Rumah > Artikel > pembangunan bahagian belakang > Membina Sistem Ejen Perusahaan: Reka Bentuk dan Pengoptimuman Komponen Teras
Membina ejen AI gred perusahaan memerlukan pertimbangan yang teliti terhadap reka bentuk komponen, seni bina sistem dan amalan kejuruteraan. Artikel ini meneroka komponen utama dan amalan terbaik untuk membina sistem ejen yang teguh dan berskala.
from typing import Protocol, Dict from jinja2 import Template class PromptTemplate(Protocol): def render(self, **kwargs) -> str: pass class JinjaPromptTemplate: def __init__(self, template_string: str): self.template = Template(template_string) def render(self, **kwargs) -> str: return self.template.render(**kwargs) class PromptLibrary: def __init__(self): self.templates: Dict[str, PromptTemplate] = {} def register_template(self, name: str, template: PromptTemplate): self.templates[name] = template def get_template(self, name: str) -> PromptTemplate: return self.templates[name]
class PromptVersion: def __init__(self, version: str, template: str, metadata: dict): self.version = version self.template = template self.metadata = metadata self.test_cases = [] def add_test_case(self, inputs: dict, expected_output: str): self.test_cases.append((inputs, expected_output)) def validate(self) -> bool: template = JinjaPromptTemplate(self.template) for inputs, expected in self.test_cases: result = template.render(**inputs) if not self._validate_output(result, expected): return False return True
from typing import Any, List from datetime import datetime class MemoryEntry: def __init__(self, content: Any, importance: float): self.content = content self.importance = importance self.timestamp = datetime.now() self.access_count = 0 class MemoryLayer: def __init__(self, capacity: int): self.capacity = capacity self.memories: List[MemoryEntry] = [] def add(self, entry: MemoryEntry): if len(self.memories) >= self.capacity: self._evict() self.memories.append(entry) def _evict(self): # Implement memory eviction strategy self.memories.sort(key=lambda x: x.importance * x.access_count) self.memories.pop(0) class HierarchicalMemory: def __init__(self): self.working_memory = MemoryLayer(capacity=5) self.short_term = MemoryLayer(capacity=50) self.long_term = MemoryLayer(capacity=1000) def store(self, content: Any, importance: float): entry = MemoryEntry(content, importance) if importance > 0.8: self.working_memory.add(entry) elif importance > 0.5: self.short_term.add(entry) else: self.long_term.add(entry)
from typing import List, Tuple import numpy as np from sklearn.metrics.pairwise import cosine_similarity class MemoryIndex: def __init__(self, embedding_model): self.embedding_model = embedding_model self.embeddings = [] self.memories = [] def add(self, memory: MemoryEntry): embedding = self.embedding_model.embed(memory.content) self.embeddings.append(embedding) self.memories.append(memory) def search(self, query: str, k: int = 5) -> List[Tuple[MemoryEntry, float]]: query_embedding = self.embedding_model.embed(query) similarities = cosine_similarity( [query_embedding], self.embeddings )[0] top_k_indices = np.argsort(similarities)[-k:] return [ (self.memories[i], similarities[i]) for i in top_k_indices ]
from typing import List, Optional from dataclasses import dataclass import uuid @dataclass class ThoughtNode: content: str confidence: float supporting_evidence: List[str] class ReasoningChain: def __init__(self): self.chain_id = str(uuid.uuid4()) self.nodes: List[ThoughtNode] = [] self.metadata = {} def add_thought(self, thought: ThoughtNode): self.nodes.append(thought) def get_path(self) -> List[str]: return [node.content for node in self.nodes] def get_confidence(self) -> float: if not self.nodes: return 0.0 return sum(n.confidence for n in self.nodes) / len(self.nodes)
import logging from opentelemetry import trace from prometheus_client import Histogram reasoning_time = Histogram( 'reasoning_chain_duration_seconds', 'Time spent in reasoning chain' ) class ChainMonitor: def __init__(self): self.tracer = trace.get_tracer(__name__) def monitor_chain(self, chain: ReasoningChain): with self.tracer.start_as_current_span("reasoning_chain") as span: span.set_attribute("chain_id", chain.chain_id) with reasoning_time.time(): for node in chain.nodes: with self.tracer.start_span("thought") as thought_span: thought_span.set_attribute( "confidence", node.confidence ) logging.info( f"Thought: {node.content} " f"(confidence: {node.confidence})" )
from abc import ABC, abstractmethod from typing import Generic, TypeVar T = TypeVar('T') class Component(ABC, Generic[T]): @abstractmethod def process(self, input_data: T) -> T: pass class Pipeline: def __init__(self): self.components: List[Component] = [] def add_component(self, component: Component): self.components.append(component) def process(self, input_data: Any) -> Any: result = input_data for component in self.components: result = component.process(result) return result
class ComponentRegistry: _instance = None def __new__(cls): if cls._instance is None: cls._instance = super().__new__(cls) cls._instance.components = {} return cls._instance def register(self, name: str, component: Component): self.components[name] = component def get(self, name: str) -> Optional[Component]: return self.components.get(name) def create_pipeline(self, component_names: List[str]) -> Pipeline: pipeline = Pipeline() for name in component_names: component = self.get(name) if component: pipeline.add_component(component) return pipeline
from dataclasses import dataclass from typing import Dict import time @dataclass class PerformanceMetrics: latency: float memory_usage: float token_count: int success_rate: float class PerformanceMonitor: def __init__(self): self.metrics: Dict[str, List[PerformanceMetrics]] = {} def record_operation( self, operation_name: str, metrics: PerformanceMetrics ): if operation_name not in self.metrics: self.metrics[operation_name] = [] self.metrics[operation_name].append(metrics) def get_average_metrics( self, operation_name: str ) -> Optional[PerformanceMetrics]: if operation_name not in self.metrics: return None metrics_list = self.metrics[operation_name] return PerformanceMetrics( latency=sum(m.latency for m in metrics_list) / len(metrics_list), memory_usage=sum(m.memory_usage for m in metrics_list) / len(metrics_list), token_count=sum(m.token_count for m in metrics_list) / len(metrics_list), success_rate=sum(m.success_rate for m in metrics_list) / len(metrics_list) )
class PerformanceOptimizer: def __init__(self, monitor: PerformanceMonitor): self.monitor = monitor self.thresholds = { 'latency': 1.0, # seconds 'memory_usage': 512, # MB 'token_count': 1000, 'success_rate': 0.95 } def analyze_performance(self, operation_name: str) -> List[str]: metrics = self.monitor.get_average_metrics(operation_name) if not metrics: return [] recommendations = [] if metrics.latency > self.thresholds['latency']: recommendations.append( "Consider implementing caching or parallel processing" ) if metrics.memory_usage > self.thresholds['memory_usage']: recommendations.append( "Optimize memory usage through batch processing" ) if metrics.token_count > self.thresholds['token_count']: recommendations.append( "Implement prompt optimization to reduce token usage" ) if metrics.success_rate < self.thresholds['success_rate']: recommendations.append( "Review error handling and implement retry mechanisms" ) return recommendations
Membina sistem Ejen gred perusahaan memerlukan perhatian yang teliti terhadap:
Atas ialah kandungan terperinci Membina Sistem Ejen Perusahaan: Reka Bentuk dan Pengoptimuman Komponen Teras. Untuk maklumat lanjut, sila ikut artikel berkaitan lain di laman web China PHP!