diff --git a/frontend/src/components/ModelEditorDialog.vue b/frontend/src/components/ModelEditorDialog.vue index b1918d30..7db7a57a 100644 --- a/frontend/src/components/ModelEditorDialog.vue +++ b/frontend/src/components/ModelEditorDialog.vue @@ -360,7 +360,7 @@ const fallbackProviderOptions = computed(() => [ value: 'nvidia', label: t('model.editor.providers.nvidia.label'), defaultUrls: { - chat: 'https://integrate.api.nvidia.com/v1', + chat: 'https://integrate.api.nvidia.com/v1/chat/completions', embedding: 'https://integrate.api.nvidia.com/v1', rerank: 'https://ai.api.nvidia.com/v1/retrieval/nvidia/reranking', vllm: 'https://integrate.api.nvidia.com/v1', diff --git a/internal/models/chat/chat.go b/internal/models/chat/chat.go index 781804fb..2cb1752a 100644 --- a/internal/models/chat/chat.go +++ b/internal/models/chat/chat.go @@ -121,6 +121,9 @@ func NewRemoteChat(config *ChatConfig) (Chat, error) { case provider.ProviderGeneric: // Generic provider (如 vLLM) 使用 ChatTemplateKwargs return NewGenericChat(config) + case provider.ProviderNvidia: + // NVIDIA provider 使用BaseURL为请求地址 + return NewNvidiaChat(config) default: // 其他 provider 使用标准 OpenAI 兼容实现 return NewRemoteAPIChat(config) diff --git a/internal/models/chat/nvidia.go b/internal/models/chat/nvidia.go new file mode 100644 index 00000000..08ad176b --- /dev/null +++ b/internal/models/chat/nvidia.go @@ -0,0 +1,34 @@ +package chat + +import ( + "github.com/Tencent/WeKnora/internal/models/provider" +) + +// NvidiaChat NVIDIA 模型聊天实现 +// NVIDIA 模型需要自定义请求地址 +type NvidiaChat struct { + *RemoteAPIChat +} + +// NewNvidiaChat 创建 NVIDIA 聊天实例 +func NewNvidiaChat(config *ChatConfig) (*NvidiaChat, error) { + config.Provider = string(provider.ProviderAliyun) + + remoteChat, err := NewRemoteAPIChat(config) + if err != nil { + return nil, err + } + + chat := &NvidiaChat{ + RemoteAPIChat: remoteChat, + } + + // 设置请求地址自定义器 + remoteChat.SetEndpointCustomizer(chat.endpointCustomizer) + return chat, nil +} + +// customizeRequest 自定义 Qwen 请求 +func (c *NvidiaChat) endpointCustomizer(baseURL string, modelID string, isStream bool) string { + return baseURL +} diff --git a/internal/models/chat/remote_api.go b/internal/models/chat/remote_api.go index 1900fd05..c7ad5c6f 100644 --- a/internal/models/chat/remote_api.go +++ b/internal/models/chat/remote_api.go @@ -28,6 +28,10 @@ type RemoteAPIChat struct { // 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 requestCustomizer func(req *openai.ChatCompletionRequest, opts *ChatOptions, isStream bool) (customReq any, useRawHTTP bool) + + // endpointCustomizer 允许子类自定义请求的 endpoint + // 返回是否使用自定义请求地址, 返回空则使用默认OpenAI格式地址 + endpointCustomizer func(baseURL string, modelID string, isStream bool) (endpoint string) } // NewRemoteAPIChat 创建远程 API 聊天实例 @@ -58,6 +62,11 @@ func (c *RemoteAPIChat) SetRequestCustomizer(customizer func(req *openai.ChatCom c.requestCustomizer = customizer } +// SetEndpointCustomizer 设置请求地址自定义器 +func (c *RemoteAPIChat) SetEndpointCustomizer(customizer func(baseURL string, modelID string, isStream bool) string) { + c.endpointCustomizer = customizer +} + // ConvertMessages 转换消息格式为 OpenAI 格式(导出供子类使用) func (c *RemoteAPIChat) ConvertMessages(messages []Message) []openai.ChatCompletionMessage { openaiMessages := make([]openai.ChatCompletionMessage, 0, len(messages)) @@ -178,15 +187,23 @@ func (c *RemoteAPIChat) logRequest(ctx context.Context, req any, isStream bool) // Chat 进行非流式聊天 func (c *RemoteAPIChat) Chat(ctx context.Context, messages []Message, opts *ChatOptions) (*types.ChatResponse, error) { req := c.BuildChatCompletionRequest(messages, opts, false) - + var customEndpoint string + if c.endpointCustomizer != nil { + customEndpoint = c.endpointCustomizer(c.baseURL, c.modelID, true) + } // 检查是否需要自定义请求 if c.requestCustomizer != nil { customReq, useRawHTTP := c.requestCustomizer(&req, opts, false) if useRawHTTP && customReq != nil { - return c.chatWithRawHTTP(ctx, customReq) + return c.chatWithRawHTTP(ctx, customEndpoint, customReq) } } + // 使用自定义请求地址 + if customEndpoint != "" { + return c.chatWithRawHTTP(ctx, customEndpoint, &req) + } + c.logRequest(ctx, req, false) resp, err := c.client.CreateChatCompletion(ctx, req) @@ -198,15 +215,16 @@ func (c *RemoteAPIChat) Chat(ctx context.Context, messages []Message, opts *Chat } // chatWithRawHTTP 使用原始 HTTP 请求进行聊天(供自定义请求使用) -func (c *RemoteAPIChat) chatWithRawHTTP(ctx context.Context, customReq any) (*types.ChatResponse, error) { +func (c *RemoteAPIChat) chatWithRawHTTP(ctx context.Context, endpoint string, customReq any) (*types.ChatResponse, error) { jsonData, err := json.Marshal(customReq) if err != nil { return nil, fmt.Errorf("marshal request: %w", err) } logger.Infof(ctx, "[LLM Request] model=%s, raw HTTP request:\n%s", c.modelName, string(jsonData)) - - endpoint := c.baseURL + "/chat/completions" + if endpoint == "" { + endpoint = c.baseURL + "/chat/completions" + } httpReq, err := http.NewRequestWithContext(ctx, "POST", endpoint, bytes.NewBuffer(jsonData)) if err != nil { return nil, fmt.Errorf("create request: %w", err) @@ -304,14 +322,22 @@ func removeThinkingContent(content string) string { func (c *RemoteAPIChat) ChatStream(ctx context.Context, messages []Message, opts *ChatOptions) (<-chan types.StreamResponse, error) { req := c.BuildChatCompletionRequest(messages, opts, true) + var customEndpoint string + if c.endpointCustomizer != nil { + customEndpoint = c.endpointCustomizer(c.baseURL, c.modelID, true) + } + // 检查是否需要自定义请求 if c.requestCustomizer != nil { customReq, useRawHTTP := c.requestCustomizer(&req, opts, true) if useRawHTTP && customReq != nil { - return c.chatStreamWithRawHTTP(ctx, customReq) + return c.chatStreamWithRawHTTP(ctx, customEndpoint, customReq) } } - + // 使用自定义请求地址 + if customEndpoint != "" { + return c.chatStreamWithRawHTTP(ctx, customEndpoint, &req) + } c.logRequest(ctx, req, true) streamChan := make(chan types.StreamResponse) @@ -328,7 +354,7 @@ func (c *RemoteAPIChat) ChatStream(ctx context.Context, messages []Message, opts } // chatStreamWithRawHTTP 使用原始 HTTP 请求进行流式聊天 -func (c *RemoteAPIChat) chatStreamWithRawHTTP(ctx context.Context, customReq any) (<-chan types.StreamResponse, error) { +func (c *RemoteAPIChat) chatStreamWithRawHTTP(ctx context.Context, endpoint string, customReq any) (<-chan types.StreamResponse, error) { jsonData, err := json.Marshal(customReq) if err != nil { return nil, fmt.Errorf("marshal request: %w", err) @@ -336,7 +362,9 @@ func (c *RemoteAPIChat) chatStreamWithRawHTTP(ctx context.Context, customReq any logger.Infof(ctx, "[LLM Stream] model=%s", c.modelName) - endpoint := c.baseURL + "/chat/completions" + if endpoint == "" { + endpoint = c.baseURL + "/chat/completions" + } httpReq, err := http.NewRequestWithContext(ctx, "POST", endpoint, bytes.NewBuffer(jsonData)) if err != nil { return nil, fmt.Errorf("create request: %w", err) diff --git a/internal/models/provider/nvidia.go b/internal/models/provider/nvidia.go index 9fb87c0d..6f2a9305 100644 --- a/internal/models/provider/nvidia.go +++ b/internal/models/provider/nvidia.go @@ -8,7 +8,7 @@ import ( const ( // NvidiaChatBaseURL NVIDIA Chat 的默认 BaseURL - NvidiaChatBaseURL = "https://integrate.api.nvidia.com/v1" + NvidiaChatBaseURL = "https://integrate.api.nvidia.com/v1/chat/completions" // NvidiaRerankBaseURL NVIDIA Rerank 的默认 BaseURL NvidiaRerankBaseURL = "https://ai.api.nvidia.com/v1/retrieval/nvidia/reranking" )