技能 人工智能 大模型运维与生产部署

大模型运维与生产部署

v20260927
llm-ops
本指南全面覆盖大模型运维(LLMOps)全流程知识。集成了RAG、向量数据库(如Pinecone、Chroma)、嵌入、微调、高级提示工程和成本估算等核心技术。旨在帮助用户将AI原型转化为可信赖、可扩展、高效率的生产级AI产品。
获取技能
74 次下载
概览

LLM-OPS -- IA de Producao

Overview

LLM Operations -- RAG, embeddings, vector databases, fine-tuning, prompt engineering avancado, custos de LLM, evals de qualidade e arquiteturas de IA para producao. Ativar para: implementar RAG, criar pipeline de embeddings, Pinecone/Chroma/pgvector, fine-tuning, prompt engineering, reducao de custos de LLM, evals, cache semantico, streaming, agents.

When to Use This Skill

  • When you need specialized assistance with this domain

Do Not Use This Skill When

  • The task is unrelated to llm ops
  • A simpler, more specific tool can handle the request
  • The user needs general-purpose assistance without domain expertise

How It Works

A diferenca entre um prototipo de IA e um produto de IA e operabilidade.
LLM-Ops e a engenharia que torna IA confiavel, escalavel e economica.

Arquitetura Rag Completa

[Documentos] -> [Chunking] -> [Embeddings] -> [Vector DB]
|
[Query] -> [Embed query] -> [Semantic Search] -> [Top K chunks]
|
[LLM + Context] -> [Resposta]

Pipeline De Indexacao

from anthropic import Anthropic
import chromadb

client = Anthropic()
chroma = chromadb.PersistentClient(path="./chroma_db")

def chunktext(text, chunksize=500, overlap=50):
words = text.split()
chunks = []
for i in range(0, len(words), chunk_size - overlap):
chunk = " ".join(words[i:i + chunk_size])
if chunk: chunks.append(chunk)
return chunks

def indexdocument(docid, content_text, metadata=None):
chunks = chunktext(contenttext)
ids = [f"{docid}chunk_{i}" for i in range(len(chunks))]
collection.upsert(ids=ids, documents=chunks)
return len(chunks)

Pipeline De Query Com Rag

def ragquery(query, topk=5, system=None):
results = collection.query(
querytexts=[query], nresults=top_k,
include=["documents", "metadatas", "distances"])
context_parts = []
for doc, meta, dist in zip(results["documents"][0],
results["metadatas"][0],
results["distances"][0]):
if dist < 1.5:
src = meta.get("source", "doc")
context_parts.append(f"[Fonte: {src}]
{doc}")
context = "


".join(context_parts)
response = client.messages.create(
model="claude-opus-4-20250805", max_tokens=1024,
system=system or "Responda baseado no contexto.",
messages=[{"role": "user", "content": f"Contexto:
{context}

{query}"}])
return response.content[0].text


Escolha Do Vector Db

DBMelhor ParaHostingCusto
ChromaDesenvolvimento, localSelf-hostedGratis
pgvectorJa usa PostgreSQLSelf/CloudGratis
PineconeProducao gerenciadaCloudUSD 70+/mes
WeaviateMulti-modalSelf/CloudGratis+
QdrantAlta performanceSelf/CloudGratis+

Pgvector

CREATE EXTENSION IF NOT EXISTS vector;
CREATE TABLE knowledge_embeddings (
id UUID PRIMARY KEY DEFAULT genrandomuuid(),
content TEXT NOT NULL,
embedding vector(1536),
metadata JSONB,
created_at TIMESTAMPTZ DEFAULT NOW()
);
CREATE INDEX ON knowledge_embeddings
USING ivfflat (embedding vectorcosineops) WITH (lists = 100);
SELECT content, 1 - (embedding <=> QUERY_VECTOR) AS similarity
FROM knowledge_embeddings ORDER BY similarity DESC LIMIT 5;


Estrutura De Prompt De Elite

Componentes do system prompt Auri:

  • Identidade: Nome (Auri), Tom (Natural, caloroso, direto), Plataforma (Amazon Alexa)
  • Regras: Maximo 3 paragrafos curtos, sem markdown, linguagem conversacional
  • Capacidades: analise de negocios, conselho baseado em dados, criatividade
  • Limitacoes: sem internet tempo real, sem transacoes financeiras
  • Personalizacao: {username}, {userpreferences}, {relevant_history}

Chain-Of-Thought

def cot_analysis(problem: str) -> str:
steps = [
"1. O que exatamente esta sendo pedido?",
"2. Que informacoes sao criticas para resolver?",
"3. Quais abordagens possiveis existem?",
"4. Qual abordagem e melhor e por que?",
"5. Quais riscos ou limitacoes existem?",
]
prompt = f"Analise passo a passo:

PROBLEMA: {problem}

"
prompt += "
".join(steps) + "

Resposta final (concisa, para voz):"
return call_claude(prompt)


Cache Semantico

class SemanticCache:
def init(self, similarity_threshold=0.95):
self.threshold = similarity_threshold
self.cache = {}

def get_cached(self, query, embedding):
for cachedemb, (response, ) in self.cache.items():
if cosinesimilarity(embedding, cachedemb) >= self.threshold:
return response
return None

def set_cache(self, query, embedding, response):
self.cache[tuple(embedding)] = (response, query)

Estimativa De Custos Claude

PRICING = {
"claude-opus-4-20250805": {"input": 15.00, "output": 75.00},
"claude-sonnet-4-5": {"input": 3.00, "output": 15.00},
"claude-haiku-3-5": {"input": 0.80, "output": 4.00},
}

def estimatemonthlycost(model, avginput, avgoutput, reqperday):
p = PRICING[model]
daily = (avginput + avgoutput) * reqperday / 1e6
monthly = daily p["input"] 30
return {"model": model, "monthly_cost": "USD %.2f" % monthly}


Framework De Avaliacao

from anthropic import Anthropic
client = Anthropic()

def evaluate_response(question, expected, actual, criteria):
criteria_text = "
".join(f"- {c}" for c in criteria)
eval_prompt = (
f"Avalie a resposta do assistente de IA.

"
f"PERGUNTA: {question}
RESPOSTA ESPERADA: {expected}
"
f"RESPOSTA ATUAL: {actual}

Criterios:
{criteria_text}

"
"Nota 0-10 e justificativa para cada criterio. Formato JSON."
)
response = client.messages.create(
model="claude-haiku-3-5", max_tokens=1024,
messages=[{"role": "user", "content": eval_prompt}]
)
import json
return json.loads(response.content[0].text)

AURI_EVALS = [
{
"question": "Quais sao os principais riscos de abrir startup agora?",
"criteria": ["precisaofactual", "relevancia", "clarezapara_voz"]
},
]


6. Comandos

ComandoAcao
/rag-setupConfigura pipeline RAG completo
/embed-docsIndexa documentos no vector DB
/prompt-optimizeOtimiza prompt para qualidade e custo
/cost-estimateEstima custo mensal do LLM
/eval-runRoda suite de evals de qualidade
/cache-setupConfigura cache semantico
/model-selectEscolhe modelo ideal para o caso de uso

Best Practices

  • Provide clear, specific context about your project and requirements
  • Review all suggestions before applying them to production code
  • Combine with other complementary skills for comprehensive analysis

Common Pitfalls

  • Using this skill for tasks outside its domain expertise
  • Applying recommendations without understanding your specific context
  • Not providing enough project context for accurate analysis

Example

User request:

Use @llm-ops for this task: LLM Operations -- RAG, embeddings, vector databases, fine-tuning, prompt engineering avancado, custos de LLM, evals de qualidade e arquiteturas de IA para producao.

Limitations

  • Use this skill only when the task clearly matches the scope described above.
  • Do not treat the output as a substitute for environment-specific validation, testing, or expert review.
  • Stop and ask for clarification if required inputs, permissions, safety boundaries, or success criteria are missing.
信息
分类 人工智能
名称 llm-ops
版本 v20260927
大小 8.33KB
更新时间 2026-09-28
语言