GuangYu/QuestionProject
4
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(agent): persist production generation settings

Browse Source
This commit is contained in:
Nelson
2026-07-16 16:07:11 +08:00
parent 245b6b3eca
commit 6c02662d3b
9 changed files with 353 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
ai_knowledge_base_v2/apps/admin-web/src/App.vue
View File

@@ -103,7 +103,7 @@ const modelForm = reactive({
topK: null as number | null,
presencePenalty: null as number | null,
frequencyPenalty: null as number | null,
maxToken: null as number | null,
maxToken: 8192 as number | null,
contextWindow: null as number | null,
streamEnabled: 1,
responseFormat: "",
@@ -471,18 +471,6 @@ function buildModelPayload() {
payload[field] = null;
}
});
Object.assign(payload, {
temperature: null,
topP: null,
topK: null,
presencePenalty: null,
frequencyPenalty: null,
maxToken: null,
contextWindow: null,
streamEnabled: 1,
responseFormat: null,
extraParams: null,
});
return payload;
}
@@ -513,7 +501,7 @@ async function quickAddModel() {
topK: null,
presencePenalty: null,
frequencyPenalty: null,
maxToken: null,
maxToken: 8192,
contextWindow: null,
streamEnabled: 1,
responseFormat: null,
@@ -594,13 +582,13 @@ function editModel(row: ModelItem) {
apiKey: "",
authType: row.authType,
apiVersion: row.apiVersion ?? "",
temperature: null,
topP: null,
topK: null,
presencePenalty: null,
frequencyPenalty: null,
maxToken: null,
contextWindow: null,
temperature: row.temperature ?? null,
topP: row.topP ?? null,
topK: row.topK ?? null,
presencePenalty: row.presencePenalty ?? null,
frequencyPenalty: row.frequencyPenalty ?? null,
maxToken: row.maxToken ?? 8192,
contextWindow: row.contextWindow ?? null,
streamEnabled: row.streamEnabled,
responseFormat: row.responseFormat ?? "",
extraParams: row.extraParams ?? "",
@@ -626,7 +614,7 @@ function resetModelForm() {
topK: null,
presencePenalty: null,
frequencyPenalty: null,
maxToken: null,
maxToken: 8192,
contextWindow: null,
streamEnabled: 1,
responseFormat: "",

49
ai_knowledge_base_v2/apps/admin-web/src/components/AgentGenerationParameters.vue Normal file
View File

@@ -0,0 +1,49 @@
<script setup lang="ts">
defineProps<{
temperature: number | null;
topP: number | null;
topK: number | null;
presencePenalty: number | null;
frequencyPenalty: number | null;
maxToken: number | null;
disabled?: boolean;
}>();
defineEmits<{
"update:temperature": [value: number | null];
"update:topP": [value: number | null];
"update:topK": [value: number | null];
"update:presencePenalty": [value: number | null];
"update:frequencyPenalty": [value: number | null];
"update:maxToken": [value: number | null];
}>();
</script>
<template>
<div class="agent-config-grid">
<el-form-item>
<template #label><span class="agent-parameter-label">回答随机性<el-tooltip placement="top" :show-after="200" content="取值 0–2。数值越小，回答越稳定、一致；数值越大，表达越多样，但也更容易发散。答疑场景建议使用较低数值。"><button type="button" class="agent-parameter-help" aria-label="回答随机性说明">?</button></el-tooltip></span></template>
<el-input-number :model-value="temperature" :disabled="disabled" :min="0" :max="2" :step="0.1" @update:model-value="$emit('update:temperature', $event)" />
</el-form-item>
<el-form-item>
<template #label><span class="agent-parameter-label">核采样概率<el-tooltip placement="top" :show-after="200" content="取值 0–1。模型只从累计概率达到该数值的候选词中选择。数值越小越专注，越大越多样。留空表示使用模型供应商默认值。"><button type="button" class="agent-parameter-help" aria-label="核采样概率说明">?</button></el-tooltip></span></template>
<el-input-number :model-value="topP" :disabled="disabled" :min="0" :max="1" :step="0.05" @update:model-value="$emit('update:topP', $event)" />
</el-form-item>
<el-form-item>
<template #label><span class="agent-parameter-label">候选词数量<el-tooltip placement="top" :show-after="200" content="取值 1–1000。每次生成只从概率最高的前 K 个候选词中选择。数值越小越保守，越大选择越多。留空表示使用供应商默认值，部分模型可能不支持。"><button type="button" class="agent-parameter-help" aria-label="候选词数量说明">?</button></el-tooltip></span></template>
<el-input-number :model-value="topK" :disabled="disabled" :min="1" :max="1000" @update:model-value="$emit('update:topK', $event)" />
</el-form-item>
<el-form-item>
<template #label><span class="agent-parameter-label">最大输出长度<el-tooltip placement="top" :show-after="200" content="限制模型本次最多生成的 Token 数，不等于字数。建议正式答疑使用 4096–8192；数值越大允许回答越长，也会增加耗时和费用，最终仍受模型自身上限约束。"><button type="button" class="agent-parameter-help" aria-label="最大输出长度说明">?</button></el-tooltip></span></template>
<el-input-number :model-value="maxToken" :disabled="disabled" :min="256" :max="100000" :step="256" @update:model-value="$emit('update:maxToken', $event)" />
</el-form-item>
<el-form-item>
<template #label><span class="agent-parameter-label">主题重复惩罚<el-tooltip placement="top" :show-after="200" content="取值 -2–2，0 表示不调整。正数会降低已出现内容再次出现的概率，鼓励引入新主题；负数会更容易重复已出现内容。数值过高可能导致回答跳跃。"><button type="button" class="agent-parameter-help" aria-label="主题重复惩罚说明">?</button></el-tooltip></span></template>
<el-input-number :model-value="presencePenalty" :disabled="disabled" :min="-2" :max="2" :step="0.1" @update:model-value="$emit('update:presencePenalty', $event)" />
</el-form-item>
<el-form-item>
<template #label><span class="agent-parameter-label">高频重复惩罚<el-tooltip placement="top" :show-after="200" content="取值 -2–2，0 表示不调整。正数会根据词语已出现的次数降低其再次出现的概率，用于减少反复表达；负数会增加重复倾向。数值过高可能影响语句连贯性。"><button type="button" class="agent-parameter-help" aria-label="高频重复惩罚说明">?</button></el-tooltip></span></template>
<el-input-number :model-value="frequencyPenalty" :disabled="disabled" :min="-2" :max="2" :step="0.1" @update:model-value="$emit('update:frequencyPenalty', $event)" />
</el-form-item>
</div>
</template>

132
ai_knowledge_base_v2/apps/admin-web/src/components/AgentManagementView.vue
View File

@@ -3,7 +3,8 @@ import { ElMessage, ElMessageBox } from "element-plus";
import { computed, onMounted, reactive, ref, watch } from "vue";
import { api } from "../services/api";
import type { AgentDebugResult, KnowledgeItem, ModelItem, PromptDetail, PromptHistoryItem } from "../types/api";
import type { AgentDebugResult, AgentRuntimeConfig, KnowledgeItem, ModelItem, PromptDetail, PromptHistoryItem } from "../types/api";
import AgentGenerationParameters from "./AgentGenerationParameters.vue";
import AdminPagination from "./AdminPagination.vue";
const props = defineProps<{ previewKnowledgeId?: number | null }>();
@@ -11,6 +12,7 @@ const emit = defineEmits<{ consumedPreview: [] }>();
const loading = ref(false);
const promptSaving = ref(false);
const runtimeSaving = ref(false);
const historyLoading = ref(false);
const agentDebugging = ref(false);
const activeConfigTab = ref("prompt");
@@ -37,9 +39,18 @@ const agentForm = reactive({
topK: null as number | null,
presencePenalty: null as number | null,
frequencyPenalty: null as number | null,
maxToken: 1024 as number | null,
maxToken: 8192 as number | null,
question: "",
});
const runtimeConfig = ref<AgentRuntimeConfig | null>(null);
const runtimeForm = reactive({
temperature: null as number | null,
topP: null as number | null,
topK: null as number | null,
presencePenalty: null as number | null,
frequencyPenalty: null as number | null,
maxToken: 8192 as number | null,
});
const promptDirty = computed(() => promptContent.value !== savedPromptContent.value);
const selectedModelName = computed(() => {
@@ -63,13 +74,15 @@ watch(() => props.previewKnowledgeId, (knowledgeId) => {
async function load() {
loading.value = true;
try {
const [promptResult, modelRows, knowledgeRows] = await Promise.all([
api.prompt(), api.models(), api.knowledgeOptions(),
const [promptResult, modelRows, knowledgeRows, formalConfig] = await Promise.all([
api.prompt(), api.models(), api.knowledgeOptions(), api.agentRuntimeConfig(),
]);
applyPrompt(promptResult);
models.value = modelRows;
knowledge.value = knowledgeRows;
agentForm.modelId = modelRows.find((item) => item.enabled === 1)?.id ?? modelRows[0]?.id;
applyRuntimeConfig(formalConfig);
applyDebugModelDefaults(agentForm.modelId);
agentForm.knowledgeIds = knowledgeRows
.filter((item) => item.status === 1 && item.lifecycleStatus === "active")
.map((item) => item.id);
@@ -82,6 +95,56 @@ async function load() {
}
}
function applyRuntimeConfig(value: AgentRuntimeConfig) {
runtimeConfig.value = value;
Object.assign(runtimeForm, {
temperature: value.temperature,
topP: value.topP,
topK: value.topK,
presencePenalty: value.presencePenalty,
frequencyPenalty: value.frequencyPenalty,
maxToken: value.maxToken,
});
}
function applyDebugModelDefaults(modelId?: number) {
const model = models.value.find((item) => item.id === modelId);
if (!model) return;
Object.assign(agentForm, {
temperature: model.temperature ?? null,
topP: model.topP ?? null,
topK: model.topK ?? null,
presencePenalty: model.presencePenalty ?? null,
frequencyPenalty: model.frequencyPenalty ?? null,
maxToken: model.maxToken ?? (model.enabled === 1 ? runtimeConfig.value?.maxToken : null) ?? 8192,
});
}
async function saveRuntimeConfig() {
if (!runtimeConfig.value?.modelId) return ElMessage.error("请先在模型管理中启用一个正式模型");
if (!runtimeForm.maxToken) return ElMessage.warning("最大输出长度不能为空");
runtimeSaving.value = true;
try {
const saved = await api.saveAgentRuntimeConfig({
temperature: runtimeForm.temperature,
topP: runtimeForm.topP,
topK: runtimeForm.topK,
presencePenalty: runtimeForm.presencePenalty,
frequencyPenalty: runtimeForm.frequencyPenalty,
maxToken: runtimeForm.maxToken,
});
applyRuntimeConfig(saved);
const model = models.value.find((item) => item.id === saved.modelId);
if (model) Object.assign(model, runtimeForm);
if (agentForm.modelId === saved.modelId) applyDebugModelDefaults(saved.modelId ?? undefined);
ElMessage.success("正式运行参数已保存，将从用户端下一条消息开始生效");
} catch (error) {
ElMessage.error(errorMessage(error, "正式运行参数保存失败"));
} finally {
runtimeSaving.value = false;
}
}
function applyPreviewKnowledge(knowledgeId: number) {
if (knowledge.value.some((item) => item.id === knowledgeId)) {
agentForm.knowledgeIds = [knowledgeId];
@@ -262,7 +325,7 @@ function errorMessage(error: unknown, fallback: string) {
<div class="page-head agent-page-head">
<div>
<h2>Agent 管理</h2>
<p>维护主提示词及历史版本，配置本次调试参数，并验证真实检索与回答效果。</p>
<p>维护主提示词、用户端正式生成参数和历史版本，并验证真实检索与回答效果。</p>
</div>
<div class="agent-current-summary">
<el-tag type="success">当前版本 {{ prompt?.id ? `#${prompt.id}` : "系统默认" }}</el-tag>
@@ -286,11 +349,34 @@ function errorMessage(error: unknown, fallback: string) {
</div>
</el-tab-pane>
<el-tab-pane label="正式运行配置" name="runtime">
<div class="agent-section-intro">
<div><h3>用户端正式运行参数</h3><p>保存后从用户端下一条消息开始生效。当前正式模型：{{ runtimeConfig?.modelName || '未启用模型' }}</p></div>
<el-button type="primary" :loading="runtimeSaving" :disabled="!runtimeConfig?.modelId" @click="saveRuntimeConfig">保存正式参数</el-button>
</div>
<el-alert v-if="!runtimeConfig?.modelId" title="当前没有已启用模型，请先到模型管理中启用一个正式模型。" type="warning" :closable="false" show-icon />
<el-alert v-else title="最大输出长度默认提升为 8192 Token；设置过低会导致长回答在完成前被模型截断。实际可用上限仍受模型供应商限制。" type="info" :closable="false" show-icon />
<el-form label-position="top" class="agent-runtime-form">
<section class="agent-advanced-settings">
<div class="agent-subsection-title"><h4>正式生成参数</h4><span>用户端正式对话使用</span></div>
<AgentGenerationParameters
v-model:temperature="runtimeForm.temperature"
v-model:top-p="runtimeForm.topP"
v-model:top-k="runtimeForm.topK"
v-model:presence-penalty="runtimeForm.presencePenalty"
v-model:frequency-penalty="runtimeForm.frequencyPenalty"
v-model:max-token="runtimeForm.maxToken"
:disabled="!runtimeConfig?.modelId"
/>
</section>
</el-form>
</el-tab-pane>
<el-tab-pane label="调试配置" name="debug">
<div class="agent-section-intro"><div><h3>本次调试配置</h3><p>配置只作用于后台预览，不会修改模型和知识库的正式状态。</p></div></div>
<el-form label-position="top">
<el-form-item label="调试模型">
<el-select v-model="agentForm.modelId" placeholder="选择已接入模型">
<el-select v-model="agentForm.modelId" placeholder="选择已接入模型" @change="applyDebugModelDefaults">
<el-option v-for="model in models" :key="model.id" :label="model.displayName || model.modelName" :value="model.id" />
</el-select>
</el-form-item>
@@ -301,32 +387,14 @@ function errorMessage(error: unknown, fallback: string) {
</el-form-item>
<section class="agent-advanced-settings">
<div class="agent-subsection-title"><h4>高级生成参数</h4><span>仅影响本次后台调试</span></div>
<div class="agent-config-grid">
<el-form-item>
<template #label><span class="agent-parameter-label">回答随机性<el-tooltip placement="top" :show-after="200" content="取值 0–2。数值越小，回答越稳定、一致；数值越大，表达越多样，但也更容易发散。答疑场景建议使用较低数值。"><button type="button" class="agent-parameter-help" aria-label="回答随机性说明">?</button></el-tooltip></span></template>
<el-input-number v-model="agentForm.temperature" :min="0" :max="2" :step="0.1" />
</el-form-item>
<el-form-item>
<template #label><span class="agent-parameter-label">核采样概率<el-tooltip placement="top" :show-after="200" content="取值 0–1。模型只从累计概率达到该数值的候选词中选择。数值越小越专注，越大越多样。留空表示使用模型供应商默认值。"><button type="button" class="agent-parameter-help" aria-label="核采样概率说明">?</button></el-tooltip></span></template>
<el-input-number v-model="agentForm.topP" :min="0" :max="1" :step="0.05" />
</el-form-item>
<el-form-item>
<template #label><span class="agent-parameter-label">候选词数量<el-tooltip placement="top" :show-after="200" content="取值 1–1000。每次生成只从概率最高的前 K 个候选词中选择。数值越小越保守，越大选择越多。留空表示使用供应商默认值，部分模型可能不支持。"><button type="button" class="agent-parameter-help" aria-label="候选词数量说明">?</button></el-tooltip></span></template>
<el-input-number v-model="agentForm.topK" :min="1" :max="1000" />
</el-form-item>
<el-form-item>
<template #label><span class="agent-parameter-label">最大输出长度<el-tooltip placement="top" :show-after="200" content="限制模型本次最多生成的 Token 数，不等于字数，也不代表一定会生成到该长度。数值越大允许回答越长，同时可能增加耗时和费用，最终仍受模型上限约束。"><button type="button" class="agent-parameter-help" aria-label="最大输出长度说明">?</button></el-tooltip></span></template>
<el-input-number v-model="agentForm.maxToken" :min="1" :max="100000" />
</el-form-item>
<el-form-item>
<template #label><span class="agent-parameter-label">主题重复惩罚<el-tooltip placement="top" :show-after="200" content="取值 -2–2，0 表示不调整。正数会降低已出现内容再次出现的概率，鼓励引入新主题；负数会更容易重复已出现内容。数值过高可能导致回答跳跃。"><button type="button" class="agent-parameter-help" aria-label="主题重复惩罚说明">?</button></el-tooltip></span></template>
<el-input-number v-model="agentForm.presencePenalty" :min="-2" :max="2" :step="0.1" />
</el-form-item>
<el-form-item>
<template #label><span class="agent-parameter-label">高频重复惩罚<el-tooltip placement="top" :show-after="200" content="取值 -2–2，0 表示不调整。正数会根据词语已出现的次数降低其再次出现的概率，用于减少反复表达；负数会增加重复倾向。数值过高可能影响语句连贯性。"><button type="button" class="agent-parameter-help" aria-label="高频重复惩罚说明">?</button></el-tooltip></span></template>
<el-input-number v-model="agentForm.frequencyPenalty" :min="-2" :max="2" :step="0.1" />
</el-form-item>
</div>
<AgentGenerationParameters
v-model:temperature="agentForm.temperature"
v-model:top-p="agentForm.topP"
v-model:top-k="agentForm.topK"
v-model:presence-penalty="agentForm.presencePenalty"
v-model:frequency-penalty="agentForm.frequencyPenalty"
v-model:max-token="agentForm.maxToken"
/>
</section>
</el-form>
</el-tab-pane>

5
ai_knowledge_base_v2/apps/admin-web/src/services/api.ts
View File

@@ -2,6 +2,8 @@ import type {
AdminProfile,
AdminUser,
AgentDebugResult,
AgentGenerationConfig,
AgentRuntimeConfig,
AiLogRecord,
ApiResponse,
ChatDetail,
@@ -178,6 +180,9 @@ export const api = {
restorePrompt: (id: number) => request<PromptDetail>(`/admin/prompt/history/${id}/restore`, { method: "POST", body: "{}" }),
debugAgent: (payload: Record<string, unknown>) =>
request<AgentDebugResult>("/admin/agent/debug", { method: "POST", body: JSON.stringify(payload) }),
agentRuntimeConfig: () => request<AgentRuntimeConfig>("/admin/agent/runtime-config"),
saveAgentRuntimeConfig: (payload: AgentGenerationConfig) =>
request<AgentRuntimeConfig>("/admin/agent/runtime-config", { method: "PUT", body: JSON.stringify(payload) }),
models: () => request<ModelItem[]>("/admin/model/list"),
createModel: (payload: Record<string, unknown>) =>
request<ModelItem>("/admin/model", { method: "POST", body: JSON.stringify(payload) }),

14
ai_knowledge_base_v2/apps/admin-web/src/types/api.ts
View File

@@ -133,6 +133,20 @@ export interface ModelItem {
enabled: number;
}
export interface AgentGenerationConfig {
temperature: number | null;
topP: number | null;
topK: number | null;
presencePenalty: number | null;
frequencyPenalty: number | null;
maxToken: number;
}
export interface AgentRuntimeConfig extends AgentGenerationConfig {
modelId: number | null;
modelName: string | null;
}
export interface AgentDebugResult {
ok: boolean;
message: string;

63
ai_knowledge_base_v2/apps/backend/app/api/admin_settings.py
View File

@@ -13,6 +13,7 @@ from app.models.ai_config import ModelConfig, Prompt, SystemConfig
from app.models.knowledge import Knowledge
from app.schemas.admin import (
AgentDebugRequest,
AgentRuntimeConfigSaveRequest,
EnableModelRequest,
ModelSaveRequest,
PromptSaveRequest,
@@ -184,6 +185,46 @@ async def debug_agent(
return api_success(result)
@router.get("/agent/runtime-config")
def get_agent_runtime_config(
db: Session = Depends(get_db),
current_admin: Admin = Depends(get_current_admin),
) -> dict:
model = _enabled_model(db)
return api_success(_agent_runtime_config_dict(model))
@router.put("/agent/runtime-config")
def save_agent_runtime_config(
payload: AgentRuntimeConfigSaveRequest,
db: Session = Depends(get_db),
current_admin: Admin = Depends(get_current_admin),
) -> dict:
model = _enabled_model(db)
if model is None:
raise HTTPException(
status_code=status.HTTP_409_CONFLICT,
detail="当前没有已启用模型，请先在模型管理中启用一个模型",
)
model.temperature = payload.temperature
model.top_p = payload.topP
model.top_k = payload.topK
model.presence_penalty = payload.presencePenalty
model.frequency_penalty = payload.frequencyPenalty
model.max_token = payload.maxToken
db.add(model)
OperationLogService.write(
db,
admin_id=current_admin.id,
module="agent",
action="save_runtime_config",
target_id=model.id,
)
db.commit()
db.refresh(model)
return api_success(_agent_runtime_config_dict(model))
@router.get("/model/list")
def list_models(db: Session = Depends(get_db), current_admin: Admin = Depends(get_current_admin)) -> dict:
models = db.scalars(select(ModelConfig).order_by(ModelConfig.id.desc())).all()
@@ -376,6 +417,28 @@ def _model_dict(model: ModelConfig) -> dict:
}
def _enabled_model(db: Session) -> ModelConfig | None:
return db.scalar(
select(ModelConfig)
.where(ModelConfig.enabled == 1)
.order_by(ModelConfig.id.desc())
.limit(1)
)
def _agent_runtime_config_dict(model: ModelConfig | None) -> dict:
return {
"modelId": model.id if model is not None else None,
"modelName": (model.display_name or model.model_name) if model is not None else None,
"temperature": float(model.temperature) if model is not None and model.temperature is not None else None,
"topP": float(model.top_p) if model is not None and model.top_p is not None else None,
"topK": model.top_k if model is not None else None,
"presencePenalty": float(model.presence_penalty) if model is not None and model.presence_penalty is not None else None,
"frequencyPenalty": float(model.frequency_penalty) if model is not None and model.frequency_penalty is not None else None,
"maxToken": model.max_token if model is not None and model.max_token is not None else 8192,
}
def _default_prompt_detail() -> dict:
return {
"id": None,

11
ai_knowledge_base_v2/apps/backend/app/schemas/admin.py
View File

@@ -89,7 +89,16 @@ class AgentDebugRequest(BaseModel):
topK: int | None = Field(default=None, ge=1, le=1000)
presencePenalty: float | None = Field(default=None, ge=-2, le=2)
frequencyPenalty: float | None = Field(default=None, ge=-2, le=2)
maxToken: int | None = Field(default=1024, ge=1, le=100000)
maxToken: int | None = Field(default=8192, ge=1, le=100000)
class AgentRuntimeConfigSaveRequest(BaseModel):
temperature: float | None = Field(default=None, ge=0, le=2)
topP: float | None = Field(default=None, ge=0, le=1)
topK: int | None = Field(default=None, ge=1, le=1000)
presencePenalty: float | None = Field(default=None, ge=-2, le=2)
frequencyPenalty: float | None = Field(default=None, ge=-2, le=2)
maxToken: int = Field(default=8192, ge=256, le=100000)
class ModelSaveRequest(BaseModel):

2
ai_knowledge_base_v2/apps/backend/app/services/model_service.py
View File

@@ -494,7 +494,7 @@ def _agent_system_prompt() -> str:
def _max_output_tokens(model: ModelConfig) -> int:
configured = max(1, model.max_token or 1024)
configured = max(1, model.max_token or 8192)
if not model.context_window:
return configured
return min(configured, max(1, model.context_window // 3))

101
ai_knowledge_base_v2/apps/backend/tests/test_agent_runtime_config.py Normal file
View File

@@ -0,0 +1,101 @@
from decimal import Decimal
from sqlalchemy import create_engine
from sqlalchemy.orm import Session
from sqlalchemy.pool import StaticPool
from app.api.admin_settings import get_agent_runtime_config, save_agent_runtime_config
from app.models import Base
from app.models.admin import Admin
from app.models.ai_config import ModelConfig
from app.schemas.admin import AgentRuntimeConfigSaveRequest
from app.services.model_stream_service import _openai_stream_payload
from app.services.model_service import _max_output_tokens
from app.services.rag_service import RagResult
def _database() -> Session:
engine = create_engine(
"sqlite:///:memory:",
connect_args={"check_same_thread": False},
poolclass=StaticPool,
)
Base.metadata.create_all(engine)
return Session(engine)
def _admin() -> Admin:
return Admin(id=1, username="admin", password="hash", name="系统管理员", status=1)
def _model() -> ModelConfig:
return ModelConfig(
id=1,
provider="openai-compatible",
display_name="正式模型",
api_type="openai_compatible",
model_name="production-model",
base_url="https://example.com/v1",
api_url="",
api_key="encrypted",
auth_type="bearer",
max_token=None,
stream_enabled=1,
timeout_second=30,
enabled=1,
)
def test_runtime_config_defaults_to_long_answer_safe_max_tokens():
with _database() as db:
admin = _admin()
model = _model()
db.add_all([admin, model])
db.commit()
result = get_agent_runtime_config(db=db, current_admin=admin)["data"]
assert result["modelId"] == model.id
assert result["modelName"] == "正式模型"
assert result["maxToken"] == 8192
assert _max_output_tokens(model) == 8192
def test_saved_runtime_config_is_persisted_on_enabled_model():
with _database() as db:
admin = _admin()
model = _model()
db.add_all([admin, model])
db.commit()
result = save_agent_runtime_config(
AgentRuntimeConfigSaveRequest(
temperature=0.3,
topP=0.9,
topK=40,
presencePenalty=0.2,
frequencyPenalty=0.4,
maxToken=12000,
),
db=db,
current_admin=admin,
)["data"]
db.refresh(model)
assert result["maxToken"] == 12000
assert model.max_token == 12000
assert model.temperature == Decimal("0.30")
assert model.top_p == Decimal("0.90")
assert model.top_k == 40
assert model.presence_penalty == Decimal("0.20")
assert model.frequency_penalty == Decimal("0.40")
payload = _openai_stream_payload(
model,
RagResult(question="测试", knowledge_scopes=[], chunks=[], prompt="测试", allow_general_knowledge=True),
)
assert payload["max_tokens"] == 12000
assert payload["temperature"] == 0.3
assert payload["top_p"] == 0.9
assert payload["presence_penalty"] == 0.2
assert payload["frequency_penalty"] == 0.4