ab1589921a
## 安全修复 (12项) - Webhook接口添加全局Token认证，过滤敏感请求头 - 修复JWT Base64 padding公式，防止签名验证绕过 - 数据库密码/飞书Token从源码移除，改为环境变量 - 工作流引擎添加路径遍历防护 (_resolve_safe_path) - eval()添加模板长度上限检查 - 审批API添加认证依赖 - 前端v-html增强XSS转义，console.log仅开发模式输出 - 500错误不再暴露内部异常详情 ## Agent运行时修复 (7项) - 删除_inject_knowledge_context中未定义db变量的finally块 - 工具执行添加try/except保护，异常不崩溃Agent - LLM重试计入budget计数器 - self_review异常时passed=False - max_iterations截断标记success=False - 工具参数JSON解析失败时记录警告日志 - run()开始时重置_llm_invocations计数器 ## 配置与基础设施 - DEBUG默认False，SQL_ECHO独立配置项 - init_db()补全13个缺失模型导入 - 新增WEBHOOK_AUTH_TOKEN/SQL_ECHO配置项 - 新增.env.example模板文件 ## 前端修复 (12项) - 登录改用URLSearchParams替代FormData - 401拦截器通过Pinia store统一清理状态 - SSE流超时从60s延长至300s - final/error事件时清除streamTimeout - localStorage聊天记录添加24h TTL - safeParseArgCount替代模板中裸JSON.parse - fetchUser 401时同时清除user对象 ## 新增模块 - 知识进化: knowledge_extractor/retriever/tasks - 数字孪生: shadow_executor/comparison模型 - 行为采集: behavior_middleware/collector/fingerprint_engine - 代码审查: code_review_agent/document_review_agent - 反馈学习: feedback_learner - 瓶颈检测/优化引擎/成本估算/需求估算 - 速率限制器 (rate_limiter) - Alembic迁移 015-020 ## 文档 - 商业化落地计划 - 8篇docs文档 (架构/API/部署/开发/贡献等) - Docker Compose生产配置 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
79 lines
2.3 KiB
Python
79 lines
2.3 KiB
Python
"""LLM 成本估算 — 基于模型定价和 Token 用量估算费用"""
|
||
from __future__ import annotations
|
||
|
||
import logging
|
||
from typing import Dict, Optional, Tuple
|
||
|
||
logger = logging.getLogger(__name__)
|
||
|
||
# 模型定价 (per 1M tokens, USD) — 2024 Q4 参考值
|
||
MODEL_PRICING: Dict[str, Tuple[float, float]] = {
|
||
# (input_price_per_1M, output_price_per_1M)
|
||
"gpt-4o": (2.50, 10.00),
|
||
"gpt-4o-mini": (0.15, 0.60),
|
||
"gpt-4-turbo": (10.00, 30.00),
|
||
"gpt-3.5-turbo": (0.50, 1.50),
|
||
"deepseek-chat": (0.14, 0.28),
|
||
"deepseek-v3": (0.27, 1.10),
|
||
"deepseek-r1": (0.55, 2.19),
|
||
"deepseek-v4-flash": (0.14, 0.28),
|
||
"deepseek-v4-pro": (0.27, 1.10),
|
||
"claude-3-opus": (15.00, 75.00),
|
||
"claude-3-sonnet": (3.00, 15.00),
|
||
"claude-3-haiku": (0.25, 1.25),
|
||
"claude-3.5-sonnet": (3.00, 15.00),
|
||
"claude-3.5-haiku": (1.00, 5.00),
|
||
}
|
||
|
||
# 缓存未命中模型的默认定价
|
||
DEFAULT_PRICING = (1.00, 4.00)
|
||
|
||
|
||
def estimate_cost(
|
||
model: str,
|
||
prompt_tokens: int = 0,
|
||
completion_tokens: int = 0,
|
||
) -> float:
|
||
"""估算单次 LLM 调用的费用（USD）。
|
||
|
||
Args:
|
||
model: 模型名称（模糊匹配）
|
||
prompt_tokens: 输入 token 数
|
||
completion_tokens: 输出 token 数
|
||
"""
|
||
input_price, output_price = _get_price(model)
|
||
cost = (prompt_tokens / 1_000_000) * input_price + (
|
||
completion_tokens / 1_000_000
|
||
) * output_price
|
||
return round(cost, 6)
|
||
|
||
|
||
def estimate_cost_yuan(
|
||
model: str,
|
||
prompt_tokens: int = 0,
|
||
completion_tokens: int = 0,
|
||
exchange_rate: float = 7.2,
|
||
) -> float:
|
||
"""估算费用（人民币）。"""
|
||
return round(estimate_cost(model, prompt_tokens, completion_tokens) * exchange_rate, 4)
|
||
|
||
|
||
def _get_price(model: str) -> Tuple[float, float]:
|
||
"""模糊匹配模型定价。"""
|
||
model_lower = model.lower().replace("-", "").replace(".", "")
|
||
for key, price in MODEL_PRICING.items():
|
||
if key.replace("-", "").replace(".", "") in model_lower:
|
||
return price
|
||
return DEFAULT_PRICING
|
||
|
||
|
||
def get_model_pricing_table() -> Dict[str, dict]:
|
||
"""返回模型定价表（供前端展示）。"""
|
||
return {
|
||
model: {
|
||
"input_per_1M": input_p,
|
||
"output_per_1M": output_p,
|
||
}
|
||
for model, (input_p, output_p) in MODEL_PRICING.items()
|
||
}
|