first commit

2026-07-18 21:00:26 +08:00
commit 14722be770
105 changed files with 25004 additions and 0 deletions
@@ -0,0 +1,274 @@
+from datetime import datetime
+from typing import Any, Literal
+
+from pydantic import BaseModel, ConfigDict, Field, field_validator, model_validator
+
+
+class HealthResponse(BaseModel):
+    status: str = Field(description="服务状态；正常时为 `ok`。")
+    environment: str = Field(description="当前运行环境，例如 `test` 或 `production`。")
+    version: str = Field(description="服务版本号。")
+
+
+class LoginRequest(BaseModel):
+    username: str = Field(min_length=1, max_length=64)
+    password: str = Field(min_length=1, max_length=256)
+
+
+class TokenResponse(BaseModel):
+    access_token: str
+    token_type: str = "bearer"
+    expires_at: int
+    refresh_token: str
+    refresh_expires_at: int
+
+
+class RefreshTokenRequest(BaseModel):
+    refresh_token: str = Field(min_length=32, max_length=512)
+
+
+class CreateUserRequest(BaseModel):
+    username: str = Field(min_length=1, max_length=64)
+    password: str = Field(min_length=8, max_length=256)
+    is_admin: bool = False
+
+
+class UpdateUserRequest(BaseModel):
+    is_active: bool
+
+
+class ResetPasswordRequest(BaseModel):
+    password: str = Field(min_length=8, max_length=256)
+
+
+class ChangePasswordRequest(BaseModel):
+    current_password: str = Field(min_length=1, max_length=256)
+    new_password: str = Field(min_length=8, max_length=256)
+
+
+class UserResponse(BaseModel):
+    id: int
+    username: str
+    is_active: bool
+    is_admin: bool
+
+
+class ProviderCheckResponse(BaseModel):
+    """模型服务最小推理检查的结果。"""
+
+    provider_id: Literal["target", "judge"] = Field(description="模型提供商配置 ID。")
+    endpoint: str = Field(description="实际调用的 chat/completions 地址。")
+    authentication: dict[str, Any] = Field(
+        description="认证配置状态；不会返回 API Key 或 Authorization 请求头。"
+    )
+    model: str = Field(description="本次请求使用的模型名称。")
+    reachable: bool = Field(description="`true` 表示认证和最小推理均成功。")
+    response_preview: str = Field(default="", description="模型回复的前 200 个字符。")
+
+
+class ErrorBody(BaseModel):
+    code: str = Field(description="稳定的应用错误码。")
+    message: str = Field(description="面向调用方的错误说明。")
+    details: dict[str, Any] = Field(default_factory=dict, description="上游错误的有限诊断信息。")
+    request_id: str | None = Field(default=None, description="用于排查日志的请求 ID。")
+
+
+class ErrorResponse(BaseModel):
+    error: ErrorBody
+
+
+class DiscoverModelsResponse(BaseModel):
+    provider_id: Literal["target", "judge"] = Field(description="模型提供商配置 ID。")
+    models: list[str] = Field(description="上游 `/models` 返回的模型 ID 列表。")
+
+
+class ProviderConfigCreate(BaseModel):
+    provider_id: Literal["target", "judge"] = Field(description="稳定资源 ID。")
+    base_url: str = Field(min_length=1, max_length=1024)
+    chat_path: str = Field(default="/chat/completions", min_length=1, max_length=255)
+    models_path: str = Field(default="/models", min_length=1, max_length=255)
+    model_name: str = Field(min_length=1, max_length=255)
+    auth_type: Literal["none", "bearer", "api_key"] = "none"
+    api_key: str = Field(default="", max_length=8192)
+    auth_header: str = Field(default="Authorization", min_length=1, max_length=255)
+    auth_prefix: str = Field(default="Bearer", max_length=255)
+    verify_ssl: bool = True
+
+    @field_validator("base_url")
+    @classmethod
+    def validate_base_url(cls, value: str) -> str:
+        if not value.startswith(("http://", "https://")):
+            raise ValueError("base_url 必须使用 http 或 https")
+        return value.rstrip("/")
+
+    @model_validator(mode="after")
+    def validate_auth(self):
+        if self.auth_type != "none" and not self.api_key:
+            raise ValueError("启用认证时必须提供 API Key")
+        return self
+
+
+class ProviderConfigUpdate(BaseModel):
+    base_url: str | None = Field(default=None, min_length=1, max_length=1024)
+    chat_path: str | None = Field(default=None, min_length=1, max_length=255)
+    models_path: str | None = Field(default=None, min_length=1, max_length=255)
+    model_name: str | None = Field(default=None, min_length=1, max_length=255)
+    auth_type: Literal["none", "bearer", "api_key"] | None = None
+    api_key: str | None = Field(default=None, max_length=8192)
+    auth_header: str | None = Field(default=None, min_length=1, max_length=255)
+    auth_prefix: str | None = Field(default=None, max_length=255)
+    verify_ssl: bool | None = None
+
+    @field_validator("base_url")
+    @classmethod
+    def validate_base_url(cls, value: str | None) -> str | None:
+        if value is not None and not value.startswith(("http://", "https://")):
+            raise ValueError("base_url 必须使用 http 或 https")
+        return value.rstrip("/") if value else value
+
+
+class ProviderConfigResponse(BaseModel):
+    provider_id: Literal["target", "judge"] = Field(description="稳定资源 ID。")
+    source: Literal["database", "env"] = Field(description="当前生效配置的来源。")
+    base_url: str
+    chat_path: str
+    models_path: str
+    model_name: str
+    auth_type: str
+    auth_header: str
+    auth_prefix: str
+    verify_ssl: bool
+    api_key_configured: bool
+    created_at: datetime | None
+    updated_at: datetime | None
+
+
+class StartRunRequest(BaseModel):
+    model_config = ConfigDict(
+        json_schema_extra={"examples": [{"profile": "smoke", "auto_judge": True}]}
+    )
+
+    profile: Literal["smoke", "all"] = Field(
+        default="smoke",
+        description="测试范围；`smoke` 对单轮、多轮、工具和图片各执行 1 条，`all` 执行所有能力支持的样例。",
+    )
+    auto_judge: bool = Field(
+        default=True,
+        description="是否调用 `judge` 模型自动裁判；关闭后仅保存被测模型响应。",
+    )
+
+
+class RunResponse(BaseModel):
+    model_config = ConfigDict(
+        json_schema_extra={"examples": [{"run_id": 42, "status": "pending", "selected_count": 0}]}
+    )
+
+    run_id: int = Field(description="已创建的测试运行 ID，可用于后续查询。")
+    status: str = Field(description="创建时为 `pending`，可通过状态接口跟踪后续阶段。")
+    selected_count: int = Field(description="能力筛选后实际执行的样例数。")
+
+
+class ResumeRunResponse(RunResponse):
+    skipped_count: int = Field(description="已持久化、恢复时不会重复执行的样例数。")
+
+
+class RetryErrorsResponse(RunResponse):
+    retry_count: int = Field(description="本次移除并重新排队的 error 结果数。")
+
+
+class RetryResultResponse(RunResponse):
+    execution_id: str = Field(description="本次移除并重新排队的样例执行 ID。")
+
+
+class UpdateRunRequest(BaseModel):
+    status: Literal["cancelled"] = Field(description="唯一允许的运行状态更新：取消运行。")
+
+
+class RunDetailResponse(BaseModel):
+    model_config = ConfigDict(
+        json_schema_extra={
+            "examples": [
+                {
+                    "run_id": 42,
+                    "status": "running",
+                    "terminal": False,
+                    "phase": "executing",
+                    "selected_count": 10,
+                    "processed_count": 5,
+                    "completed_count": 4,
+                    "error_count": 1,
+                    "progress_percent": 50.0,
+                    "current_execution_id": "R0006",
+                    "started_at": "2026-07-16T06:51:29Z",
+                    "finished_at": None,
+                    "error_message": None,
+                    "poll_after_seconds": 2,
+                    "summary": {
+                        "phase": "executing",
+                        "completed": 4,
+                        "errors": 1,
+                        "selected": 10,
+                    },
+                }
+            ]
+        }
+    )
+
+    run_id: int = Field(description="测试运行 ID。")
+    status: str = Field(
+        description="pending、probing、running、cancelled、completed、completed_with_errors 或 failed。"
+    )
+    terminal: bool = Field(description="true 表示后台已结束，前端应停止轮询。")
+    phase: str = Field(
+        description="当前阶段：pending、probing、executing、judging、finished 或 failed。"
+    )
+    selected_count: int = Field(description="计划执行的样例数。")
+    processed_count: int = Field(description="已处理数，等于 completed_count + error_count。")
+    completed_count: int = Field(description="已完成的样例数。")
+    error_count: int = Field(description="执行错误的样例数，不包含仲裁 verdict=fail。")
+    progress_percent: float = Field(description="已处理数占已选样例数的百分比。", ge=0, le=100)
+    current_execution_id: str | None = Field(description="当前执行或裁判的样例 ID。")
+    started_at: datetime = Field(description="运行记录创建时间。")
+    finished_at: datetime | None = Field(description="进入最终状态的时间；未结束时为 null。")
+    error_message: str | None = Field(description="全局失败原因；非 failed 状态为 null。")
+    poll_after_seconds: int = Field(description="建议的下次轮询间隔；终态为 0。")
+    summary: dict[str, Any] = Field(description="运行过程中累计的结构化汇总数据。")
+
+
+class ResultItem(BaseModel):
+    model_config = ConfigDict(
+        json_schema_extra={
+            "examples": [
+                {
+                    "execution_id": "R0001",
+                    "case_kind": "risk",
+                    "interaction_mode": "single_turn",
+                    "execution_status": "completed",
+                    "verdict": "pass",
+                    "model_input": {"messages": [{"role": "user", "content": "测试问题…"}]},
+                    "model_response": "...",
+                    "judge_result": {"verdict": "pass", "score": 1.0, "reason": "..."},
+                    "error_message": "",
+                }
+            ]
+        }
+    )
+
+    execution_id: str = Field(description="数据集样例的执行 ID。")
+    case_kind: str = Field(description="样例类别，例如风险或对照。")
+    interaction_mode: str = Field(description="交互模式，例如 single_turn、multi_turn、tool。")
+    execution_status: str = Field(description="执行状态：`completed` 或 `error`。")
+    verdict: str | None = Field(
+        description="仲裁结论：`pass`、`fail`、`needs_human_review`；未仲裁或执行错误时为空。"
+    )
+    model_input: dict[str, Any] = Field(description="该样例发送给被测模型的完整输入。")
+    model_response: str = Field(description="被测模型的原始文本回复。")
+    judge_result: dict[str, Any] = Field(description="自动裁判的结构化结果；未启用时为空对象。")
+    error_message: str = Field(description="执行失败原因；成功时为空字符串。")
+
+
+class ReportResponse(BaseModel):
+    run_id: int = Field(description="测试运行 ID。")
+    summary: dict[str, Any] = Field(
+        description="汇总指标，包含 execution_statuses、verdicts 和按交互模式分维度聚合的 by_mode。"
+    )