From fd3e2992b1c1ef406605bdb3297a42b48099aac2 Mon Sep 17 00:00:00 2001 From: wizardchen Date: Thu, 23 Apr 2026 23:13:10 +0800 Subject: [PATCH] feat(models): support custom HTTP headers across all remote model calls MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Introduce per-model `custom_headers` config (similar to OpenAI Python SDK's `extra_headers`) so users can inject gateway auth tokens, trace IDs, etc. into every outbound API request for chat / embedding / rerank / VLLM / ASR models. Reserved headers like Authorization / Content-Type are always preserved to avoid breaking auth or signing flows. Along the way, unify production and "test connection" paths onto a single `ConfigFromModel(*types.Model, appID, appSecret)` constructor in each model package. Both `service.modelService.GetXxxModel` and the four `handler.initialization.Check*Model` / `TestEmbeddingModel` endpoints now go through the same field mapping, so new parameters only need to be added in one place. Backend: - Add `CustomHeaders map[string]string` to `types.ModelParameters` with JSON/YAML `omitempty` tags for forward compatibility. - New `internal/utils/extraheaders.go` providing `ApplyCustomHeaders` (for hand-rolled HTTP paths) and `WrapHTTPClientWithHeaders` / `CustomHeadersRoundTripper` (for SDK paths like go-openai). Reserved headers (Authorization, api-key, Content-Type, Accept, Host, Content-Length, User-Agent) are filtered out. - Inject headers in every remote model path: chat (SDK + raw HTTP), embedding (openai/aliyun/jina/volcengine/nvidia/azure_openai), rerank (remote_api/aliyun/jina/nvidia/zhipu — via shared `customHeaderSetter` interface), VLM and ASR (via http.Client wrapping). - Add `ConfigFromModel` to each of chat/embedding/rerank/vlm/asr, consolidating field mapping (ExtraConfig, CustomHeaders, InterfaceType defaulting, WeKnoraCloud credentials, etc.) and cover it with dedicated unit tests per package. - Refactor `service/model.go`: replace ~100 lines of hand-written Config literals with one-line `ConfigFromModel` calls; drop the now-unused `stringMapToAnyMap` helper. - Refactor `handler/initialization.go` test-connection endpoints: merge four ad-hoc request structs into one `ModelTestRequest`, extract `buildTestModel` and `resolveTenantWeKnoraCloudCreds` helpers, and route all four endpoints through `ConfigFromModel` + `NewXxx` so the test path is now behaviorally identical to production. Frontend: - Add `custom_headers` to the `ModelConfig` API type. - `ModelEditorDialog.vue`: new Key-Value header editor (add/remove rows), auto-converts map <-> array when loading/saving, and passes the current header set to the `Test Connection` button so the preview reflects exactly what production will send. - `ModelSettings.vue`: serialize the header array back into a map for the backend (drops empty rows). - i18n: add custom-header labels / descriptions / placeholders to zh-CN, en-US, ru-RU, ko-KR. Tests: - `internal/utils/extraheaders_test.go` covers reserved-header filtering and round-tripper wrapping (including nil-client fallback). - New `config_from_model_test.go` in chat / embedding / rerank / vlm / asr verifies all fields (CustomHeaders, ExtraConfig, InterfaceType defaulting, AppID/AppSecret) are propagated end-to-end. --- frontend/src/api/initialization/index.ts | 19 +- frontend/src/api/model/index.ts | 3 + frontend/src/components/ModelEditorDialog.vue | 157 +++++++++- frontend/src/i18n/locales/en-US.ts | 5 + frontend/src/i18n/locales/ko-KR.ts | 5 + frontend/src/i18n/locales/ru-RU.ts | 5 + frontend/src/i18n/locales/zh-CN.ts | 5 + frontend/src/views/settings/ModelSettings.vue | 19 +- internal/application/service/model.go | 97 +------ internal/handler/initialization.go | 272 ++++++++---------- internal/models/asr/asr.go | 19 ++ internal/models/asr/config_from_model_test.go | 36 +++ internal/models/asr/openai.go | 10 +- internal/models/chat/chat.go | 28 +- .../models/chat/config_from_model_test.go | 59 ++++ internal/models/chat/remote_api.go | 36 ++- internal/models/embedding/aliyun.go | 8 + internal/models/embedding/azure_openai.go | 8 + .../embedding/config_from_model_test.go | 43 +++ internal/models/embedding/embedder.go | 71 ++++- internal/models/embedding/jina.go | 24 +- internal/models/embedding/nvidia.go | 24 +- internal/models/embedding/openai.go | 9 + internal/models/embedding/volcengine.go | 8 + internal/models/rerank/aliyun_reranker.go | 18 +- .../models/rerank/config_from_model_test.go | 32 +++ internal/models/rerank/jina_reranker.go | 18 +- internal/models/rerank/nvidia_reranker.go | 18 +- internal/models/rerank/remote_api.go | 18 +- internal/models/rerank/reranker.go | 55 +++- internal/models/rerank/zhipu_reranker.go | 18 +- internal/models/vlm/config_from_model_test.go | 59 ++++ internal/models/vlm/remote_api.go | 10 +- internal/models/vlm/vlm.go | 44 +++ internal/types/model.go | 7 +- internal/utils/extraheaders.go | 86 ++++++ internal/utils/extraheaders_test.go | 86 ++++++ 37 files changed, 1105 insertions(+), 334 deletions(-) create mode 100644 internal/models/asr/config_from_model_test.go create mode 100644 internal/models/chat/config_from_model_test.go create mode 100644 internal/models/embedding/config_from_model_test.go create mode 100644 internal/models/rerank/config_from_model_test.go create mode 100644 internal/models/vlm/config_from_model_test.go create mode 100644 internal/utils/extraheaders.go create mode 100644 internal/utils/extraheaders_test.go diff --git a/frontend/src/api/initialization/index.ts b/frontend/src/api/initialization/index.ts index e079816b..e0c168be 100644 --- a/frontend/src/api/initialization/index.ts +++ b/frontend/src/api/initialization/index.ts @@ -250,13 +250,22 @@ export function getCurrentConfigByKB(kbId: string): Promise; + extraConfig?: Record; + interfaceType?: string; +} + // 检查远程API模型 export function checkRemoteModel(modelConfig: { modelName: string; baseUrl: string; apiKey?: string; provider?: string; -}): Promise<{ +} & BaseModelTestPayload): Promise<{ available: boolean; message?: string; }> { @@ -280,7 +289,7 @@ export function testEmbeddingModel(modelConfig: { apiKey?: string; dimension?: number; provider?: string; -}): Promise<{ available: boolean; message?: string; dimension?: number }> { +} & BaseModelTestPayload): Promise<{ available: boolean; message?: string; dimension?: number }> { return new Promise((resolve, reject) => { post('/api/v1/initialization/embedding/test', modelConfig) .then((response: any) => { @@ -298,7 +307,8 @@ export function checkRerankModel(modelConfig: { modelName: string; baseUrl: string; apiKey?: string; -}): Promise<{ + provider?: string; +} & BaseModelTestPayload): Promise<{ available: boolean; message?: string; }> { @@ -319,7 +329,8 @@ export function checkASRModel(modelConfig: { modelName: string; baseUrl: string; apiKey?: string; -}): Promise<{ + provider?: string; +} & BaseModelTestPayload): Promise<{ available: boolean; message?: string; }> { diff --git a/frontend/src/api/model/index.ts b/frontend/src/api/model/index.ts index f7357ac6..c108de4e 100644 --- a/frontend/src/api/model/index.ts +++ b/frontend/src/api/model/index.ts @@ -22,6 +22,9 @@ export interface ModelConfig { interface_type?: 'ollama' | 'openai'; // VLLM专用 parameter_size?: string; // Ollama模型参数大小 (e.g., "7B", "13B", "70B") extra_config?: Record; // Provider-specific configuration + // 自定义 HTTP 请求头(类似 Python OpenAI SDK 的 extra_headers), + // 会在调用远程模型 API 时附加到每个请求上。Authorization、Content-Type 等保留头会被忽略。 + custom_headers?: Record; supports_vision?: boolean; // Whether the model accepts image/multimodal input }; is_default?: boolean; diff --git a/frontend/src/components/ModelEditorDialog.vue b/frontend/src/components/ModelEditorDialog.vue index 75fcba2e..3d8ef378 100644 --- a/frontend/src/components/ModelEditorDialog.vue +++ b/frontend/src/components/ModelEditorDialog.vue @@ -218,6 +218,46 @@ /> + +
+
+ + + + {{ $t('model.editor.customHeadersAdd') }} + +
+

{{ $t('model.editor.customHeadersDesc') }}

+
+
+ + + + + +
+
+
+
@@ -314,6 +354,11 @@ import { getWeKnoraCloudStatus } from '@/api/model' import { useI18n } from 'vue-i18n' import { useUIStore } from '@/stores/ui' +interface CustomHeaderItem { + key: string + value: string +} + interface ModelFormData { id: string name: string @@ -326,6 +371,8 @@ interface ModelFormData { interfaceType?: 'ollama' | 'openai' isDefault: boolean supportsVision?: boolean + // 自定义 HTTP 请求头(类似 OpenAI Python SDK 的 extra_headers) + customHeaders?: CustomHeaderItem[] } interface Props { @@ -571,7 +618,8 @@ const formData = ref({ dimension: undefined, interfaceType: 'ollama', isDefault: false, - supportsVision: false + supportsVision: false, + customHeaders: [] }) const rules = computed(() => ({ @@ -693,7 +741,12 @@ watch(() => props.visible, (val) => { loadProviders() if (props.modelData) { - formData.value = { ...props.modelData } + formData.value = { + ...props.modelData, + customHeaders: Array.isArray(props.modelData.customHeaders) + ? props.modelData.customHeaders.map(h => ({ key: h.key, value: h.value })) + : [] + } } else { resetForm() } @@ -726,7 +779,8 @@ const resetForm = () => { dimension: undefined, // 默认不填,让用户手动输入或通过检测按钮获取 interfaceType: undefined, isDefault: false, - supportsVision: false + supportsVision: false, + customHeaders: [] } modelChecked.value = false modelAvailable.value = false @@ -765,6 +819,19 @@ const generateId = () => { return `model_${Date.now()}_${Math.random().toString(36).substr(2, 9)}` } +// 自定义 HTTP Header 编辑 +const addCustomHeader = () => { + if (!Array.isArray(formData.value.customHeaders)) { + formData.value.customHeaders = [] + } + formData.value.customHeaders.push({ key: '', value: '' }) +} + +const removeCustomHeader = (idx: number) => { + if (!Array.isArray(formData.value.customHeaders)) return + formData.value.customHeaders.splice(idx, 1) +} + // 过滤后的模型列表 const filteredOllamaModels = computed(() => { if (!searchKeyword.value) return ollamaModelList.value @@ -900,7 +967,23 @@ const checkRemoteAPI = async () => { try { let result: any - + + // 把表单里 Key-Value 数组形式的自定义 Header 转成后端期望的 map。 + // 跟 ModelSettings.vue 保存时一致,空行自动丢弃,保证测试连接与真正保存后的 + // 生产调用使用完全相同的 Header 集合。 + const customHeaders: Record = {} + if (Array.isArray(formData.value.customHeaders)) { + for (const item of formData.value.customHeaders) { + const key = (item?.key ?? '').trim() + const value = (item?.value ?? '').trim() + if (key && value) customHeaders[key] = value + } + } + // 只在非空时带上字段,避免在 URL query / 日志里出现空对象 + const headerPayload = Object.keys(customHeaders).length > 0 + ? { customHeaders } + : {} + // 根据模型类型调用不同的校验接口 switch (props.modelType) { case 'chat': @@ -909,7 +992,8 @@ const checkRemoteAPI = async () => { modelName: formData.value.modelName, baseUrl: formData.value.baseUrl, apiKey: formData.value.apiKey || '', - provider: formData.value.provider + provider: formData.value.provider, + ...headerPayload, }) break @@ -921,7 +1005,8 @@ const checkRemoteAPI = async () => { baseUrl: formData.value.baseUrl, apiKey: formData.value.apiKey || '', dimension: formData.value.dimension, - provider: formData.value.provider + provider: formData.value.provider, + ...headerPayload, }) // 如果测试成功且返回了维度,自动填充 if (result.available && result.dimension) { @@ -935,7 +1020,9 @@ const checkRemoteAPI = async () => { result = await checkRerankModel({ modelName: formData.value.modelName, baseUrl: formData.value.baseUrl, - apiKey: formData.value.apiKey || '' + apiKey: formData.value.apiKey || '', + provider: formData.value.provider, + ...headerPayload, }) break @@ -946,7 +1033,8 @@ const checkRemoteAPI = async () => { modelName: formData.value.modelName, baseUrl: formData.value.baseUrl, apiKey: formData.value.apiKey || '', - provider: formData.value.provider + provider: formData.value.provider, + ...headerPayload, }) break @@ -955,7 +1043,9 @@ const checkRemoteAPI = async () => { result = await checkASRModel({ modelName: formData.value.modelName, baseUrl: formData.value.baseUrl, - apiKey: formData.value.apiKey || '' + apiKey: formData.value.apiKey || '', + provider: formData.value.provider, + ...headerPayload, }) break @@ -1682,6 +1772,55 @@ const handleOverlayMouseUp = () => { } } +// 自定义 HTTP Header 区域 +.custom-headers-header { + display: flex; + align-items: center; + justify-content: space-between; + margin-bottom: 6px; +} + +.custom-headers-desc { + margin: 0 0 10px 0; + font-size: 12px; + line-height: 1.5; + color: var(--td-text-color-placeholder); +} + +.custom-headers-list { + display: flex; + flex-direction: column; + gap: 8px; +} + +.custom-header-row { + display: flex; + align-items: center; + gap: 8px; + + .custom-header-key { + flex: 0 0 38%; + } + + .custom-header-value { + flex: 1; + } + + :deep(.t-button) { + flex-shrink: 0; + width: 32px; + height: 32px; + padding: 0; + } +} + +.form-desc { + margin: 4px 0 0 0; + font-size: 12px; + line-height: 1.5; + color: var(--td-text-color-placeholder); +} + // Ollama不可用提示样式 .ollama-unavailable-tip { display: flex; diff --git a/frontend/src/i18n/locales/en-US.ts b/frontend/src/i18n/locales/en-US.ts index 480c6bfc..9c4823b8 100755 --- a/frontend/src/i18n/locales/en-US.ts +++ b/frontend/src/i18n/locales/en-US.ts @@ -2349,6 +2349,11 @@ export default { baseUrlPlaceholderAsr: 'e.g. https://api.openai.com/v1', apiKeyOptional: 'API Key (optional)', apiKeyPlaceholder: 'Enter API Key', + customHeadersLabel: 'Custom Request Headers (optional)', + customHeadersDesc: 'Extra HTTP headers appended to requests to the remote model API (e.g. for enterprise gateway auth or tracing). Reserved headers like Authorization / Content-Type are ignored.', + customHeadersAdd: 'Add Header', + customHeadersKeyPlaceholder: 'Header name', + customHeadersValuePlaceholder: 'Header value', connectionTest: 'Connection Test', testing: 'Testing...', testConnection: 'Test Connection', diff --git a/frontend/src/i18n/locales/ko-KR.ts b/frontend/src/i18n/locales/ko-KR.ts index 51cf6241..994ed95a 100755 --- a/frontend/src/i18n/locales/ko-KR.ts +++ b/frontend/src/i18n/locales/ko-KR.ts @@ -1686,6 +1686,11 @@ export default { baseUrlPlaceholderAsr: "예: https://api.openai.com/v1", apiKeyOptional: "API 키 (선택)", apiKeyPlaceholder: "API 키 입력", + customHeadersLabel: "사용자 정의 요청 헤더 (선택)", + customHeadersDesc: "원격 모델 API 호출 시 추가되는 HTTP 헤더입니다. 기업 게이트웨이 인증이나 추적 등에 사용할 수 있으며, Authorization / Content-Type 같은 예약 헤더는 무시됩니다.", + customHeadersAdd: "헤더 추가", + customHeadersKeyPlaceholder: "헤더 이름", + customHeadersValuePlaceholder: "헤더 값", connectionTest: "연결 테스트", testing: "테스트 중...", testConnection: "연결 테스트", diff --git a/frontend/src/i18n/locales/ru-RU.ts b/frontend/src/i18n/locales/ru-RU.ts index 00758e0e..b5f4f7fc 100755 --- a/frontend/src/i18n/locales/ru-RU.ts +++ b/frontend/src/i18n/locales/ru-RU.ts @@ -1527,6 +1527,11 @@ export default { baseUrlPlaceholderAsr: 'например: https://api.openai.com/v1', apiKeyOptional: 'API Key (опционально)', apiKeyPlaceholder: 'Введите API Key', + customHeadersLabel: 'Пользовательские заголовки запроса (опционально)', + customHeadersDesc: 'Дополнительные HTTP-заголовки, добавляемые к запросам к удалённому API модели (например, для авторизации корпоративного шлюза или трассировки). Зарезервированные заголовки, такие как Authorization / Content-Type, игнорируются.', + customHeadersAdd: 'Добавить заголовок', + customHeadersKeyPlaceholder: 'Имя заголовка', + customHeadersValuePlaceholder: 'Значение заголовка', connectionTest: 'Проверка соединения', testing: 'Проверка...', testConnection: 'Проверить соединение', diff --git a/frontend/src/i18n/locales/zh-CN.ts b/frontend/src/i18n/locales/zh-CN.ts index 46ad3318..0e3205b7 100755 --- a/frontend/src/i18n/locales/zh-CN.ts +++ b/frontend/src/i18n/locales/zh-CN.ts @@ -1698,6 +1698,11 @@ export default { baseUrlPlaceholderAsr: "例如:https://api.openai.com/v1", apiKeyOptional: "API Key(可选)", apiKeyPlaceholder: "输入 API Key", + customHeadersLabel: "自定义请求头(可选)", + customHeadersDesc: "调用远程模型 API 时附加的 HTTP 请求头,常用于企业网关鉴权、链路追踪等场景;Authorization、Content-Type 等保留头会被自动忽略。", + customHeadersAdd: "添加请求头", + customHeadersKeyPlaceholder: "Header 名称", + customHeadersValuePlaceholder: "Header 值", connectionTest: "连接测试", testing: "测试中...", testConnection: "测试连接", diff --git a/frontend/src/views/settings/ModelSettings.vue b/frontend/src/views/settings/ModelSettings.vue index 94ae37d4..408a2dd4 100644 --- a/frontend/src/views/settings/ModelSettings.vue +++ b/frontend/src/views/settings/ModelSettings.vue @@ -350,7 +350,11 @@ function convertToLegacyFormat(model: ModelConfig) { provider: model.parameters.provider || '', // 添加 provider 字段 dimension: model.parameters.embedding_parameters?.dimension, isBuiltin: model.is_builtin || false, - supportsVision: model.parameters.supports_vision || false + supportsVision: model.parameters.supports_vision || false, + // 将后端 map 形式转换为前端可编辑的数组形式 + customHeaders: model.parameters.custom_headers + ? Object.entries(model.parameters.custom_headers).map(([key, value]) => ({ key, value: String(value) })) + : [] } } @@ -426,6 +430,18 @@ const handleModelSave = async (modelData: any) => { } } + // 将前端 Key-Value 数组形式的自定义 Header 转换成后端期望的 map + const customHeadersMap: Record = {} + if (Array.isArray(modelData.customHeaders)) { + for (const item of modelData.customHeaders) { + const key = (item?.key ?? '').trim() + const value = (item?.value ?? '').trim() + if (key && value) { + customHeadersMap[key] = value + } + } + } + // 将前端格式转换为后端格式 const apiModelData: ModelConfig = { name: modelData.modelName.trim(), // 使用 modelName 作为 name,并去除首尾空格 @@ -436,6 +452,7 @@ const handleModelSave = async (modelData: any) => { base_url: modelData.baseUrl?.trim() || '', api_key: modelData.apiKey?.trim() || '', provider: modelData.provider || '', // 添加 provider 字段 + ...(Object.keys(customHeadersMap).length > 0 ? { custom_headers: customHeadersMap } : {}), ...(currentModelType.value === 'embedding' && modelData.dimension ? { embedding_parameters: { dimension: modelData.dimension, diff --git a/internal/application/service/model.go b/internal/application/service/model.go index da42d58d..bd81799b 100644 --- a/internal/application/service/model.go +++ b/internal/application/service/model.go @@ -298,20 +298,7 @@ func (s *modelService) GetEmbeddingModel(ctx context.Context, modelId string) (e appID, appSecret := s.resolveWeKnoraCloudCredentials(ctx, &model.Parameters) - // Initialize the embedder with model configuration - embedder, err := embedding.NewEmbedder(embedding.Config{ - Source: model.Source, - BaseURL: model.Parameters.BaseURL, - APIKey: model.Parameters.APIKey, - ModelID: model.ID, - ModelName: model.Name, - Dimensions: model.Parameters.EmbeddingParameters.Dimension, - TruncatePromptTokens: model.Parameters.EmbeddingParameters.TruncatePromptTokens, - Provider: model.Parameters.Provider, - ExtraConfig: model.Parameters.ExtraConfig, - AppID: appID, - AppSecret: appSecret, - }, s.pooler, s.ollamaService) + embedder, err := embedding.NewEmbedder(embedding.ConfigFromModel(model, appID, appSecret), s.pooler, s.ollamaService) if err != nil { logger.ErrorWithFields(ctx, err, map[string]interface{}{ "model_id": model.ID, @@ -358,20 +345,7 @@ func (s *modelService) GetEmbeddingModelForTenant(ctx context.Context, modelId s appID, appSecret := s.resolveWeKnoraCloudCredentials(ctx, &model.Parameters) - // Initialize the embedder with model configuration - embedder, err := embedding.NewEmbedder(embedding.Config{ - Source: model.Source, - BaseURL: model.Parameters.BaseURL, - APIKey: model.Parameters.APIKey, - ModelID: model.ID, - ModelName: model.Name, - Dimensions: model.Parameters.EmbeddingParameters.Dimension, - TruncatePromptTokens: model.Parameters.EmbeddingParameters.TruncatePromptTokens, - Provider: model.Parameters.Provider, - ExtraConfig: model.Parameters.ExtraConfig, - AppID: appID, - AppSecret: appSecret, - }, s.pooler, s.ollamaService) + embedder, err := embedding.NewEmbedder(embedding.ConfigFromModel(model, appID, appSecret), s.pooler, s.ollamaService) if err != nil { logger.ErrorWithFields(ctx, err, map[string]interface{}{ "model_id": model.ID, @@ -401,18 +375,7 @@ func (s *modelService) GetRerankModel(ctx context.Context, modelId string) (rera appID, appSecret := s.resolveWeKnoraCloudCredentials(ctx, &model.Parameters) - // Initialize the reranker with model configuration - reranker, err := rerank.NewReranker(&rerank.RerankerConfig{ - ModelID: model.ID, - APIKey: model.Parameters.APIKey, - BaseURL: model.Parameters.BaseURL, - ModelName: model.Name, - Source: model.Source, - Provider: model.Parameters.Provider, - ExtraConfig: model.Parameters.ExtraConfig, - AppID: appID, - AppSecret: appSecret, - }) + reranker, err := rerank.NewReranker(rerank.ConfigFromModel(model, appID, appSecret)) if err != nil { logger.ErrorWithFields(ctx, err, map[string]interface{}{ "model_id": model.ID, @@ -455,18 +418,7 @@ func (s *modelService) GetChatModel(ctx context.Context, modelId string) (chat.C appID, appSecret := s.resolveWeKnoraCloudCredentials(ctx, &model.Parameters) - // Initialize the chat model with model configuration - chatModel, err := chat.NewChat(&chat.ChatConfig{ - ModelID: model.ID, - APIKey: model.Parameters.APIKey, - BaseURL: model.Parameters.BaseURL, - ModelName: model.Name, - Source: model.Source, - Provider: model.Parameters.Provider, - ExtraConfig: model.Parameters.ExtraConfig, - AppID: appID, - AppSecret: appSecret, - }, s.ollamaService) + chatModel, err := chat.NewChat(chat.ConfigFromModel(model, appID, appSecret), s.ollamaService) if err != nil { logger.ErrorWithFields(ctx, err, map[string]interface{}{ "model_id": model.ID, @@ -501,29 +453,9 @@ func (s *modelService) GetVLMModel(ctx context.Context, modelId string) (vlm.VLM logger.Infof(ctx, "Getting VLM model: %s, source: %s", model.Name, model.Source) - ifType := model.Parameters.InterfaceType - if ifType == "" { - if model.Source == types.ModelSourceLocal { - ifType = "ollama" - } else { - ifType = "openai" - } - } - appID, appSecret := s.resolveWeKnoraCloudCredentials(ctx, &model.Parameters) - vlmModel, err := vlm.NewVLM(&vlm.Config{ - ModelID: model.ID, - APIKey: model.Parameters.APIKey, - BaseURL: model.Parameters.BaseURL, - ModelName: model.Name, - Source: model.Source, - InterfaceType: ifType, - Provider: model.Parameters.Provider, - Extra: stringMapToAnyMap(model.Parameters.ExtraConfig), - AppID: appID, - AppSecret: appSecret, - }, s.ollamaService) + vlmModel, err := vlm.NewVLM(vlm.ConfigFromModel(model, appID, appSecret), s.ollamaService) if err != nil { logger.ErrorWithFields(ctx, err, map[string]interface{}{ "model_id": model.ID, @@ -561,13 +493,7 @@ func (s *modelService) GetASRModel(ctx context.Context, modelId string) (asr.ASR logger.Infof(ctx, "Getting ASR model: %s, source: %s", model.Name, model.Source) - sttModel, err := asr.NewASR(&asr.Config{ - ModelID: model.ID, - APIKey: model.Parameters.APIKey, - BaseURL: model.Parameters.BaseURL, - ModelName: model.Name, - Source: model.Source, - }) + sttModel, err := asr.NewASR(asr.ConfigFromModel(model)) if err != nil { logger.ErrorWithFields(ctx, err, map[string]interface{}{ "model_id": model.ID, @@ -578,14 +504,3 @@ func (s *modelService) GetASRModel(ctx context.Context, modelId string) (asr.ASR return sttModel, nil } - -func stringMapToAnyMap(m map[string]string) map[string]any { - if m == nil { - return nil - } - result := make(map[string]any, len(m)) - for k, v := range m { - result[k] = v - } - return result -} diff --git a/internal/handler/initialization.go b/internal/handler/initialization.go index cc2ab671..5356e865 100644 --- a/internal/handler/initialization.go +++ b/internal/handler/initialization.go @@ -1468,12 +1468,76 @@ func (h *InitializationHandler) buildConfigResponse(ctx context.Context, models return config } -// RemoteModelCheckRequest 远程模型检查请求结构 -type RemoteModelCheckRequest struct { - ModelName string `json:"modelName" binding:"required"` - BaseURL string `json:"baseUrl" binding:"required"` - APIKey string `json:"apiKey"` - Provider string `json:"provider"` +// ModelTestRequest 统一的"测试连接"请求体。 +// +// 四种模型(chat/embedding/rerank/asr)的测试接口共享同一份结构,以便: +// - 前端只需维护一份表单 → 后端映射。 +// - 后端可以直接把请求转成 *types.Model,再调用各包的 ConfigFromModel, +// 与生产路径(service.modelService.GetXxxModel)走完全相同的装配流程, +// 彻底消除过去每个测试端点手工拼 Config 的样板代码。 +// +// 所有 provider/model 通用字段都在这里集中声明;若未来新增字段(比如现在的 +// custom_headers),只需改一处,生产路径和测试路径会同时生效。 +type ModelTestRequest struct { + Source string `json:"source"` // 为空时按需默认为 "remote" + ModelName string `json:"modelName" binding:"required"` + BaseURL string `json:"baseUrl"` + APIKey string `json:"apiKey"` + Provider string `json:"provider"` + InterfaceType string `json:"interfaceType,omitempty"` + Dimension int `json:"dimension,omitempty"` + CustomHeaders map[string]string `json:"customHeaders,omitempty"` + ExtraConfig map[string]string `json:"extraConfig,omitempty"` +} + +// RemoteModelCheckRequest 兼容旧 swagger 定义。 +// +// Deprecated: 保留是为了不破坏已生成的 API 文档,新代码请直接使用 ModelTestRequest。 +type RemoteModelCheckRequest = ModelTestRequest + +// buildTestModel 把测试连接请求转成一个临时的 *types.Model(不落库), +// 供 ConfigFromModel 使用。source 为空时按 defaultSource 兜底(chat/rerank/asr +// 默认 remote,embedding 会根据前端传入的 source 决定)。 +func (h *InitializationHandler) buildTestModel( + req *ModelTestRequest, modelType types.ModelType, defaultSource types.ModelSource, +) *types.Model { + source := types.ModelSource(strings.ToLower(req.Source)) + if source == "" { + source = defaultSource + } + return &types.Model{ + Name: req.ModelName, + Type: modelType, + Source: source, + Parameters: types.ModelParameters{ + BaseURL: req.BaseURL, + APIKey: req.APIKey, + Provider: req.Provider, + InterfaceType: req.InterfaceType, + ExtraConfig: req.ExtraConfig, + CustomHeaders: req.CustomHeaders, + EmbeddingParameters: types.EmbeddingParameters{ + Dimension: req.Dimension, + TruncatePromptTokens: 256, + }, + }, + } +} + +// resolveTenantWeKnoraCloudCreds 从当前租户上下文里取出 WeKnoraCloud 凭证, +// 供测试连接端点补齐 appID/appSecret。与 service.resolveWeKnoraCloudCredentials +// 对应,但因为 handler 还没有被注入 tenantService(历史原因),暂时从 +// TenantInfoFromContext 读取,等效果相同。 +func (h *InitializationHandler) resolveTenantWeKnoraCloudCreds(ctx context.Context) (string, string, bool) { + tenantInfo, ok := types.TenantInfoFromContext(ctx) + if !ok { + return "", "", false + } + creds := tenantInfo.Credentials.GetWeKnoraCloud() + if creds == nil { + return "", "", true + } + return creds.AppID, creds.AppSecret, true } // CheckRemoteModel godoc @@ -1493,54 +1557,33 @@ func (h *InitializationHandler) CheckRemoteModel(c *gin.Context) { logger.Info(ctx, "Checking remote model connection") - var req RemoteModelCheckRequest + var req ModelTestRequest if err := c.ShouldBindJSON(&req); err != nil { logger.Error(ctx, "Failed to parse remote model check request", err) c.Error(errors.NewBadRequestError(err.Error())) return } - // 验证请求参数 if req.ModelName == "" || req.BaseURL == "" { logger.Error(ctx, "Model name and base URL are required") c.Error(errors.NewBadRequestError("模型名称和Base URL不能为空")) return } - // SSRF validation if err := utils.ValidateURLForSSRF(req.BaseURL); err != nil { logger.Warnf(ctx, "SSRF validation failed for remote model BaseURL: %v", err) c.Error(errors.NewBadRequestError(fmt.Sprintf("Base URL 未通过安全校验: %v", err))) return } - tenantInfo, ok := types.TenantInfoFromContext(ctx) + appID, appSecret, ok := h.resolveTenantWeKnoraCloudCreds(ctx) if !ok { logger.Error(ctx, "Tenant info not found") c.Error(errors.NewBadRequestError("租户信息未找到")) return } - var appID, appSecret string - if creds := tenantInfo.Credentials.GetWeKnoraCloud(); creds != nil { - appID = creds.AppID - appSecret = creds.AppSecret - } - // 创建模型配置进行测试 - modelConfig := &types.Model{ - Name: req.ModelName, - Source: "remote", - Parameters: types.ModelParameters{ - BaseURL: req.BaseURL, - APIKey: req.APIKey, - Provider: req.Provider, - AppID: appID, - AppSecret: appSecret, - }, - Type: "llm", // 默认类型,实际检查时不区分具体类型 - } - - // 检查远程模型连接 - available, message := h.checkRemoteModelConnection(ctx, modelConfig) + model := h.buildTestModel(&req, types.ModelTypeKnowledgeQA, types.ModelSourceRemote) + available, message := h.checkChatModelConnection(ctx, model, appID, appSecret) logger.Infof(ctx, "Remote model check completed, available: %v, message: %s", available, message) @@ -1570,22 +1613,16 @@ func (h *InitializationHandler) TestEmbeddingModel(c *gin.Context) { logger.Info(ctx, "Testing embedding model connectivity and functionality") - var req struct { - Source string `json:"source" binding:"required"` - ModelName string `json:"modelName" binding:"required"` - BaseURL string `json:"baseUrl"` - APIKey string `json:"apiKey"` - Dimension int `json:"dimension"` - Provider string `json:"provider"` - } - + var req ModelTestRequest if err := c.ShouldBindJSON(&req); err != nil { logger.Error(ctx, "Failed to parse embedding test request", err) c.Error(errors.NewBadRequestError(err.Error())) return } + if req.Source == "" { + req.Source = string(types.ModelSourceRemote) + } - // SSRF validation for embedding BaseURL if req.BaseURL != "" { if err := utils.ValidateURLForSSRF(req.BaseURL); err != nil { logger.Warnf(ctx, "SSRF validation failed for embedding BaseURL: %v", err) @@ -1594,7 +1631,7 @@ func (h *InitializationHandler) TestEmbeddingModel(c *gin.Context) { } } - // 检查是否是阿里云多模态 embedding 模型(暂不支持) + // 阿里云多模态 Embedding 模型暂不支持 if strings.ToLower(req.Provider) == "aliyun" { modelNameLower := strings.ToLower(req.ModelName) if strings.Contains(modelNameLower, "vision") || strings.Contains(modelNameLower, "multimodal") { @@ -1610,33 +1647,16 @@ func (h *InitializationHandler) TestEmbeddingModel(c *gin.Context) { return } } - tenantInfo, ok := types.TenantInfoFromContext(ctx) + + appID, appSecret, ok := h.resolveTenantWeKnoraCloudCreds(ctx) if !ok { logger.Error(ctx, "Tenant info not found") c.Error(errors.NewBadRequestError("租户信息未找到")) return } - var appID, appSecret string - if creds := tenantInfo.Credentials.GetWeKnoraCloud(); creds != nil { - appID = creds.AppID - appSecret = creds.AppSecret - } - // 构造 embedder 配置 - cfg := embedding.Config{ - Source: types.ModelSource(strings.ToLower(req.Source)), - BaseURL: req.BaseURL, - ModelName: req.ModelName, - APIKey: req.APIKey, - TruncatePromptTokens: 256, - Dimensions: req.Dimension, - ModelID: "", - Provider: req.Provider, - AppID: appID, - AppSecret: appSecret, - } - - emb, err := embedding.NewEmbedder(cfg, h.pooler, h.ollamaService) + model := h.buildTestModel(&req, types.ModelTypeEmbedding, types.ModelSourceRemote) + emb, err := embedding.NewEmbedder(embedding.ConfigFromModel(model, appID, appSecret), h.pooler, h.ollamaService) if err != nil { logger.ErrorWithFields(ctx, err, map[string]interface{}{"model": utils.SanitizeForLog(req.ModelName)}) c.JSON(http.StatusOK, gin.H{ @@ -1646,11 +1666,9 @@ func (h *InitializationHandler) TestEmbeddingModel(c *gin.Context) { return } - // 执行一次最小化 embedding 调用 - sample := "hello" - vec, err := emb.Embed(ctx, sample) + vec, err := emb.Embed(ctx, "hello") if err != nil { - logger.Error(ctx, "Failed to create embedder", err) + logger.Error(ctx, "Failed to call embedder", err) c.JSON(http.StatusOK, gin.H{ "success": true, "data": gin.H{`available`: false, `message`: fmt.Sprintf("调用Embedding失败: %v", err), `dimension`: 0}, @@ -1665,44 +1683,23 @@ func (h *InitializationHandler) TestEmbeddingModel(c *gin.Context) { }) } -// checkRemoteModelConnection 检查远程模型连接的内部方法 -func (h *InitializationHandler) checkRemoteModelConnection(ctx context.Context, - model *types.Model, +// checkChatModelConnection 使用 chat 模块做一次最小化调用来测试连通性与鉴权。 +// 与生产路径走完全相同的 ConfigFromModel → NewChat 流程,因此 CustomHeaders、 +// ExtraConfig、Provider 等字段都会被正确透传。 +func (h *InitializationHandler) checkChatModelConnection( + ctx context.Context, model *types.Model, appID, appSecret string, ) (bool, string) { - // 使用 models/chat 进行连接检查 - // 创建聊天配置 - chatConfig := &chat.ChatConfig{ - Source: types.ModelSourceRemote, - BaseURL: model.Parameters.BaseURL, - ModelName: model.Name, - APIKey: model.Parameters.APIKey, - ModelID: model.Name, - Provider: model.Parameters.Provider, - AppID: model.Parameters.AppID, - AppSecret: model.Parameters.AppSecret, - } - - // 创建聊天实例 - chatInstance, err := chat.NewChat(chatConfig, h.ollamaService) + chatInstance, err := chat.NewChat(chat.ConfigFromModel(model, appID, appSecret), h.ollamaService) if err != nil { return false, fmt.Sprintf("创建聊天实例失败: %v", err) } - // 构造测试消息 - testMessages := []chat.Message{ - { - Role: "user", - Content: "test", - }, - } - - // 构造测试选项 + testMessages := []chat.Message{{Role: "user", Content: "test"}} testOptions := &chat.ChatOptions{ MaxTokens: 1, Thinking: &[]bool{false}[0], // for dashscope.aliyuncs qwen3-32b } - // 使用聊天实例进行测试 _, err = chatInstance.Chat(ctx, testMessages, testOptions) if err != nil { errMsg := err.Error() @@ -1728,54 +1725,24 @@ func (h *InitializationHandler) checkRemoteModelConnection(ctx context.Context, return true, "连接正常,模型可用" } -// checkRerankModelConnection 检查Rerank模型连接和功能的内部方法 -func (h *InitializationHandler) checkRerankModelConnection(ctx context.Context, - modelName, baseURL, apiKey string, +// checkRerankModelConnection 使用 rerank 模块做一次最小化调用来测试连通性与鉴权。 +// 与生产路径共用 ConfigFromModel,所有字段(CustomHeaders 等)都透传。 +func (h *InitializationHandler) checkRerankModelConnection( + ctx context.Context, model *types.Model, appID, appSecret string, ) (bool, string) { - // 创建Reranker配置 - config := &rerank.RerankerConfig{ - APIKey: apiKey, - BaseURL: baseURL, - ModelName: modelName, - Source: types.ModelSourceRemote, // 默认值,实际会根据URL判断 - } - tenantInfo, ok := types.TenantInfoFromContext(ctx) - if !ok { - logger.Error(ctx, "Tenant info not found") - return false, "租户信息未找到" - } - var appID, appSecret string - if creds := tenantInfo.Credentials.GetWeKnoraCloud(); creds != nil { - appID = creds.AppID - appSecret = creds.AppSecret - } - config.AppID = appID - config.AppSecret = appSecret - - // 创建Reranker实例 - reranker, err := rerank.NewReranker(config) + reranker, err := rerank.NewReranker(rerank.ConfigFromModel(model, appID, appSecret)) if err != nil { return false, fmt.Sprintf("创建Reranker失败: %v", err) } - // 简化的测试数据 - testQuery := "ping" - testDocuments := []string{ - "pong", - } - - // 使用Reranker进行测试 - results, err := reranker.Rerank(ctx, testQuery, testDocuments) + results, err := reranker.Rerank(ctx, "ping", []string{"pong"}) if err != nil { return false, fmt.Sprintf("重排测试失败: %v", err) } - - // 检查结果 if len(results) > 0 { return true, fmt.Sprintf("重排功能正常,返回%d个结果", len(results)) - } else { - return false, "重排接口连接成功,但未返回重排结果" } + return false, "重排接口连接成功,但未返回重排结果" } // CheckRerankModel godoc @@ -1795,36 +1762,34 @@ func (h *InitializationHandler) CheckRerankModel(c *gin.Context) { logger.Info(ctx, "Checking rerank model connection and functionality") - var req struct { - ModelName string `json:"modelName" binding:"required"` - BaseURL string `json:"baseUrl" binding:"required"` - APIKey string `json:"apiKey"` - } - + var req ModelTestRequest if err := c.ShouldBindJSON(&req); err != nil { logger.Error(ctx, "Failed to parse rerank model check request", err) c.Error(errors.NewBadRequestError(err.Error())) return } - // 验证请求参数 if req.ModelName == "" || req.BaseURL == "" { logger.Error(ctx, "Model name and base URL are required") c.Error(errors.NewBadRequestError("模型名称和Base URL不能为空")) return } - // SSRF validation if err := utils.ValidateURLForSSRF(req.BaseURL); err != nil { logger.Warnf(ctx, "SSRF validation failed for rerank BaseURL: %v", err) c.Error(errors.NewBadRequestError(fmt.Sprintf("Base URL 未通过安全校验: %v", err))) return } - // 检查Rerank模型连接和功能 - available, message := h.checkRerankModelConnection( - ctx, req.ModelName, req.BaseURL, req.APIKey, - ) + appID, appSecret, ok := h.resolveTenantWeKnoraCloudCreds(ctx) + if !ok { + logger.Error(ctx, "Tenant info not found") + c.Error(errors.NewBadRequestError("租户信息未找到")) + return + } + + model := h.buildTestModel(&req, types.ModelTypeRerank, types.ModelSourceRemote) + available, message := h.checkRerankModelConnection(ctx, model, appID, appSecret) logger.Infof(ctx, "Rerank model check completed, available: %v, message: %s", available, message) @@ -1854,12 +1819,7 @@ func (h *InitializationHandler) CheckASRModel(c *gin.Context) { logger.Info(ctx, "Checking ASR model connection") - var req struct { - ModelName string `json:"modelName" binding:"required"` - BaseURL string `json:"baseUrl" binding:"required"` - APIKey string `json:"apiKey"` - } - + var req ModelTestRequest if err := c.ShouldBindJSON(&req); err != nil { logger.Error(ctx, "Failed to parse ASR model check request", err) c.Error(errors.NewBadRequestError(err.Error())) @@ -1872,20 +1832,16 @@ func (h *InitializationHandler) CheckASRModel(c *gin.Context) { return } - // SSRF validation if err := utils.ValidateURLForSSRF(req.BaseURL); err != nil { logger.Warnf(ctx, "SSRF validation failed for ASR BaseURL: %v", err) c.Error(errors.NewBadRequestError(fmt.Sprintf("Base URL 未通过安全校验: %v", err))) return } - // 使用 ASR 模块测试连接:发送一段极短的静默 WAV 音频 - asrInstance, err := asr.NewASR(&asr.Config{ - BaseURL: req.BaseURL, - ModelName: req.ModelName, - APIKey: req.APIKey, - Source: "remote", - }) + // 用统一构造器生成测试用 *types.Model(ASR 不涉及 WeKnoraCloud 凭证), + // 发送一段极短的静默 WAV 音频验证 /v1/audio/transcriptions 端点可达。 + model := h.buildTestModel(&req, types.ModelTypeASR, types.ModelSourceRemote) + asrInstance, err := asr.NewASR(asr.ConfigFromModel(model)) if err != nil { logger.Errorf(ctx, "Failed to create ASR instance for check: %v", err) c.JSON(http.StatusOK, gin.H{ diff --git a/internal/models/asr/asr.go b/internal/models/asr/asr.go index 02d70e53..5f7c17a7 100644 --- a/internal/models/asr/asr.go +++ b/internal/models/asr/asr.go @@ -36,6 +36,25 @@ type Config struct { APIKey string ModelID string Language string // optional: specify language for transcription + // CustomHeaders 允许在调用远程 API 时附加自定义 HTTP 请求头(类似 OpenAI Python SDK 的 extra_headers)。 + CustomHeaders map[string]string +} + +// ConfigFromModel 根据 types.Model 构造 asr.Config。 +// 生产路径(从 DB 拉起)和测试连接路径(临时表单)共享这份映射。 +// 当前 ASR 不涉及 WeKnoraCloud 凭证,所以签名不含 appID/appSecret。 +func ConfigFromModel(m *types.Model) *Config { + if m == nil { + return nil + } + return &Config{ + ModelID: m.ID, + APIKey: m.Parameters.APIKey, + BaseURL: m.Parameters.BaseURL, + ModelName: m.Name, + Source: m.Source, + CustomHeaders: m.Parameters.CustomHeaders, + } } // NewASR creates an ASR instance based on the provided configuration. diff --git a/internal/models/asr/config_from_model_test.go b/internal/models/asr/config_from_model_test.go new file mode 100644 index 00000000..7d2275ec --- /dev/null +++ b/internal/models/asr/config_from_model_test.go @@ -0,0 +1,36 @@ +package asr + +import ( + "testing" + + "github.com/Tencent/WeKnora/internal/types" +) + +func TestConfigFromModel(t *testing.T) { + m := &types.Model{ + ID: "asr-1", + Name: "whisper-1", + Source: types.ModelSourceRemote, + Parameters: types.ModelParameters{ + BaseURL: "https://api.example.com/v1", + APIKey: "sk", + CustomHeaders: map[string]string{"X": "y"}, + }, + } + cfg := ConfigFromModel(m) + if cfg == nil || cfg.ModelID != "asr-1" || cfg.ModelName != "whisper-1" { + t.Fatalf("identity mismatch: %+v", cfg) + } + if cfg.BaseURL != "https://api.example.com/v1" || cfg.APIKey != "sk" { + t.Errorf("connection fields mismatch: %+v", cfg) + } + if cfg.CustomHeaders["X"] != "y" { + t.Errorf("CustomHeaders not propagated: %+v", cfg.CustomHeaders) + } +} + +func TestConfigFromModel_Nil(t *testing.T) { + if got := ConfigFromModel(nil); got != nil { + t.Fatalf("expected nil, got %+v", got) + } +} diff --git a/internal/models/asr/openai.go b/internal/models/asr/openai.go index ea72756c..6244a5e5 100644 --- a/internal/models/asr/openai.go +++ b/internal/models/asr/openai.go @@ -11,6 +11,7 @@ import ( "time" "github.com/Tencent/WeKnora/internal/logger" + secutils "github.com/Tencent/WeKnora/internal/utils" openai "github.com/sashabaranov/go-openai" ) @@ -33,7 +34,14 @@ func NewOpenAIASR(config *Config) (*OpenAIASR, error) { if config.BaseURL != "" { apiCfg.BaseURL = config.BaseURL } - apiCfg.HTTPClient = &http.Client{Timeout: asrDefaultTimeout} + httpClient := &http.Client{Timeout: asrDefaultTimeout} + + // 注入用户自定义 HTTP header(类似 OpenAI Python SDK 的 extra_headers) + if len(config.CustomHeaders) > 0 { + apiCfg.HTTPClient = secutils.WrapHTTPClientWithHeaders(httpClient, config.CustomHeaders) + } else { + apiCfg.HTTPClient = httpClient + } return &OpenAIASR{ modelName: config.ModelName, diff --git a/internal/models/chat/chat.go b/internal/models/chat/chat.go index f4eb8529..478be8dc 100644 --- a/internal/models/chat/chat.go +++ b/internal/models/chat/chat.go @@ -101,8 +101,32 @@ type ChatConfig struct { ModelID string Provider string ExtraConfig map[string]string - AppID string - AppSecret string // 加密值,由工厂函数调用方传入,在 NewWeKnoraCloudChat 中使用前已解密 + // CustomHeaders 允许在调用远程 OpenAI 兼容 API 时附加自定义 HTTP 请求头(类似 OpenAI Python SDK 的 extra_headers)。 + CustomHeaders map[string]string + AppID string + AppSecret string // 加密值,由工厂函数调用方传入,在 NewWeKnoraCloudChat 中使用前已解密 +} + +// ConfigFromModel 根据 types.Model 构造 ChatConfig。 +// 保证生产路径(service 层根据 DB 中的模型配置拉起实例)和测试路径 +// (handler 层根据前端表单临时拉起实例)走完全相同的字段映射,避免重复样板。 +// appID / appSecret 是已经解密/解析好的 WeKnoraCloud 凭证,调用方负责传入。 +func ConfigFromModel(m *types.Model, appID, appSecret string) *ChatConfig { + if m == nil { + return nil + } + return &ChatConfig{ + ModelID: m.ID, + APIKey: m.Parameters.APIKey, + BaseURL: m.Parameters.BaseURL, + ModelName: m.Name, + Source: m.Source, + Provider: m.Parameters.Provider, + ExtraConfig: m.Parameters.ExtraConfig, + CustomHeaders: m.Parameters.CustomHeaders, + AppID: appID, + AppSecret: appSecret, + } } // NewChat 创建聊天实例 diff --git a/internal/models/chat/config_from_model_test.go b/internal/models/chat/config_from_model_test.go new file mode 100644 index 00000000..9e0e9fe5 --- /dev/null +++ b/internal/models/chat/config_from_model_test.go @@ -0,0 +1,59 @@ +package chat + +import ( + "testing" + + "github.com/Tencent/WeKnora/internal/types" +) + +// TestConfigFromModel 验证 ConfigFromModel 能把 types.Model 完整映射到 ChatConfig, +// 避免后续新加字段时只在生产路径或测试路径其中一侧更新导致功能漏配。 +func TestConfigFromModel(t *testing.T) { + m := &types.Model{ + ID: "model-123", + Name: "gpt-4o-mini", + Source: types.ModelSourceRemote, + Parameters: types.ModelParameters{ + BaseURL: "https://api.example.com/v1", + APIKey: "sk-xxx", + Provider: "openai", + ExtraConfig: map[string]string{ + "deployment": "prod", + }, + CustomHeaders: map[string]string{ + "X-Gateway-Token": "abc", + "X-Trace-ID": "t-1", + }, + }, + } + + cfg := ConfigFromModel(m, "app-id", "app-secret") + if cfg == nil { + t.Fatal("ConfigFromModel returned nil") + } + if cfg.ModelID != "model-123" || cfg.ModelName != "gpt-4o-mini" { + t.Errorf("model identity not propagated: %+v", cfg) + } + if cfg.BaseURL != "https://api.example.com/v1" || cfg.APIKey != "sk-xxx" { + t.Errorf("connection fields not propagated: %+v", cfg) + } + if cfg.Provider != "openai" { + t.Errorf("provider not propagated: %q", cfg.Provider) + } + if cfg.ExtraConfig["deployment"] != "prod" { + t.Errorf("ExtraConfig not propagated: %+v", cfg.ExtraConfig) + } + if cfg.CustomHeaders["X-Gateway-Token"] != "abc" || + cfg.CustomHeaders["X-Trace-ID"] != "t-1" { + t.Errorf("CustomHeaders not propagated: %+v", cfg.CustomHeaders) + } + if cfg.AppID != "app-id" || cfg.AppSecret != "app-secret" { + t.Errorf("cloud credentials not propagated: %+v", cfg) + } +} + +func TestConfigFromModel_Nil(t *testing.T) { + if got := ConfigFromModel(nil, "", ""); got != nil { + t.Fatalf("expected nil for nil model, got %+v", got) + } +} diff --git a/internal/models/chat/remote_api.go b/internal/models/chat/remote_api.go index 256de4c5..1f06facc 100644 --- a/internal/models/chat/remote_api.go +++ b/internal/models/chat/remote_api.go @@ -41,6 +41,8 @@ type RemoteAPIChat struct { provider provider.ProviderName appID string appSecret string + // customHeaders 为用户在模型配置中指定的自定义 HTTP 请求头(类似 OpenAI Python SDK 的 extra_headers)。 + customHeaders map[string]string // requestCustomizer 允许子类自定义请求 // 返回自定义请求体(如果为 nil 则使用标准请求)和是否需要使用原始 HTTP 请求 @@ -86,6 +88,17 @@ func NewRemoteAPIChat(chatConfig *ChatConfig) (*RemoteAPIChat, error) { } } + // 如果指定了 CustomHeaders,则给 SDK 使用的 HTTPClient 挂一层 RoundTripper, + // 在每个请求上自动注入这些 header(raw HTTP 路径会在发送前单独处理)。 + if len(chatConfig.CustomHeaders) > 0 { + if httpClient, ok := config.HTTPClient.(*http.Client); ok { + config.HTTPClient = secutils.WrapHTTPClientWithHeaders(httpClient, chatConfig.CustomHeaders) + } else { + // SDK 默认未显式设置时 HTTPClient 为 nil,此时构造一个新的注入了 header 的 client。 + config.HTTPClient = secutils.WrapHTTPClientWithHeaders(nil, chatConfig.CustomHeaders) + } + } + modelName := chatConfig.ModelName if chatConfig.ExtraConfig != nil { if override := strings.TrimSpace(chatConfig.ExtraConfig["remote_model_name"]); override != "" { @@ -102,14 +115,15 @@ func NewRemoteAPIChat(chatConfig *ChatConfig) (*RemoteAPIChat, error) { } return &RemoteAPIChat{ - modelName: modelName, - client: openai.NewClientWithConfig(config), - modelID: chatConfig.ModelID, - baseURL: chatConfig.BaseURL, - apiKey: apiKey, - provider: providerName, - appID: chatConfig.AppID, - appSecret: chatConfig.AppSecret, + modelName: modelName, + client: openai.NewClientWithConfig(config), + modelID: chatConfig.ModelID, + baseURL: chatConfig.BaseURL, + apiKey: apiKey, + provider: providerName, + appID: chatConfig.AppID, + appSecret: chatConfig.AppSecret, + customHeaders: chatConfig.CustomHeaders, }, nil } @@ -368,6 +382,9 @@ func (c *RemoteAPIChat) chatWithRawHTTP(ctx context.Context, endpoint string, cu httpReq.Header.Set("Authorization", "Bearer "+c.apiKey) } + // 注入用户自定义 header(保留头会在工具内部自动跳过) + secutils.ApplyCustomHeaders(httpReq, c.customHeaders) + logger.Infof(ctx, "[LLM Request] Remote HTTP, endpoint=%s, model=%s", endpoint, c.modelName) @@ -538,6 +555,9 @@ func (c *RemoteAPIChat) chatStreamWithRawHTTP(ctx context.Context, endpoint stri } httpReq.Header.Set("Accept", "text/event-stream") + // 注入用户自定义 header(保留头会在工具内部自动跳过) + secutils.ApplyCustomHeaders(httpReq, c.customHeaders) + resp, err := rawHTTPClient.Do(httpReq) if err != nil { return nil, fmt.Errorf("send request: %w", err) diff --git a/internal/models/embedding/aliyun.go b/internal/models/embedding/aliyun.go index 68246f14..0505fd21 100644 --- a/internal/models/embedding/aliyun.go +++ b/internal/models/embedding/aliyun.go @@ -11,6 +11,7 @@ import ( "time" "github.com/Tencent/WeKnora/internal/logger" + secutils "github.com/Tencent/WeKnora/internal/utils" ) const ( @@ -29,9 +30,15 @@ type AliyunEmbedder struct { httpClient *http.Client timeout time.Duration maxRetries int + customHeaders map[string]string EmbedderPooler } +// SetCustomHeaders 设置用户自定义 HTTP 请求头(类似 OpenAI Python SDK 的 extra_headers)。 +func (e *AliyunEmbedder) SetCustomHeaders(headers map[string]string) { + e.customHeaders = headers +} + // AliyunEmbedRequest represents an Aliyun DashScope multimodal embedding request type AliyunEmbedRequest struct { Model string `json:"model"` @@ -154,6 +161,7 @@ func (e *AliyunEmbedder) doRequestWithRetry(ctx context.Context, jsonData []byte } req.Header.Set("Content-Type", "application/json") req.Header.Set("Authorization", "Bearer "+e.apiKey) + secutils.ApplyCustomHeaders(req, e.customHeaders) resp, err = e.httpClient.Do(req) if err == nil { diff --git a/internal/models/embedding/azure_openai.go b/internal/models/embedding/azure_openai.go index f4dcbe3f..1a3d58bf 100644 --- a/internal/models/embedding/azure_openai.go +++ b/internal/models/embedding/azure_openai.go @@ -11,6 +11,7 @@ import ( "time" "github.com/Tencent/WeKnora/internal/logger" + secutils "github.com/Tencent/WeKnora/internal/utils" ) // AzureOpenAIEmbedder implements text vectorization using Azure OpenAI API @@ -24,9 +25,15 @@ type AzureOpenAIEmbedder struct { apiVersion string httpClient *http.Client maxRetries int + customHeaders map[string]string EmbedderPooler } +// SetCustomHeaders 设置用户自定义 HTTP 请求头(类似 OpenAI Python SDK 的 extra_headers)。 +func (e *AzureOpenAIEmbedder) SetCustomHeaders(headers map[string]string) { + e.customHeaders = headers +} + type azureOpenAIEmbedRequest struct { Model string `json:"model"` Input []string `json:"input"` @@ -157,6 +164,7 @@ func (e *AzureOpenAIEmbedder) doRequestWithRetry(ctx context.Context, jsonData [ } req.Header.Set("Content-Type", "application/json") req.Header.Set("api-key", e.apiKey) + secutils.ApplyCustomHeaders(req, e.customHeaders) resp, err = e.httpClient.Do(req) if err == nil { diff --git a/internal/models/embedding/config_from_model_test.go b/internal/models/embedding/config_from_model_test.go new file mode 100644 index 00000000..dc4a335e --- /dev/null +++ b/internal/models/embedding/config_from_model_test.go @@ -0,0 +1,43 @@ +package embedding + +import ( + "testing" + + "github.com/Tencent/WeKnora/internal/types" +) + +func TestConfigFromModel(t *testing.T) { + m := &types.Model{ + ID: "emb-1", + Name: "text-embedding-3-small", + Source: types.ModelSourceRemote, + Parameters: types.ModelParameters{ + BaseURL: "https://api.example.com/v1", + APIKey: "sk-xxx", + Provider: "openai", + EmbeddingParameters: types.EmbeddingParameters{ + Dimension: 1536, + TruncatePromptTokens: 512, + }, + ExtraConfig: map[string]string{"region": "us-east"}, + CustomHeaders: map[string]string{"X-Gateway": "g1"}, + }, + } + + cfg := ConfigFromModel(m, "app", "secret") + if cfg.ModelID != "emb-1" || cfg.ModelName != "text-embedding-3-small" { + t.Errorf("identity mismatch: %+v", cfg) + } + if cfg.Dimensions != 1536 || cfg.TruncatePromptTokens != 512 { + t.Errorf("embedding params mismatch: %+v", cfg) + } + if cfg.CustomHeaders["X-Gateway"] != "g1" { + t.Errorf("CustomHeaders not propagated: %+v", cfg.CustomHeaders) + } + if cfg.ExtraConfig["region"] != "us-east" { + t.Errorf("ExtraConfig not propagated: %+v", cfg.ExtraConfig) + } + if cfg.AppID != "app" || cfg.AppSecret != "secret" { + t.Errorf("cloud creds mismatch: %+v", cfg) + } +} diff --git a/internal/models/embedding/embedder.go b/internal/models/embedding/embedder.go index c56b1b1e..5f8aeec9 100644 --- a/internal/models/embedding/embedder.go +++ b/internal/models/embedding/embedder.go @@ -49,8 +49,33 @@ type Config struct { ModelID string `json:"model_id"` Provider string `json:"provider"` ExtraConfig map[string]string `json:"extra_config"` - AppID string - AppSecret string // 加密值,工厂函数调用方传入,使用前已解密 + // CustomHeaders 允许在调用远程 API 时附加自定义 HTTP 请求头(类似 OpenAI Python SDK 的 extra_headers)。 + CustomHeaders map[string]string `json:"custom_headers"` + AppID string + AppSecret string // 加密值,工厂函数调用方传入,使用前已解密 +} + +// ConfigFromModel 根据 types.Model 构造 embedding.Config。 +// 生产路径(从 DB 拉起)和测试连接路径(临时表单)共享这份映射。 +// appID / appSecret 是已解密的 WeKnoraCloud 凭证,调用方负责传入。 +func ConfigFromModel(m *types.Model, appID, appSecret string) Config { + if m == nil { + return Config{} + } + return Config{ + Source: m.Source, + BaseURL: m.Parameters.BaseURL, + APIKey: m.Parameters.APIKey, + ModelID: m.ID, + ModelName: m.Name, + Dimensions: m.Parameters.EmbeddingParameters.Dimension, + TruncatePromptTokens: m.Parameters.EmbeddingParameters.TruncatePromptTokens, + Provider: m.Parameters.Provider, + ExtraConfig: m.Parameters.ExtraConfig, + CustomHeaders: m.Parameters.CustomHeaders, + AppID: appID, + AppSecret: appSecret, + } } // NewEmbedder creates an embedder based on the configuration @@ -97,46 +122,62 @@ func newEmbedder(config Config, pooler EmbedderPooler, ollamaService *ollama.Oll baseURL = strings.Replace(baseURL, "/compatible-mode/v1", "", 1) baseURL = strings.Replace(baseURL, "/compatible-mode", "", 1) } - embedder, err = NewAliyunEmbedder(config.APIKey, + aliyunEmb, aErr := NewAliyunEmbedder(config.APIKey, baseURL, config.ModelName, config.TruncatePromptTokens, config.Dimensions, config.ModelID, pooler) + if aliyunEmb != nil { + aliyunEmb.SetCustomHeaders(config.CustomHeaders) + } + embedder, err = aliyunEmb, aErr } else { baseURL := config.BaseURL if baseURL == "" || !strings.Contains(baseURL, "/compatible-mode/") { baseURL = "https://dashscope.aliyuncs.com/compatible-mode/v1" } - embedder, err = NewOpenAIEmbedder(config.APIKey, + openaiEmb, oErr := NewOpenAIEmbedder(config.APIKey, baseURL, config.ModelName, config.TruncatePromptTokens, config.Dimensions, config.ModelID, pooler) + if openaiEmb != nil { + openaiEmb.SetCustomHeaders(config.CustomHeaders) + } + embedder, err = openaiEmb, oErr } return embedder, err case provider.ProviderVolcengine: // Volcengine Ark uses multimodal embedding API - embedder, err = NewVolcengineEmbedder(config.APIKey, + volcEmb, vErr := NewVolcengineEmbedder(config.APIKey, config.BaseURL, config.ModelName, config.TruncatePromptTokens, config.Dimensions, config.ModelID, pooler) + if volcEmb != nil { + volcEmb.SetCustomHeaders(config.CustomHeaders) + } + embedder, err = volcEmb, vErr return embedder, err case provider.ProviderJina: // Jina AI uses different API format (truncate instead of truncate_prompt_tokens) - embedder, err = NewJinaEmbedder(config.APIKey, + jinaEmb, jErr := NewJinaEmbedder(config.APIKey, config.BaseURL, config.ModelName, config.TruncatePromptTokens, config.Dimensions, config.ModelID, pooler) + if jinaEmb != nil { + jinaEmb.SetCustomHeaders(config.CustomHeaders) + } + embedder, err = jinaEmb, jErr return embedder, err case provider.ProviderAzureOpenAI: apiVersion := "2024-10-21" @@ -145,7 +186,7 @@ func newEmbedder(config Config, pooler EmbedderPooler, ollamaService *ollama.Oll apiVersion = v } } - embedder, err = NewAzureOpenAIEmbedder(config.APIKey, + azureEmb, azErr := NewAzureOpenAIEmbedder(config.APIKey, config.BaseURL, config.ModelName, config.TruncatePromptTokens, @@ -153,27 +194,39 @@ func newEmbedder(config Config, pooler EmbedderPooler, ollamaService *ollama.Oll config.ModelID, apiVersion, pooler) + if azureEmb != nil { + azureEmb.SetCustomHeaders(config.CustomHeaders) + } + embedder, err = azureEmb, azErr return embedder, err case provider.ProviderNvidia: - embedder, err = NewNvidiaEmbedder(config.APIKey, + nvEmb, nErr := NewNvidiaEmbedder(config.APIKey, config.BaseURL, config.ModelName, config.Dimensions, config.ModelID, pooler) + if nvEmb != nil { + nvEmb.SetCustomHeaders(config.CustomHeaders) + } + embedder, err = nvEmb, nErr return embedder, err case provider.ProviderWeKnoraCloud: embedder, err = NewWeKnoraCloudEmbedder(config) return embedder, err default: // Use OpenAI-compatible embedder for other providers - embedder, err = NewOpenAIEmbedder(config.APIKey, + openaiEmb, oErr := NewOpenAIEmbedder(config.APIKey, config.BaseURL, config.ModelName, config.TruncatePromptTokens, config.Dimensions, config.ModelID, pooler) + if openaiEmb != nil { + openaiEmb.SetCustomHeaders(config.CustomHeaders) + } + embedder, err = openaiEmb, oErr return embedder, err } default: diff --git a/internal/models/embedding/jina.go b/internal/models/embedding/jina.go index c19dd33a..78711db1 100644 --- a/internal/models/embedding/jina.go +++ b/internal/models/embedding/jina.go @@ -10,22 +10,29 @@ import ( "time" "github.com/Tencent/WeKnora/internal/logger" + secutils "github.com/Tencent/WeKnora/internal/utils" ) // JinaEmbedder implements text vectorization functionality using Jina AI API // Jina API is mostly OpenAI-compatible but does NOT support truncate_prompt_tokens type JinaEmbedder struct { - apiKey string - baseURL string - modelName string - dimensions int - modelID string - httpClient *http.Client - timeout time.Duration - maxRetries int + apiKey string + baseURL string + modelName string + dimensions int + modelID string + httpClient *http.Client + timeout time.Duration + maxRetries int + customHeaders map[string]string EmbedderPooler } +// SetCustomHeaders 设置用户自定义 HTTP 请求头(类似 OpenAI Python SDK 的 extra_headers)。 +func (e *JinaEmbedder) SetCustomHeaders(headers map[string]string) { + e.customHeaders = headers +} + // JinaEmbedRequest represents a Jina embedding request // Note: Jina uses 'truncate' (boolean) instead of 'truncate_prompt_tokens' (integer) type JinaEmbedRequest struct { @@ -118,6 +125,7 @@ func (e *JinaEmbedder) doRequestWithRetry(ctx context.Context, jsonData []byte) } req.Header.Set("Content-Type", "application/json") req.Header.Set("Authorization", "Bearer "+e.apiKey) + secutils.ApplyCustomHeaders(req, e.customHeaders) resp, err = e.httpClient.Do(req) if err == nil { diff --git a/internal/models/embedding/nvidia.go b/internal/models/embedding/nvidia.go index 2c37dbbd..f645da79 100644 --- a/internal/models/embedding/nvidia.go +++ b/internal/models/embedding/nvidia.go @@ -11,21 +11,28 @@ import ( "github.com/Tencent/WeKnora/internal/logger" "github.com/Tencent/WeKnora/internal/types" + secutils "github.com/Tencent/WeKnora/internal/utils" ) // NvidiaEmbedder implements text vectorization functionality using NVIDIA API type NvidiaEmbedder struct { - apiKey string - baseURL string - modelName string - dimensions int - modelID string - httpClient *http.Client - timeout time.Duration - maxRetries int + apiKey string + baseURL string + modelName string + dimensions int + modelID string + httpClient *http.Client + timeout time.Duration + maxRetries int + customHeaders map[string]string EmbedderPooler } +// SetCustomHeaders 设置用户自定义 HTTP 请求头(类似 OpenAI Python SDK 的 extra_headers)。 +func (e *NvidiaEmbedder) SetCustomHeaders(headers map[string]string) { + e.customHeaders = headers +} + // NvidiaEmbedRequest represents an NVIDIA embedding request type NvidiaEmbedRequest struct { Model string `json:"model"` @@ -118,6 +125,7 @@ func (e *NvidiaEmbedder) doRequestWithRetry(ctx context.Context, jsonData []byte } req.Header.Set("Content-Type", "application/json") req.Header.Set("Authorization", "Bearer "+e.apiKey) + secutils.ApplyCustomHeaders(req, e.customHeaders) resp, err = e.httpClient.Do(req) if err == nil { diff --git a/internal/models/embedding/openai.go b/internal/models/embedding/openai.go index 7e495e13..0c70bfd8 100644 --- a/internal/models/embedding/openai.go +++ b/internal/models/embedding/openai.go @@ -10,6 +10,7 @@ import ( "time" "github.com/Tencent/WeKnora/internal/logger" + secutils "github.com/Tencent/WeKnora/internal/utils" ) // OpenAIEmbedder implements text vectorization functionality using OpenAI API @@ -23,6 +24,7 @@ type OpenAIEmbedder struct { httpClient *http.Client timeout time.Duration maxRetries int + customHeaders map[string]string EmbedderPooler } @@ -79,6 +81,12 @@ func NewOpenAIEmbedder(apiKey, baseURL, modelName string, }, nil } +// SetCustomHeaders 设置用户自定义 HTTP 请求头(类似 OpenAI Python SDK 的 extra_headers)。 +// 保留头(Authorization、Content-Type 等)会在发送时被自动跳过。 +func (e *OpenAIEmbedder) SetCustomHeaders(headers map[string]string) { + e.customHeaders = headers +} + // Embed converts text to vector func (e *OpenAIEmbedder) Embed(ctx context.Context, text string) ([]float32, error) { for range 3 { @@ -122,6 +130,7 @@ func (e *OpenAIEmbedder) doRequestWithRetry(ctx context.Context, jsonData []byte } req.Header.Set("Content-Type", "application/json") req.Header.Set("Authorization", "Bearer "+e.apiKey) + secutils.ApplyCustomHeaders(req, e.customHeaders) resp, err = e.httpClient.Do(req) if err == nil { diff --git a/internal/models/embedding/volcengine.go b/internal/models/embedding/volcengine.go index 369def29..c5a3e617 100644 --- a/internal/models/embedding/volcengine.go +++ b/internal/models/embedding/volcengine.go @@ -11,6 +11,7 @@ import ( "time" "github.com/Tencent/WeKnora/internal/logger" + secutils "github.com/Tencent/WeKnora/internal/utils" ) const ( @@ -29,9 +30,15 @@ type VolcengineEmbedder struct { httpClient *http.Client timeout time.Duration maxRetries int + customHeaders map[string]string EmbedderPooler } +// SetCustomHeaders 设置用户自定义 HTTP 请求头(类似 OpenAI Python SDK 的 extra_headers)。 +func (e *VolcengineEmbedder) SetCustomHeaders(headers map[string]string) { + e.customHeaders = headers +} + // VolcengineEmbedRequest represents a Volcengine Ark multimodal embedding request type VolcengineEmbedRequest struct { Model string `json:"model"` @@ -165,6 +172,7 @@ func (e *VolcengineEmbedder) doRequestWithRetry(ctx context.Context, jsonData [] } req.Header.Set("Content-Type", "application/json") req.Header.Set("Authorization", "Bearer "+e.apiKey) + secutils.ApplyCustomHeaders(req, e.customHeaders) resp, err = e.httpClient.Do(req) if err == nil { diff --git a/internal/models/rerank/aliyun_reranker.go b/internal/models/rerank/aliyun_reranker.go index 75103c87..ccd0c307 100644 --- a/internal/models/rerank/aliyun_reranker.go +++ b/internal/models/rerank/aliyun_reranker.go @@ -9,15 +9,22 @@ import ( "net/http" "github.com/Tencent/WeKnora/internal/logger" + secutils "github.com/Tencent/WeKnora/internal/utils" ) // AliyunReranker implements a reranking system based on Aliyun DashScope models type AliyunReranker struct { - modelName string // Name of the model used for reranking - modelID string // Unique identifier of the model - apiKey string // API key for authentication - baseURL string // Base URL for API requests - client *http.Client // HTTP client for making API requests + modelName string // Name of the model used for reranking + modelID string // Unique identifier of the model + apiKey string // API key for authentication + baseURL string // Base URL for API requests + client *http.Client // HTTP client for making API requests + customHeaders map[string]string +} + +// SetCustomHeaders 设置用户自定义 HTTP 请求头(类似 OpenAI Python SDK 的 extra_headers)。 +func (r *AliyunReranker) SetCustomHeaders(headers map[string]string) { + r.customHeaders = headers } // AliyunRerankRequest represents a request to rerank documents using Aliyun DashScope API @@ -111,6 +118,7 @@ func (r *AliyunReranker) Rerank(ctx context.Context, query string, documents []s } req.Header.Set("Content-Type", "application/json") req.Header.Set("Authorization", fmt.Sprintf("Bearer %s", r.apiKey)) + secutils.ApplyCustomHeaders(req, r.customHeaders) logger.Debugf(ctx, "%s", buildRerankRequestDebug(r.modelName, r.baseURL, query, documents)) diff --git a/internal/models/rerank/config_from_model_test.go b/internal/models/rerank/config_from_model_test.go new file mode 100644 index 00000000..39085ac1 --- /dev/null +++ b/internal/models/rerank/config_from_model_test.go @@ -0,0 +1,32 @@ +package rerank + +import ( + "testing" + + "github.com/Tencent/WeKnora/internal/types" +) + +func TestConfigFromModel(t *testing.T) { + m := &types.Model{ + ID: "rr-1", + Name: "bge-reranker-v2-m3", + Source: types.ModelSourceRemote, + Parameters: types.ModelParameters{ + BaseURL: "https://api.example.com/v1", + APIKey: "sk-xxx", + Provider: "siliconflow", + ExtraConfig: map[string]string{"flag": "on"}, + CustomHeaders: map[string]string{"X-Gateway": "g"}, + }, + } + cfg := ConfigFromModel(m, "app", "secret") + if cfg == nil || cfg.ModelID != "rr-1" || cfg.ModelName != "bge-reranker-v2-m3" { + t.Fatalf("identity mismatch: %+v", cfg) + } + if cfg.Provider != "siliconflow" || cfg.CustomHeaders["X-Gateway"] != "g" { + t.Errorf("provider/headers not propagated: %+v", cfg) + } + if cfg.AppID != "app" || cfg.AppSecret != "secret" { + t.Errorf("cloud creds mismatch: %+v", cfg) + } +} diff --git a/internal/models/rerank/jina_reranker.go b/internal/models/rerank/jina_reranker.go index 5067123e..27f93af6 100644 --- a/internal/models/rerank/jina_reranker.go +++ b/internal/models/rerank/jina_reranker.go @@ -9,16 +9,23 @@ import ( "net/http" "github.com/Tencent/WeKnora/internal/logger" + secutils "github.com/Tencent/WeKnora/internal/utils" ) // JinaReranker implements a reranking system using Jina AI API // Jina API uses different parameters than standard OpenAI-compatible APIs type JinaReranker struct { - modelName string // Name of the model used for reranking - modelID string // Unique identifier of the model - apiKey string // API key for authentication - baseURL string // Base URL for API requests - client *http.Client // HTTP client for making API requests + modelName string // Name of the model used for reranking + modelID string // Unique identifier of the model + apiKey string // API key for authentication + baseURL string // Base URL for API requests + client *http.Client // HTTP client for making API requests + customHeaders map[string]string +} + +// SetCustomHeaders 设置用户自定义 HTTP 请求头(类似 OpenAI Python SDK 的 extra_headers)。 +func (r *JinaReranker) SetCustomHeaders(headers map[string]string) { + r.customHeaders = headers } // JinaRerankRequest represents a Jina rerank request @@ -79,6 +86,7 @@ func (r *JinaReranker) Rerank(ctx context.Context, query string, documents []str } req.Header.Set("Content-Type", "application/json") req.Header.Set("Authorization", fmt.Sprintf("Bearer %s", r.apiKey)) + secutils.ApplyCustomHeaders(req, r.customHeaders) logger.Debugf(ctx, "%s", buildRerankRequestDebug(r.modelName, fmt.Sprintf("%s/rerank", r.baseURL), query, documents)) diff --git a/internal/models/rerank/nvidia_reranker.go b/internal/models/rerank/nvidia_reranker.go index 13097ed8..bdc8bd72 100644 --- a/internal/models/rerank/nvidia_reranker.go +++ b/internal/models/rerank/nvidia_reranker.go @@ -9,16 +9,23 @@ import ( "net/http" "github.com/Tencent/WeKnora/internal/logger" + secutils "github.com/Tencent/WeKnora/internal/utils" ) // NvidiaReranker implements a reranking system using Jina AI API // Jina API uses different parameters than standard OpenAI-compatible APIs type NvidiaReranker struct { - modelName string // Name of the model used for reranking - modelID string // Unique identifier of the model - apiKey string // API key for authentication - baseURL string // Base URL for API requests - client *http.Client // HTTP client for making API requests + modelName string // Name of the model used for reranking + modelID string // Unique identifier of the model + apiKey string // API key for authentication + baseURL string // Base URL for API requests + client *http.Client // HTTP client for making API requests + customHeaders map[string]string +} + +// SetCustomHeaders 设置用户自定义 HTTP 请求头(类似 OpenAI Python SDK 的 extra_headers)。 +func (r *NvidiaReranker) SetCustomHeaders(headers map[string]string) { + r.customHeaders = headers } type NvidiaRerankDocument struct { Text string `json:"text"` @@ -88,6 +95,7 @@ func (r *NvidiaReranker) Rerank(ctx context.Context, query string, documents []s } req.Header.Set("Content-Type", "application/json") req.Header.Set("Authorization", fmt.Sprintf("Bearer %s", r.apiKey)) + secutils.ApplyCustomHeaders(req, r.customHeaders) // Log the curl equivalent for debugging (API key masked for security) logger.GetLogger(ctx).Infof( diff --git a/internal/models/rerank/remote_api.go b/internal/models/rerank/remote_api.go index d9b77771..ce78be20 100644 --- a/internal/models/rerank/remote_api.go +++ b/internal/models/rerank/remote_api.go @@ -9,15 +9,22 @@ import ( "net/http" "github.com/Tencent/WeKnora/internal/logger" + secutils "github.com/Tencent/WeKnora/internal/utils" ) // OpenAIReranker implements a reranking system based on OpenAI models type OpenAIReranker struct { - modelName string // Name of the model used for reranking - modelID string // Unique identifier of the model - apiKey string // API key for authentication - baseURL string // Base URL for API requests - client *http.Client // HTTP client for making API requests + modelName string // Name of the model used for reranking + modelID string // Unique identifier of the model + apiKey string // API key for authentication + baseURL string // Base URL for API requests + client *http.Client // HTTP client for making API requests + customHeaders map[string]string +} + +// SetCustomHeaders 设置用户自定义 HTTP 请求头(类似 OpenAI Python SDK 的 extra_headers)。 +func (r *OpenAIReranker) SetCustomHeaders(headers map[string]string) { + r.customHeaders = headers } // RerankRequest represents a request to rerank documents based on relevance to a query @@ -81,6 +88,7 @@ func (r *OpenAIReranker) Rerank(ctx context.Context, query string, documents []s } req.Header.Set("Content-Type", "application/json") req.Header.Set("Authorization", fmt.Sprintf("Bearer %s", r.apiKey)) + secutils.ApplyCustomHeaders(req, r.customHeaders) logger.Debugf(ctx, "%s", buildRerankRequestDebug(r.modelName, fmt.Sprintf("%s/rerank", r.baseURL), query, documents)) diff --git a/internal/models/rerank/reranker.go b/internal/models/rerank/reranker.go index 1e8d1d86..e8c18b7e 100644 --- a/internal/models/rerank/reranker.go +++ b/internal/models/rerank/reranker.go @@ -86,8 +86,31 @@ type RerankerConfig struct { ModelID string Provider string // Provider identifier: openai, aliyun, zhipu, siliconflow, jina, generic ExtraConfig map[string]string - AppID string - AppSecret string // 加密值,工厂函数调用方传入,使用前已解密 + // CustomHeaders 允许在调用远程 API 时附加自定义 HTTP 请求头(类似 OpenAI Python SDK 的 extra_headers)。 + CustomHeaders map[string]string + AppID string + AppSecret string // 加密值,工厂函数调用方传入,使用前已解密 +} + +// ConfigFromModel 根据 types.Model 构造 RerankerConfig。 +// 生产路径(从 DB 拉起)和测试连接路径(临时表单)共享这份映射。 +// appID / appSecret 是已解密的 WeKnoraCloud 凭证,调用方负责传入。 +func ConfigFromModel(m *types.Model, appID, appSecret string) *RerankerConfig { + if m == nil { + return nil + } + return &RerankerConfig{ + ModelID: m.ID, + APIKey: m.Parameters.APIKey, + BaseURL: m.Parameters.BaseURL, + ModelName: m.Name, + Source: m.Source, + Provider: m.Parameters.Provider, + ExtraConfig: m.Parameters.ExtraConfig, + CustomHeaders: m.Parameters.CustomHeaders, + AppID: appID, + AppSecret: appSecret, + } } // NewReranker creates a reranker based on the configuration @@ -99,6 +122,11 @@ func NewReranker(config *RerankerConfig) (Reranker, error) { return &debugReranker{inner: r}, nil } +// customHeaderSetter 表示支持注入自定义 HTTP header 的 reranker 实现。 +type customHeaderSetter interface { + SetCustomHeaders(map[string]string) +} + func newReranker(config *RerankerConfig) (Reranker, error) { // Use provider field if set, otherwise detect from URL using provider registry providerName := provider.ProviderName(config.Provider) @@ -106,18 +134,29 @@ func newReranker(config *RerankerConfig) (Reranker, error) { providerName = provider.DetectProvider(config.BaseURL) } + var ( + reranker Reranker + err error + ) switch providerName { case provider.ProviderAliyun: - return NewAliyunReranker(config) + reranker, err = NewAliyunReranker(config) case provider.ProviderZhipu: - return NewZhipuReranker(config) + reranker, err = NewZhipuReranker(config) case provider.ProviderJina: - return NewJinaReranker(config) + reranker, err = NewJinaReranker(config) case provider.ProviderNvidia: - return NewNvidiaReranker(config) + reranker, err = NewNvidiaReranker(config) case provider.ProviderWeKnoraCloud: - return NewWeKnoraCloudReranker(config) + reranker, err = NewWeKnoraCloudReranker(config) default: - return NewOpenAIReranker(config) + reranker, err = NewOpenAIReranker(config) } + if err != nil { + return nil, err + } + if setter, ok := reranker.(customHeaderSetter); ok { + setter.SetCustomHeaders(config.CustomHeaders) + } + return reranker, nil } diff --git a/internal/models/rerank/zhipu_reranker.go b/internal/models/rerank/zhipu_reranker.go index ad0d801c..2e2553f2 100644 --- a/internal/models/rerank/zhipu_reranker.go +++ b/internal/models/rerank/zhipu_reranker.go @@ -9,15 +9,22 @@ import ( "net/http" "github.com/Tencent/WeKnora/internal/logger" + secutils "github.com/Tencent/WeKnora/internal/utils" ) // ZhipuReranker implements a reranking system based on Zhipu AI models type ZhipuReranker struct { - modelName string // Name of the model used for reranking - modelID string // Unique identifier of the model - apiKey string // API key for authentication - baseURL string // Base URL for API requests - client *http.Client // HTTP client for making API requests + modelName string // Name of the model used for reranking + modelID string // Unique identifier of the model + apiKey string // API key for authentication + baseURL string // Base URL for API requests + client *http.Client // HTTP client for making API requests + customHeaders map[string]string +} + +// SetCustomHeaders 设置用户自定义 HTTP 请求头(类似 OpenAI Python SDK 的 extra_headers)。 +func (r *ZhipuReranker) SetCustomHeaders(headers map[string]string) { + r.customHeaders = headers } // ZhipuRerankRequest represents a request to rerank documents using Zhipu AI API @@ -92,6 +99,7 @@ func (r *ZhipuReranker) Rerank(ctx context.Context, query string, documents []st } req.Header.Set("Content-Type", "application/json") req.Header.Set("Authorization", fmt.Sprintf("Bearer %s", r.apiKey)) + secutils.ApplyCustomHeaders(req, r.customHeaders) logger.Debugf(ctx, "%s", buildRerankRequestDebug(r.modelName, r.baseURL, query, documents)) diff --git a/internal/models/vlm/config_from_model_test.go b/internal/models/vlm/config_from_model_test.go new file mode 100644 index 00000000..fd71db2e --- /dev/null +++ b/internal/models/vlm/config_from_model_test.go @@ -0,0 +1,59 @@ +package vlm + +import ( + "testing" + + "github.com/Tencent/WeKnora/internal/types" +) + +func TestConfigFromModel_RemoteDefaultsToOpenAI(t *testing.T) { + m := &types.Model{ + ID: "v1", + Name: "gpt-4o", + Source: types.ModelSourceRemote, + Parameters: types.ModelParameters{ + BaseURL: "https://api.example.com/v1", + APIKey: "sk", + Provider: "openai", + ExtraConfig: map[string]string{"x": "y"}, + CustomHeaders: map[string]string{"H": "v"}, + }, + } + cfg := ConfigFromModel(m, "app", "secret") + if cfg.InterfaceType != "openai" { + t.Errorf("expected openai default for remote, got %q", cfg.InterfaceType) + } + if cfg.CustomHeaders["H"] != "v" { + t.Errorf("CustomHeaders not propagated: %+v", cfg.CustomHeaders) + } + if cfg.Extra["x"] != "y" { + t.Errorf("ExtraConfig not propagated as Extra: %+v", cfg.Extra) + } + if cfg.AppID != "app" || cfg.AppSecret != "secret" { + t.Errorf("cloud creds mismatch: %+v", cfg) + } +} + +func TestConfigFromModel_LocalDefaultsToOllama(t *testing.T) { + m := &types.Model{ + Name: "qwen2-vl", + Source: types.ModelSourceLocal, + } + cfg := ConfigFromModel(m, "", "") + if cfg.InterfaceType != "ollama" { + t.Errorf("expected ollama default for local, got %q", cfg.InterfaceType) + } +} + +func TestConfigFromModel_RespectsExplicitInterface(t *testing.T) { + m := &types.Model{ + Name: "qwen2-vl", + Source: types.ModelSourceRemote, + Parameters: types.ModelParameters{ + InterfaceType: "ollama", + }, + } + if got := ConfigFromModel(m, "", "").InterfaceType; got != "ollama" { + t.Errorf("expected explicit interface to win, got %q", got) + } +} diff --git a/internal/models/vlm/remote_api.go b/internal/models/vlm/remote_api.go index 3c31aa5e..03feba4f 100644 --- a/internal/models/vlm/remote_api.go +++ b/internal/models/vlm/remote_api.go @@ -10,6 +10,7 @@ import ( "github.com/Tencent/WeKnora/internal/logger" "github.com/Tencent/WeKnora/internal/models/provider" + secutils "github.com/Tencent/WeKnora/internal/utils" openai "github.com/sashabaranov/go-openai" ) @@ -53,7 +54,14 @@ func NewRemoteAPIVLM(config *Config) (*RemoteAPIVLM, error) { apiCfg.BaseURL = config.BaseURL } } - apiCfg.HTTPClient = &http.Client{Timeout: defaultTimeout} + httpClient := &http.Client{Timeout: defaultTimeout} + + // 注入用户自定义 HTTP header(类似 OpenAI Python SDK 的 extra_headers) + if len(config.CustomHeaders) > 0 { + apiCfg.HTTPClient = secutils.WrapHTTPClientWithHeaders(httpClient, config.CustomHeaders) + } else { + apiCfg.HTTPClient = httpClient + } return &RemoteAPIVLM{ modelName: config.ModelName, diff --git a/internal/models/vlm/vlm.go b/internal/models/vlm/vlm.go index 16e5f5c1..4bdb2612 100644 --- a/internal/models/vlm/vlm.go +++ b/internal/models/vlm/vlm.go @@ -30,10 +30,54 @@ type Config struct { InterfaceType string // "ollama" or "openai" (default) Provider string Extra map[string]any + // CustomHeaders 允许在调用远程 API 时附加自定义 HTTP 请求头(类似 OpenAI Python SDK 的 extra_headers)。 + CustomHeaders map[string]string AppID string AppSecret string } +// ConfigFromModel 根据 types.Model 构造 vlm.Config。 +// 生产路径(从 DB 拉起)和测试连接路径(临时表单)共享这份映射。 +// appID / appSecret 是已解密的 WeKnoraCloud 凭证,调用方负责传入。 +// InterfaceType 会根据 source / 模型参数自动回退到合理默认值。 +func ConfigFromModel(m *types.Model, appID, appSecret string) *Config { + if m == nil { + return nil + } + ifType := m.Parameters.InterfaceType + if ifType == "" { + if m.Source == types.ModelSourceLocal { + ifType = "ollama" + } else { + ifType = "openai" + } + } + return &Config{ + ModelID: m.ID, + APIKey: m.Parameters.APIKey, + BaseURL: m.Parameters.BaseURL, + ModelName: m.Name, + Source: m.Source, + InterfaceType: ifType, + Provider: m.Parameters.Provider, + Extra: stringMapToAnyMap(m.Parameters.ExtraConfig), + CustomHeaders: m.Parameters.CustomHeaders, + AppID: appID, + AppSecret: appSecret, + } +} + +func stringMapToAnyMap(in map[string]string) map[string]any { + if in == nil { + return nil + } + out := make(map[string]any, len(in)) + for k, v := range in { + out[k] = v + } + return out +} + // NewVLM creates a VLM instance based on the provided configuration. func NewVLM(config *Config, ollamaService *ollama.OllamaService) (VLM, error) { v, err := newVLM(config, ollamaService) diff --git a/internal/types/model.go b/internal/types/model.go index b9ef8122..c5f26d4a 100644 --- a/internal/types/model.go +++ b/internal/types/model.go @@ -67,7 +67,12 @@ type ModelParameters struct { ParameterSize string `yaml:"parameter_size" json:"parameter_size"` // Ollama model parameter size (e.g., "7B", "13B", "70B") Provider string `yaml:"provider" json:"provider"` // Provider identifier: openai, aliyun, zhipu, generic ExtraConfig map[string]string `yaml:"extra_config" json:"extra_config"` // Provider-specific configuration - SupportsVision bool `yaml:"supports_vision" json:"supports_vision"` // Whether the model accepts image/multimodal input + // CustomHeaders 允许在调用远程模型 API 时附加自定义 HTTP 请求头, + // 用途类似 Python OpenAI SDK 的 extra_headers 参数, + // 常见场景包括透传企业网关鉴权信息、追踪 ID、路由标识等。 + // 保留字段(Authorization、api-key、Content-Type、Accept 等)会在运行期被忽略以避免破坏签名/鉴权流程。 + CustomHeaders map[string]string `yaml:"custom_headers,omitempty" json:"custom_headers,omitempty"` + SupportsVision bool `yaml:"supports_vision" json:"supports_vision"` // Whether the model accepts image/multimodal input // WeKnoraCloud 厂商专用凭证 AppID string `yaml:"app_id,omitempty" json:"app_id,omitempty"` AppSecret string `yaml:"app_secret,omitempty" json:"app_secret,omitempty"` // AES-256 加密存储,实际承载上游 API Key diff --git a/internal/utils/extraheaders.go b/internal/utils/extraheaders.go new file mode 100644 index 00000000..0a7aab28 --- /dev/null +++ b/internal/utils/extraheaders.go @@ -0,0 +1,86 @@ +package utils + +import ( + "net/http" + "strings" +) + +// reservedHeaderKeys 列出不允许被用户自定义头覆盖的关键请求头。 +// 这些头由各 provider 的签名、鉴权或 SSE 流程控制,覆盖后可能直接导致调用失败。 +var reservedHeaderKeys = map[string]struct{}{ + "authorization": {}, + "api-key": {}, + "x-api-key": {}, + "content-type": {}, + "content-length": {}, + "accept-encoding": {}, + "host": {}, + "connection": {}, + "transfer-encoding": {}, +} + +// IsReservedHeader 判断某个 header key 是否为保留 header,保留 header 不允许被自定义头覆盖。 +func IsReservedHeader(key string) bool { + _, ok := reservedHeaderKeys[strings.ToLower(strings.TrimSpace(key))] + return ok +} + +// ApplyCustomHeaders 将用户自定义的 header 写入 http.Request。 +// 保留 header(Authorization、api-key、Content-Type 等)会被跳过以避免破坏鉴权/签名。 +// 其它 header 会直接覆盖同名条目,允许用户替换默认值(例如 Accept)。 +func ApplyCustomHeaders(req *http.Request, headers map[string]string) { + if req == nil || len(headers) == 0 { + return + } + for k, v := range headers { + name := strings.TrimSpace(k) + if name == "" { + continue + } + if IsReservedHeader(name) { + continue + } + req.Header.Set(name, v) + } +} + +// CustomHeadersRoundTripper 是一个 http.RoundTripper 包装器, +// 会在每个 HTTP 请求发出前注入用户自定义的 header。 +// 用于无法直接拿到底层 *http.Request 的场景(如 go-openai SDK)。 +type CustomHeadersRoundTripper struct { + Headers map[string]string + Base http.RoundTripper +} + +// RoundTrip 实现 http.RoundTripper 接口。 +func (t *CustomHeadersRoundTripper) RoundTrip(req *http.Request) (*http.Response, error) { + base := t.Base + if base == nil { + base = http.DefaultTransport + } + if len(t.Headers) > 0 { + // 复制请求以避免修改调用方传入的对象。 + cloned := req.Clone(req.Context()) + ApplyCustomHeaders(cloned, t.Headers) + return base.RoundTrip(cloned) + } + return base.RoundTrip(req) +} + +// WrapHTTPClientWithHeaders 返回一个新的 *http.Client,在原有 client 基础上注入自定义 header。 +// 如果 headers 为空则直接返回原 client,避免不必要的开销。 +func WrapHTTPClientWithHeaders(client *http.Client, headers map[string]string) *http.Client { + if len(headers) == 0 { + return client + } + if client == nil { + client = &http.Client{} + } + base := client.Transport + if base == nil { + base = http.DefaultTransport + } + wrapped := *client + wrapped.Transport = &CustomHeadersRoundTripper{Headers: headers, Base: base} + return &wrapped +} diff --git a/internal/utils/extraheaders_test.go b/internal/utils/extraheaders_test.go new file mode 100644 index 00000000..4b2d6fb7 --- /dev/null +++ b/internal/utils/extraheaders_test.go @@ -0,0 +1,86 @@ +package utils + +import ( + "io" + "net/http" + "net/http/httptest" + "strings" + "testing" +) + +func TestApplyCustomHeaders_SkipReserved(t *testing.T) { + req, _ := http.NewRequest("GET", "https://example.com", nil) + req.Header.Set("Authorization", "Bearer original") + req.Header.Set("Content-Type", "application/json") + + ApplyCustomHeaders(req, map[string]string{ + "Authorization": "Bearer injected", + "Content-Type": "text/plain", + "X-Trace-Id": "trace-123", + "X-Route": "edge", + "": "empty-key-should-be-skipped", + }) + + if got := req.Header.Get("Authorization"); got != "Bearer original" { + t.Fatalf("authorization overwritten: %q", got) + } + if got := req.Header.Get("Content-Type"); got != "application/json" { + t.Fatalf("content-type overwritten: %q", got) + } + if got := req.Header.Get("X-Trace-Id"); got != "trace-123" { + t.Fatalf("X-Trace-Id not injected: %q", got) + } + if got := req.Header.Get("X-Route"); got != "edge" { + t.Fatalf("X-Route not injected: %q", got) + } +} + +func TestApplyCustomHeaders_NilSafe(t *testing.T) { + ApplyCustomHeaders(nil, map[string]string{"x": "y"}) + req, _ := http.NewRequest("GET", "https://example.com", nil) + ApplyCustomHeaders(req, nil) + if len(req.Header) != 0 { + t.Fatalf("unexpected headers added: %+v", req.Header) + } +} + +func TestWrapHTTPClientWithHeaders(t *testing.T) { + gotTrace := "" + gotAuth := "" + srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { + gotTrace = r.Header.Get("X-Trace-Id") + gotAuth = r.Header.Get("Authorization") + io.Copy(io.Discard, r.Body) + w.WriteHeader(http.StatusOK) + })) + defer srv.Close() + + client := WrapHTTPClientWithHeaders(nil, map[string]string{ + "X-Trace-Id": "rt-1", + "Authorization": "Bearer should-not-override", + }) + + req, _ := http.NewRequest("POST", srv.URL, strings.NewReader("{}")) + req.Header.Set("Authorization", "Bearer kept") + + resp, err := client.Do(req) + if err != nil { + t.Fatalf("request failed: %v", err) + } + resp.Body.Close() + + if gotTrace != "rt-1" { + t.Fatalf("expected custom header injected, got %q", gotTrace) + } + if gotAuth != "Bearer kept" { + t.Fatalf("reserved header must not be overridden, got %q", gotAuth) + } +} + +func TestWrapHTTPClientWithHeaders_EmptyReturnsOriginal(t *testing.T) { + orig := &http.Client{} + wrapped := WrapHTTPClientWithHeaders(orig, nil) + if wrapped != orig { + t.Fatalf("expected original client returned when headers empty") + } +}