolla liketsahalo
    litlhahiso = fana_litlhahiso (predicted_class)
    
    khutlisetsa litlhahiso

# Fana ka tharollo ea lipatlisiso tsa ketsahalo
ketsahalo_patlisiso()

Kaholimo ho 5-10 Litaelo/Lipotlana tse Sebelisitsoeng ka ho Fetisisa

Mona ke tse 5-10 tse sebelisoang ka ho fetesisa litaelo/lipoleloana tsa theknoloji ena:

  1. aws devops agent get-logs: Bokella lintlha ho tsoa ho moemeli oa DevOps.
  2. transformers.AutoModelForSequenceClassification.from_pretrained(): Laola mofuta oa LLM.
  3. tokenizer.encode_plus(): Batla tlaleho ea ketsahalo esale pele.
  4. mohlala(litlhaloso['input_ids'], attention_mask=inputs['attention_mask']): Hlahloba tlaleho ea ketsahalo u sebelisa mohlala oa LLM.
  5. torch.argmax(outputs.logits): Ntša leibole e boletsoeng esale pele ea sehlopha.
  6. Fana_litlhahiso(sehlopha_se boletsoeng esale pele): Fana ka litlhahiso bakeng sa ho rarolla ketsahalo.
  7. agent.get_logs(): Bokella lintlha ho tsoa ho moemeli oa DevOps.
  8. hlahlobisisa_tlaleho_ea ketsahalo(tlaleho): Sekaseka tlaleho ea ketsahalo u sebelisa mofuta oa LLM.
  9. Phuputso_ea ketsahalo(): Sebelisa tharollo ea lipatlisiso tsa ketsahalo.
  10. torch.save(model.state_dict(), 'model.pth'): Boloka mofuta oa LLM.

Pono ya pono

Mona ke pono ea tharollo ea lipatlisiso tsa liketsahalo tse ikemetseng tse sebelisang Chart.js:

{
  "type": "bar",
  "data": {
    "labels": ["Patlisiso ka Manual", "Automated Investigation"],
    "lintlha tsa boitsebiso": [
      {
        "label": "Nako ea ho Rarolla",
        "data": [10, 2],
        "backgroundColor": ["rgba(255, 99, 132, 0.2)", "rgba(54, 162, 235, 0.2)"],
        "borderColor": ["rgba(255, 99, 132, 1)", "rgba(54, 162, 235, 1)"],
        "borderWidth": 1
      }
    ]
  },
  "likgetho": {
    "sehlooho": {
      "bontša": 'nete,
      "text": "Nako ea ho Rarolla Liketsahalo"
    },
    "makala": {
      "yAxes": [
        {
          "li-ticks": {
            "beginAtZero": 'nete
          }
        }
      ]
    }
  }
}

Ponahalo ena e bapisa nako ea ho rarolla liketsahalo ho sebelisoa mekhoa ea ho etsa lipatlisiso ka matsoho khahlano le mekhoa ea boithuto e ikemetseng e sebelisang li-DevOps le li-LLM.

Qetello

Ho kenya tšebetsong lipatlisiso tsa liketsahalo tse ikemetseng le baemeli ba DevOps le LLMs ho ka ntlafatsa haholo ts'ebetso le katleho ea tharollo ea liketsahalo. Ka ho kopanya baemeli ba DevOps le LLMs, mekhatlo e ka iketsetsa tlhahlobo ea li-log, ea fana ka litlhahiso tse hlokomelang maemo bakeng sa ho rarolla liketsahalo, le ho fokotsa nako ea ho rarolla liketsahalo. Re hlahlobile mateano a matekoane a DevOps, lipatlisiso tsa liketsahalo tse tsamaisoang ke AI, le karolo ea LLM ho ntlafatsa lisebelisoa tsa DevOps tse kang AWS DevOps Agent le Copado Agentia. Ka ho latela mokhoa oa mohato ka mohato o boletsoeng sehloohong sena, mekhatlo e ka fana ka tharollo ea lipatlisiso tsa liketsahalo tse scalable, tse ka tšeptjoang le tse sireletsehileng.

¿Tienes preguntas?

Hablemos

Contáctame en LinkedIn