Files
AgentLab/backend/app/providers/catalog.py
T
ThearchyHelios 03c18c4d78 界面文案整改:口语改书面、术语统一、不再露出实现细节;修掉几处说错的提示;新增文案检查防回潮
一、文案(前端约 1600 处,后端传到界面的约 880 处)
- 术语统一:工作流层面用「运行」(继续运行、重新运行、试运行),节点和工具层面用「执行」,不再用「跑」;
  总时长 / 执行时长 / 等待审批;服务端;助手;Agent;token;工作流;固定版本;模型接入;系统提示;批准 / 驳回;受管级别
- 口语改书面:取不到、没取回来、读不懂、连不上、多半、照样、兜底、揉成这类说法改成「无法 / 未能 / 失败 / 请」;去掉拟人写法
- 不再露出实现细节:界面上不写 checkpoint、同一拍、哈希(凭证处除外),不写改版历史(「以前…」「老数据」);
  报错、校验、门禁、自动修复里裸写的配置键名和枚举值换成界面上的中文标签,由新的 backend/app/engine/labels.py 统一提供;
  发布弹窗里结论句的选项不再是 off / require_citation
- 同一件事只留一种说法:生成阶段文案、出具档位说明、画布锁定提示各合成一份
- 去掉界面上的工业场景示例(MES、ERP、「张工」「王工」)和真实产品名,换成通用示例

二、说错的提示(都有测试证明)
- 数据库测连接失败时的专用提示以前从不出现(按报错文字全等比较,永远比不上):改为按错误类型判断,不再拼出重复的前缀
- 「设置 → 供应商」改指「设置 → 模型接入」;「在设置里建数据源」改指「数据」页
- 缺少 API Key、模型接入已停用的报错不再夹英文,写明去哪里填
- 两处原因说错(「这种引用在后续版本支持」「这一版还不支持」)改成真实原因
- 节点说明里不准确的说法改正:代码节点在 microVM 或系统沙箱中执行;报告撰写、口径卡的说明不再说过头
- 导出的证据 CSV 里两列同名,第三列改为「片段编号」

三、前后端对齐
- 前端按后端原文识别日志、报错的地方,新旧两种写法都认得(历史运行里存的是旧文字),逐条补了新旧原文的核对
- 后端交回模型的原指令放进单独的 for_model 字段,前端在解析响应时去掉,不会显示到界面上

四、检查
- 检查脚本里按原文匹配的断言改成新文案;改完会变成永远成立的反向断言逐条修正,并做了变异验证
- 新增 check-copy(静态扫描前端界面字符串和后端上界面的文案):再出现「跑」「后端」「多半」「令牌」、
  界面上的 Copilot、裸写的配置键名、半角标点等就报错;确需保留的写进 scripts/copy-allow.json 并注明理由
2026-09-30 08:05:15 +08:00

241 lines
8.8 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
from __future__ import annotations
from typing import Any
# --------------------------------------------------------------------------
# 模型目录
#
# pricing 单位是「美元 / 百万 token」,用于 run 的成本估算。
# 新模型的价格会变,这份表只是开箱即用的默认值,用户可以在设置页改。
# --------------------------------------------------------------------------
ANTHROPIC_MODELS: list[dict[str, Any]] = [
{"id": "claude-opus-5", "label": "Claude Opus 5", "context": 1_000_000,
"pricing": {"input": 5.0, "output": 25.0}, "thinking": "adaptive"},
{"id": "claude-fable-5-1", "label": "Claude Fable 5.1", "context": 1_000_000,
"pricing": {"input": 10.0, "output": 50.0}, "thinking": "always"},
{"id": "claude-opus-4-8", "label": "Claude Opus 4.8", "context": 1_000_000,
"pricing": {"input": 5.0, "output": 25.0}, "thinking": "adaptive"},
{"id": "claude-sonnet-5", "label": "Claude Sonnet 5", "context": 1_000_000,
"pricing": {"input": 2.0, "output": 10.0}, "thinking": "adaptive"},
{"id": "claude-haiku-4-5", "label": "Claude Haiku 4.5", "context": 200_000,
"pricing": {"input": 1.0, "output": 5.0}, "thinking": "budget"},
]
OPENAI_MODELS: list[dict[str, Any]] = [
{"id": "gpt-5.2", "label": "GPT-5.2", "context": 400_000,
"pricing": {"input": 1.25, "output": 10.0}},
{"id": "gpt-5.2-mini", "label": "GPT-5.2 mini", "context": 400_000,
"pricing": {"input": 0.25, "output": 2.0}},
{"id": "o4", "label": "o4", "context": 200_000,
"pricing": {"input": 3.0, "output": 12.0}},
]
# 常见的 OpenAI 兼容服务,一键添加用
COMPATIBLE_PRESETS: list[dict[str, Any]] = [
{
"key": "deepseek",
"label": "DeepSeek",
"base_url": "https://api.deepseek.com/v1",
"models": [
{"id": "deepseek-chat", "label": "DeepSeek Chat", "context": 128_000,
"pricing": {"input": 0.27, "output": 1.1}},
{"id": "deepseek-reasoner", "label": "DeepSeek Reasoner", "context": 128_000,
"pricing": {"input": 0.55, "output": 2.19}},
],
},
{
"key": "moonshot",
"label": "Moonshot / Kimi",
"base_url": "https://api.moonshot.cn/v1",
"models": [
{"id": "kimi-k2-turbo-preview", "label": "Kimi K2 Turbo", "context": 256_000,
"pricing": {"input": 0.6, "output": 2.5}},
{"id": "moonshot-v1-128k", "label": "Moonshot v1 128k", "context": 128_000,
"pricing": {"input": 0.8, "output": 3.0}},
],
},
{
"key": "dashscope",
"label": "阿里云百炼 / 通义千问",
"base_url": "https://dashscope.aliyuncs.com/compatible-mode/v1",
"models": [
{"id": "qwen-max", "label": "Qwen Max", "context": 32_000,
"pricing": {"input": 0.34, "output": 1.37}},
{"id": "qwen-plus", "label": "Qwen Plus", "context": 131_000,
"pricing": {"input": 0.11, "output": 0.28}},
],
},
{
"key": "zhipu",
"label": "智谱 GLM",
"base_url": "https://open.bigmodel.cn/api/paas/v4",
"models": [
{"id": "glm-4.6", "label": "GLM-4.6", "context": 200_000,
"pricing": {"input": 0.6, "output": 2.2}},
],
},
{
"key": "siliconflow",
"label": "硅基流动",
"base_url": "https://api.siliconflow.cn/v1",
"models": [
{"id": "deepseek-ai/DeepSeek-V3", "label": "DeepSeek V3", "context": 128_000,
"pricing": {"input": 0.27, "output": 1.1}},
],
},
{
"key": "openrouter",
"label": "OpenRouter",
"base_url": "https://openrouter.ai/api/v1",
"models": [
{"id": "anthropic/claude-opus-5", "label": "Claude Opus 5", "context": 1_000_000,
"pricing": {"input": 5.0, "output": 25.0}},
],
},
{
"key": "ollama",
"label": "Ollama(本地)",
"base_url": "http://localhost:11434/v1",
"api_key_optional": True,
"models": [
{"id": "qwen3:8b", "label": "Qwen3 8B(本地)", "context": 32_000,
"pricing": {"input": 0.0, "output": 0.0}},
{"id": "llama3.2", "label": "Llama 3.2(本地)", "context": 128_000,
"pricing": {"input": 0.0, "output": 0.0}},
],
},
]
MOCK_MODELS: list[dict[str, Any]] = [
{"id": "mock-fast", "label": "演示模型(无需 API Key)", "context": 128_000,
"pricing": {"input": 0.0, "output": 0.0}},
]
PROVIDER_KINDS = [
{
"kind": "anthropic",
"label": "Anthropic",
"base_url_hint": "留空使用官方地址;填写后使用自定义网关(默认读取 ANTHROPIC_BASE_URL)",
"models": ANTHROPIC_MODELS,
"default_model": "claude-opus-5",
},
{
"kind": "openai",
"label": "OpenAI",
"base_url_hint": "留空使用官方地址",
"models": OPENAI_MODELS,
"default_model": "gpt-5.2",
},
{
"kind": "openai_compatible",
"label": "OpenAI 兼容(DeepSeek / Kimi / 通义 / Ollama…)",
"base_url_hint": "必填,例如 https://api.deepseek.com/v1",
"models": [],
"presets": COMPATIBLE_PRESETS,
"default_model": "",
},
{
"kind": "mock",
"label": "演示模型(无需 API Key)",
"base_url_hint": "",
"models": MOCK_MODELS,
"default_model": "mock-fast",
},
]
# --------------------------------------------------------------------------
# 模型能力
# --------------------------------------------------------------------------
# Claude 4.6 及以后的模型移除了采样参数,传 temperature/top_p 会直接 400。
_NO_SAMPLING_PREFIXES = (
"claude-fable-5",
"claude-mythos-5",
"claude-opus-5",
"claude-opus-4-8",
"claude-opus-4-7",
"claude-opus-4-6",
"claude-sonnet-5",
"claude-sonnet-4-6",
)
# 支持 adaptive thinking(而不是老的 budget_tokens)的模型
_ADAPTIVE_THINKING_PREFIXES = _NO_SAMPLING_PREFIXES
# 支持 output_config.effort
_EFFORT_PREFIXES = _NO_SAMPLING_PREFIXES + ("claude-opus-4-5",)
# 交错思考(工具结果回来之后再想一段,而不是一轮开头想完就只剩执行)。
#
# 4.6 及以后走 adaptive thinking,交错是自带的,不需要也不该再挂 beta header
# ——在 Opus 4.6 上手动模式压根没有交错思考,只有 adaptive 有。上面那张
# _ADAPTIVE_THINKING_PREFIXES 覆盖的都属于这一档。
#
# 需要显式开的是 Claude 4 / 4.5 这一代。目录里没有列它们,但模型名是放行的
# (用户可以手填),所以这条路径要留着。Haiku 4.5 不支持,前缀也就不包含它。
_INTERLEAVED_BETA_PREFIXES = ("claude-opus-4", "claude-sonnet-4")
INTERLEAVED_BETA = "interleaved-thinking-2025-05-14"
def needs_interleaved_beta(model: str) -> bool:
"""这个模型要不要靠 beta header 才会在工具之间思考。"""
name = _norm(model)
if supports_adaptive_thinking(name):
return False # adaptive 自带,多挂一个 header 只会是噪音
return name.startswith(_INTERLEAVED_BETA_PREFIXES)
def _norm(model: str) -> str:
"""去掉 openrouter 之类的前缀,只看模型名本身。"""
return (model or "").split("/")[-1].strip()
def supports_sampling(model: str) -> bool:
"""这个模型还能不能传 temperature / top_p。"""
return not _norm(model).startswith(_NO_SAMPLING_PREFIXES)
def supports_adaptive_thinking(model: str) -> bool:
return _norm(model).startswith(_ADAPTIVE_THINKING_PREFIXES)
def supports_effort(model: str) -> bool:
return _norm(model).startswith(_EFFORT_PREFIXES)
def find_pricing(model: str) -> dict[str, float] | None:
target = _norm(model)
pools = [ANTHROPIC_MODELS, OPENAI_MODELS, MOCK_MODELS]
pools += [p["models"] for p in COMPATIBLE_PRESETS]
for pool in pools:
for entry in pool:
if _norm(entry["id"]) == target:
return entry.get("pricing")
return None
def find_context(model: str) -> int | None:
"""目录里登记的上下文窗口(token)。没登记的返回 None,由调用方取保守值。"""
target = _norm(model)
pools = [ANTHROPIC_MODELS, OPENAI_MODELS, MOCK_MODELS]
pools += [p["models"] for p in COMPATIBLE_PRESETS]
for pool in pools:
for entry in pool:
if _norm(entry["id"]) == target:
value = entry.get("context")
return int(value) if isinstance(value, (int, float)) and value > 0 else None
return None
def estimate_cost(model: str, input_tokens: int, output_tokens: int) -> float:
"""按目录价估算一次调用的花费(美元)。"""
pricing = find_pricing(model)
if not pricing:
return 0.0
return (
input_tokens / 1_000_000 * pricing.get("input", 0.0)
+ output_tokens / 1_000_000 * pricing.get("output", 0.0)
)