135 lines
5.8 KiB
Python
135 lines
5.8 KiB
Python
"""设置 API schema（模型网关与生成偏好）。
|
||
|
||
前端只见逻辑模型 / provider 密钥脱敏状态 / 生成参数偏好；
|
||
真实 model 名、endpoint URL、API Key 由后端持有，不进入响应（environment-config §3）。
|
||
"""
|
||
|
||
from pydantic import BaseModel, ConfigDict, Field
|
||
|
||
|
||
class ModelEndpointStatus(BaseModel):
|
||
"""endpoint 池状态。base_url / api_key 原值不回传（脱敏），仅回覆盖标记 + env 变量名。"""
|
||
|
||
name: str = Field(description="endpoint 名称（如 primary / fallback）。")
|
||
provider: str = Field(description="所属 provider。")
|
||
status: str = Field(description="endpoint 状态：active / inactive / unhealthy / cooldown。")
|
||
weight: int | None = Field(default=None, description="权重（weighted_random 路由用）。")
|
||
priority: int | None = Field(
|
||
default=None, description="优先级（priority_failover 用，数值越小越优先）。"
|
||
)
|
||
is_overridden: bool = Field(default=False, description="是否存在 DB 覆盖。")
|
||
has_url_override: bool = Field(
|
||
default=False, description="是否用 DB 覆盖了 base_url（优先于 .env）。"
|
||
)
|
||
has_key_override: bool = Field(
|
||
default=False, description="是否用 DB 覆盖了 api_key（优先于 .env）。"
|
||
)
|
||
base_url_env: str | None = Field(
|
||
default=None, description="base_url 的 .env 变量名（只读展示）。"
|
||
)
|
||
api_key_env: str | None = Field(
|
||
default=None, description="api_key 的 .env 变量名（只读展示）。"
|
||
)
|
||
|
||
|
||
class ProviderKeyStatus(BaseModel):
|
||
"""供应商密钥脱敏状态：configured 表示 .env 是否已配置该 provider 的 API Key。"""
|
||
|
||
provider: str = Field(description="provider 标识。")
|
||
configured: bool = Field(description="是否已配置 API Key（仅 true/false，绝不暴露值）。")
|
||
|
||
|
||
class LogicalModel(BaseModel):
|
||
"""逻辑模型（契约字典 §16）。业务层 model_hint 引用 name。"""
|
||
|
||
name: str = Field(description="逻辑模型名，业务层 model_hint 引用。")
|
||
display_name: str = Field(description="展示名。")
|
||
provider: str = Field(description="默认 provider。")
|
||
capabilities: list[str] = Field(
|
||
description="能力清单：chat / tool_use / structured_output / reasoning / long_context / vision。"
|
||
)
|
||
routing_strategy: str = Field(description="路由策略。")
|
||
supports_streaming: bool = Field(description="是否支持流式。")
|
||
supports_thinking: bool = Field(description="是否支持思考。")
|
||
supports_structured_output: bool = Field(description="是否支持结构化输出。")
|
||
context_window: int = Field(description="上下文窗口大小。")
|
||
max_output_tokens: int | None = Field(default=None, description="最大输出 token 数。")
|
||
endpoints: list[ModelEndpointStatus] | None = Field(
|
||
default=None, description="endpoint 池脱敏状态。"
|
||
)
|
||
is_default: bool | None = Field(default=None, description="是否为默认对话模型槽位。")
|
||
|
||
|
||
class ModelGatewayStatus(BaseModel):
|
||
model_config = ConfigDict(
|
||
json_schema_extra={
|
||
"examples": [
|
||
{
|
||
"default_model_hint": "default-chat",
|
||
"models": [],
|
||
"providers": [{"provider": "deepseek", "configured": True}],
|
||
}
|
||
]
|
||
}
|
||
)
|
||
|
||
models: list[LogicalModel] = Field(description="逻辑模型清单。")
|
||
providers: list[ProviderKeyStatus] = Field(description="provider 密钥脱敏状态。")
|
||
default_model_hint: str = Field(description="默认逻辑模型名。")
|
||
|
||
|
||
class GenerationPreferencesResponse(BaseModel):
|
||
"""生成参数偏好（业务层，不含密钥）。"""
|
||
|
||
model_config = ConfigDict(
|
||
json_schema_extra={
|
||
"examples": [
|
||
{
|
||
"default_model_hint": "default-chat",
|
||
"temperature": 0.7,
|
||
"top_p": 1.0,
|
||
"max_output_tokens": 4096,
|
||
"streaming": True,
|
||
}
|
||
]
|
||
}
|
||
)
|
||
|
||
default_model_hint: str = Field(description="默认对话逻辑模型名。")
|
||
temperature: float = Field(description="采样温度。")
|
||
top_p: float = Field(description="核采样概率。")
|
||
max_output_tokens: int = Field(description="最大输出 token 数。")
|
||
streaming: bool = Field(description="是否启用流式输出。")
|
||
|
||
|
||
class UpdateGenerationPreferencesRequest(BaseModel):
|
||
"""更新生成偏好；至少传一个字段。"""
|
||
|
||
default_model_hint: str | None = Field(
|
||
default=None, max_length=120, description="默认对话逻辑模型名。"
|
||
)
|
||
temperature: float | None = Field(default=None, ge=0.0, le=2.0, description="采样温度。")
|
||
top_p: float | None = Field(default=None, ge=0.0, le=1.0, description="核采样概率。")
|
||
max_output_tokens: int | None = Field(
|
||
default=None, ge=1, le=200000, description="最大输出 token 数。"
|
||
)
|
||
streaming: bool | None = Field(default=None, description="是否启用流式输出。")
|
||
|
||
|
||
class UpdateEndpointOverrideRequest(BaseModel):
|
||
"""更新 endpoint 覆盖（PATCH /settings/models/{model}/endpoints/{endpoint}）。
|
||
|
||
字段全可选；传 None 表示清除该字段覆盖、回退 yaml/env。base_url/api_key 经 DB
|
||
覆盖优先于 .env。响应脱敏，不回 base_url/api_key 原值。
|
||
"""
|
||
|
||
base_url: str | None = Field(
|
||
default=None, description="覆盖 base_url（DB 优先 env）；None 清除。"
|
||
)
|
||
api_key: str | None = Field(
|
||
default=None, description="覆盖 api_key（DB 优先 env）；None 清除。"
|
||
)
|
||
status: str | None = Field(default=None, description="覆盖 endpoint 状态。")
|
||
weight: int | None = Field(default=None, ge=0, description="覆盖权重。")
|
||
priority: int | None = Field(default=None, ge=0, description="覆盖优先级。")
|