482 lines
14 KiB
Go
482 lines
14 KiB
Go
package llm
|
||
|
||
import (
|
||
"context"
|
||
"errors"
|
||
"fmt"
|
||
"io"
|
||
"strings"
|
||
|
||
"github.com/sashabaranov/go-openai"
|
||
)
|
||
|
||
// Config 大模型统一配置
|
||
type Config struct {
|
||
Provider string // "gemini", "deepseek", "openai", "ollama", "custom"
|
||
APIKey string // 密钥 (Ollama 本地可填任意字符或留空)
|
||
BaseURL string // 自定义接口地址 (非必填，会自动根据 Provider 设置默认值)
|
||
Model string // 模型名称 (非必填，会自动设置默认主力模型)
|
||
PriceInput float64 // 用户自定义每 1M 输入 Token 美元价格 (可选，覆盖内置价格表)
|
||
PriceOutput float64 // 用户自定义每 1M 输出 Token 美元价格 (可选，覆盖内置价格表)
|
||
}
|
||
|
||
// Client 统一大模型客户端
|
||
type Client struct {
|
||
client *openai.Client
|
||
model string
|
||
priceInput float64
|
||
priceOutput float64
|
||
}
|
||
|
||
// NewClient 初始化客户端
|
||
func NewClient(cfg Config) (*Client, error) {
|
||
provider := strings.ToLower(strings.TrimSpace(cfg.Provider))
|
||
if provider == "" {
|
||
provider = "gemini"
|
||
}
|
||
|
||
apiKey := cfg.APIKey
|
||
baseURL := cfg.BaseURL
|
||
model := cfg.Model
|
||
|
||
// 智能适配各大主流模型的官方地址与主力模型
|
||
switch provider {
|
||
case "gemini", "google":
|
||
if baseURL == "" {
|
||
baseURL = "https://generativelanguage.googleapis.com/v1beta/openai/"
|
||
}
|
||
if model == "" {
|
||
model = "gemini-3.7-flash"
|
||
}
|
||
if apiKey == "" {
|
||
return nil, fmt.Errorf("gemini 必须配置 api_key")
|
||
}
|
||
|
||
case "deepseek":
|
||
if baseURL == "" {
|
||
baseURL = "https://api.deepseek.com/v1"
|
||
}
|
||
if model == "" {
|
||
model = "deepseek-chat"
|
||
}
|
||
if apiKey == "" {
|
||
return nil, fmt.Errorf("deepseek 必须配置 api_key")
|
||
}
|
||
|
||
case "siliconflow", "silicon":
|
||
if baseURL == "" {
|
||
baseURL = "https://api.siliconflow.cn/v1"
|
||
}
|
||
if model == "" {
|
||
model = "deepseek-ai/DeepSeek-V3"
|
||
}
|
||
if apiKey == "" {
|
||
return nil, fmt.Errorf("siliconflow 必须配置 api_key")
|
||
}
|
||
|
||
case "qwen", "dashscope", "aliyun":
|
||
if baseURL == "" {
|
||
baseURL = "https://dashscope.aliyuncs.com/compatible-mode/v1"
|
||
}
|
||
if model == "" {
|
||
model = "qwen-plus"
|
||
}
|
||
if apiKey == "" {
|
||
return nil, fmt.Errorf("qwen/dashscope 必须配置 api_key")
|
||
}
|
||
|
||
case "zhipu", "glm":
|
||
if baseURL == "" {
|
||
baseURL = "https://open.bigmodel.cn/api/paas/v4/"
|
||
}
|
||
if model == "" {
|
||
model = "glm-4-plus"
|
||
}
|
||
if apiKey == "" {
|
||
return nil, fmt.Errorf("zhipu 必须配置 api_key")
|
||
}
|
||
|
||
case "moonshot", "kimi":
|
||
if baseURL == "" {
|
||
baseURL = "https://api.moonshot.cn/v1"
|
||
}
|
||
if model == "" {
|
||
model = "moonshot-v1-8k"
|
||
}
|
||
if apiKey == "" {
|
||
return nil, fmt.Errorf("moonshot/kimi 必须配置 api_key")
|
||
}
|
||
|
||
case "doubao", "volcengine":
|
||
if baseURL == "" {
|
||
baseURL = "https://ark.cn-beijing.volces.com/api/v3"
|
||
}
|
||
if model == "" {
|
||
model = "doubao-pro-32k"
|
||
}
|
||
if apiKey == "" {
|
||
return nil, fmt.Errorf("doubao 必须配置 api_key (并在 model 中填写接入点 Endpoint ID)")
|
||
}
|
||
|
||
case "minimax":
|
||
if baseURL == "" {
|
||
baseURL = "https://api.minimax.chat/v1"
|
||
}
|
||
if model == "" {
|
||
model = "abab6.5s-chat"
|
||
}
|
||
if apiKey == "" {
|
||
return nil, fmt.Errorf("minimax 必须配置 api_key")
|
||
}
|
||
|
||
case "yi", "lingyi":
|
||
if baseURL == "" {
|
||
baseURL = "https://api.lingyiwanwu.com/v1"
|
||
}
|
||
if model == "" {
|
||
model = "yi-lightning"
|
||
}
|
||
if apiKey == "" {
|
||
return nil, fmt.Errorf("yi 必须配置 api_key")
|
||
}
|
||
|
||
case "groq":
|
||
if baseURL == "" {
|
||
baseURL = "https://api.groq.com/openai/v1"
|
||
}
|
||
if model == "" {
|
||
model = "llama-3.3-70b-versatile"
|
||
}
|
||
if apiKey == "" {
|
||
return nil, fmt.Errorf("groq 必须配置 api_key")
|
||
}
|
||
|
||
case "grok", "xai":
|
||
if baseURL == "" {
|
||
baseURL = "https://api.x.ai/v1"
|
||
}
|
||
if model == "" {
|
||
model = "grok-2-latest"
|
||
}
|
||
if apiKey == "" {
|
||
return nil, fmt.Errorf("grok/xai 必须配置 api_key")
|
||
}
|
||
|
||
case "claude", "anthropic":
|
||
if baseURL == "" {
|
||
baseURL = "https://openrouter.ai/api/v1"
|
||
}
|
||
if model == "" {
|
||
model = "anthropic/claude-3.7-sonnet"
|
||
}
|
||
if apiKey == "" {
|
||
return nil, fmt.Errorf("claude 必须配置 api_key (默认走 OpenRouter 统一网关，或通过 base_url 指定自建代理端点)")
|
||
}
|
||
|
||
case "openrouter":
|
||
if baseURL == "" {
|
||
baseURL = "https://openrouter.ai/api/v1"
|
||
}
|
||
if model == "" {
|
||
model = "anthropic/claude-3.7-sonnet"
|
||
}
|
||
if apiKey == "" {
|
||
return nil, fmt.Errorf("openrouter 必须配置 api_key")
|
||
}
|
||
|
||
case "ollama":
|
||
if baseURL == "" {
|
||
baseURL = "http://localhost:11434/v1"
|
||
}
|
||
if model == "" {
|
||
model = "qwen2.5-coder"
|
||
}
|
||
if apiKey == "" {
|
||
apiKey = "ollama" // 本地 ollama 传非空占位符即可
|
||
}
|
||
|
||
case "lmstudio", "vllm":
|
||
if baseURL == "" {
|
||
baseURL = "http://localhost:1234/v1"
|
||
}
|
||
if model == "" {
|
||
model = "local-model"
|
||
}
|
||
if apiKey == "" {
|
||
apiKey = "local"
|
||
}
|
||
|
||
case "openai":
|
||
if baseURL == "" {
|
||
baseURL = "https://api.openai.com/v1"
|
||
}
|
||
if model == "" {
|
||
model = "gpt-4o-mini"
|
||
}
|
||
if apiKey == "" {
|
||
return nil, fmt.Errorf("openai 必须配置 api_key")
|
||
}
|
||
|
||
default:
|
||
// custom 自定义端点（只要提供了 baseURL 就能直连）
|
||
if baseURL == "" {
|
||
return nil, fmt.Errorf("未知 Provider: '%s'，如需自定义请提供 base_url", provider)
|
||
}
|
||
if model == "" {
|
||
model = "gpt-4o"
|
||
}
|
||
}
|
||
|
||
clientConfig := openai.DefaultConfig(apiKey)
|
||
clientConfig.BaseURL = baseURL
|
||
|
||
return &Client{
|
||
client: openai.NewClientWithConfig(clientConfig),
|
||
model: model,
|
||
priceInput: cfg.PriceInput,
|
||
priceOutput: cfg.PriceOutput,
|
||
}, nil
|
||
}
|
||
|
||
// isReasoningModel 判断是否为推理或带有 Beta 限制的模型（这类模型要求 temperature 固定为 1）
|
||
func isReasoningModel(model string) bool {
|
||
m := strings.ToLower(model)
|
||
return strings.HasPrefix(m, "o1") ||
|
||
strings.HasPrefix(m, "o3") ||
|
||
strings.HasPrefix(m, "o4") ||
|
||
strings.Contains(m, "reasoner") ||
|
||
strings.Contains(m, "gpt-5") ||
|
||
strings.Contains(m, "luna") ||
|
||
strings.Contains(m, "terra") ||
|
||
strings.Contains(m, "sol")
|
||
}
|
||
|
||
// isTemperatureError 检测 API 报错是否由 temperature 参数限制触发
|
||
func isTemperatureError(err error) bool {
|
||
if err == nil {
|
||
return false
|
||
}
|
||
msg := strings.ToLower(err.Error())
|
||
return strings.Contains(msg, "temperature") || strings.Contains(msg, "beta-limitations")
|
||
}
|
||
|
||
// Review 一次性完整返回代码评审结果
|
||
func (c *Client) Review(ctx context.Context, systemPrompt, diff string) (string, error) {
|
||
temp := float32(0.2)
|
||
if isReasoningModel(c.model) {
|
||
temp = 1.0
|
||
}
|
||
|
||
req := openai.ChatCompletionRequest{
|
||
Model: c.model,
|
||
Messages: []openai.ChatCompletionMessage{
|
||
{Role: openai.ChatMessageRoleSystem, Content: systemPrompt},
|
||
{Role: openai.ChatMessageRoleUser, Content: diff},
|
||
},
|
||
Temperature: temp,
|
||
}
|
||
|
||
resp, err := c.client.CreateChatCompletion(ctx, req)
|
||
// 针对限制了 temperature 的模型，且首次非 1.0 时，自动调整为 1.0 重试
|
||
if err != nil && req.Temperature != 1.0 && isTemperatureError(err) {
|
||
req.Temperature = 1.0
|
||
resp, err = c.client.CreateChatCompletion(ctx, req)
|
||
}
|
||
|
||
if err != nil {
|
||
return "", fmt.Errorf("代码评审调用失败: %w", err)
|
||
}
|
||
|
||
if len(resp.Choices) == 0 {
|
||
return "", fmt.Errorf("模型未返回任何有效评审结果")
|
||
}
|
||
|
||
return resp.Choices[0].Message.Content, nil
|
||
}
|
||
|
||
// Usage 记录一次调用的 Token 消耗
|
||
type Usage struct {
|
||
InputTokens int
|
||
OutputTokens int
|
||
Model string
|
||
Provider string
|
||
CustomPriceInput float64 // 用户自定义输入单价（每 1M tokens 美元，>0 优先使用）
|
||
CustomPriceOutput float64 // 用户自定义输出单价（每 1M tokens 美元，>0 优先使用）
|
||
}
|
||
|
||
// reviewPriceTable 各模型每 1M token 的美元单价（输入/输出）
|
||
// 数据来源：各厂商官方定价页面，更新于 2026-08
|
||
// 如实际价格有变动，可在 config.yaml 中通过 price_input/price_output 字段自定义覆盖
|
||
var reviewPriceTable = map[string][2]float64{
|
||
|
||
// ── Google Gemini ─────────────────────────────────────────────
|
||
// 3.x 系列（当前主力）
|
||
"gemini-3.7-flash": {0.75, 3.75}, // 推介价至 2026-12-31，之后 1.50/7.50
|
||
"gemini-3.6-flash": {1.50, 7.50},
|
||
"gemini-3.5-flash": {1.50, 9.00},
|
||
"gemini-3.5-flash-lite": {0.30, 2.50},
|
||
"gemini-3.1-flash-lite": {0.25, 1.50},
|
||
"gemini-3.1-pro": {2.00, 12.00},
|
||
// 2.5 系列（2026-10 退役）
|
||
"gemini-2.5-pro": {1.25, 10.00},
|
||
"gemini-2.5-flash": {0.30, 2.50},
|
||
"gemini-2.5-flash-lite": {0.10, 0.40},
|
||
// 旧款（兼容）
|
||
"gemini-2.0-flash": {0.10, 0.40},
|
||
|
||
// ── DeepSeek ─────────────────────────────────────────────────
|
||
// 注意：DeepSeek 已启用峰谷定价，以下为非高峰标准价
|
||
"deepseek-v4-flash": {0.22, 0.66},
|
||
"deepseek-chat": {0.22, 0.66}, // V4-Flash 别名
|
||
"deepseek-reasoner": {0.55, 2.19},
|
||
|
||
// ── Anthropic Claude ──────────────────────────────────────────
|
||
"claude-opus-5": {5.00, 25.00},
|
||
"claude-sonnet-5": {2.00, 10.00}, // 推介价至 2026-08-31
|
||
"anthropic/claude-sonnet-5": {2.00, 10.00},
|
||
"claude-haiku-4-5": {1.00, 5.00},
|
||
"anthropic/claude-haiku-4-5": {1.00, 5.00},
|
||
// 旧款（兼容）
|
||
"anthropic/claude-3.7-sonnet": {3.00, 15.00},
|
||
"anthropic/claude-3.5-sonnet": {3.00, 15.00},
|
||
"anthropic/claude-3.5-haiku": {0.80, 4.00},
|
||
|
||
// ── xAI Grok ─────────────────────────────────────────────────
|
||
"grok-4.6": {2.00, 6.00}, // >200k tokens 翻倍
|
||
"grok-4.3": {1.25, 2.50},
|
||
"grok-2-latest": {2.00, 6.00}, // 旧款映射
|
||
|
||
// ── OpenAI ───────────────────────────────────────────────────
|
||
"gpt-5.6-sol": {5.00, 30.00},
|
||
"gpt-5.6-terra": {2.00, 12.00},
|
||
"gpt-5.6-luna": {0.20, 1.20},
|
||
"gpt-4o": {2.50, 10.00}, // 旧款保留
|
||
"gpt-4o-mini": {0.15, 0.60},
|
||
"o3-mini": {1.10, 4.40},
|
||
|
||
// ── SiliconFlow（聚合平台，按 DeepSeek V4 主力估算）────────────
|
||
"deepseek-ai/DeepSeek-V3": {0.22, 0.66},
|
||
"deepseek-ai/DeepSeek-R1": {0.55, 2.19},
|
||
|
||
// ── Qwen / 阿里通义千问 ───────────────────────────────────────
|
||
"qwen-3.8-max": {2.00, 6.00},
|
||
"qwen-max": {2.00, 6.00},
|
||
"qwen-plus": {0.40, 1.20},
|
||
"qwen-3.7-flash": {0.03, 0.13},
|
||
|
||
// ── Kimi / Moonshot AI ────────────────────────────────────────
|
||
"kimi-k3": {3.00, 15.00},
|
||
"moonshot-v1-8k": {0.95, 4.00},
|
||
"moonshot-v1-32k": {1.90, 8.00},
|
||
"moonshot-v1-128k": {3.50, 15.00},
|
||
|
||
// ── 智谱 GLM ─────────────────────────────────────────────────
|
||
"glm-5.3": {1.40, 4.40},
|
||
"glm-5.2": {1.40, 4.40},
|
||
"glm-4-plus": {1.40, 4.40},
|
||
|
||
// ── Doubao / 字节跳动 ─────────────────────────────────────────
|
||
"doubao-seed-2.1-pro": {0.88, 4.42},
|
||
"doubao-pro-32k": {0.88, 4.42},
|
||
|
||
// ── Groq（超快推理，按量计费）────────────────────────────────
|
||
"llama-3.3-70b-versatile": {0.59, 0.79},
|
||
"llama-3.1-8b-instant": {0.05, 0.08},
|
||
|
||
// ── 本地模型（完全免费）──────────────────────────────────────
|
||
"ollama": {0, 0},
|
||
"local-model": {0, 0},
|
||
}
|
||
|
||
// EstimateCost 估算费用（美元），返回 -1 表示无定价数据
|
||
func (u *Usage) EstimateCost() float64 {
|
||
// 1. 优先使用用户在 config.yaml 中配置的自定义单价（每 1M tokens 美元）
|
||
if u.CustomPriceInput > 0 || u.CustomPriceOutput > 0 {
|
||
inputCost := float64(u.InputTokens) / 1_000_000 * u.CustomPriceInput
|
||
outputCost := float64(u.OutputTokens) / 1_000_000 * u.CustomPriceOutput
|
||
return inputCost + outputCost
|
||
}
|
||
|
||
// 2. 回退使用内置官方模型价格表
|
||
prices, ok := reviewPriceTable[u.Model]
|
||
if !ok {
|
||
return -1
|
||
}
|
||
// Ollama / 本地模型 price = 0，直接返回 0
|
||
inputCost := float64(u.InputTokens) / 1_000_000 * prices[0]
|
||
outputCost := float64(u.OutputTokens) / 1_000_000 * prices[1]
|
||
return inputCost + outputCost
|
||
}
|
||
|
||
// ReviewStream 流式逐字返回评审结果（打字机效果），同时返回 Token 使用量
|
||
func (c *Client) ReviewStream(ctx context.Context, systemPrompt, diff string, onChunk func(chunk string)) (*Usage, error) {
|
||
temp := float32(0.2)
|
||
if isReasoningModel(c.model) {
|
||
temp = 1.0
|
||
}
|
||
|
||
req := openai.ChatCompletionRequest{
|
||
Model: c.model,
|
||
Messages: []openai.ChatCompletionMessage{
|
||
{Role: openai.ChatMessageRoleSystem, Content: systemPrompt},
|
||
{Role: openai.ChatMessageRoleUser, Content: diff},
|
||
},
|
||
Temperature: temp,
|
||
Stream: true,
|
||
StreamOptions: &openai.StreamOptions{
|
||
IncludeUsage: true,
|
||
},
|
||
}
|
||
|
||
stream, err := c.client.CreateChatCompletionStream(ctx, req)
|
||
// 针对限制了 temperature 的模型，且首次非 1.0 时，自动调整为 1.0 重试
|
||
if err != nil && req.Temperature != 1.0 && isTemperatureError(err) {
|
||
req.Temperature = 1.0
|
||
stream, err = c.client.CreateChatCompletionStream(ctx, req)
|
||
}
|
||
|
||
if err != nil {
|
||
return nil, fmt.Errorf("创建流式请求失败: %w", err)
|
||
}
|
||
if stream != nil {
|
||
defer stream.Close()
|
||
}
|
||
|
||
usage := &Usage{
|
||
Model: c.model,
|
||
CustomPriceInput: c.priceInput,
|
||
CustomPriceOutput: c.priceOutput,
|
||
}
|
||
|
||
for {
|
||
select {
|
||
case <-ctx.Done():
|
||
return nil, ctx.Err()
|
||
default:
|
||
}
|
||
|
||
response, err := stream.Recv()
|
||
if errors.Is(err, io.EOF) {
|
||
break
|
||
}
|
||
if err != nil {
|
||
return nil, fmt.Errorf("流式读取异常: %w", err)
|
||
}
|
||
|
||
// 捕获最后一个 chunk 中携带的 Usage 信息
|
||
if response.Usage != nil && response.Usage.TotalTokens > 0 {
|
||
usage.InputTokens = response.Usage.PromptTokens
|
||
usage.OutputTokens = response.Usage.CompletionTokens
|
||
}
|
||
|
||
if len(response.Choices) > 0 {
|
||
chunk := response.Choices[0].Delta.Content
|
||
if chunk != "" {
|
||
onChunk(chunk)
|
||
}
|
||
}
|
||
}
|
||
|
||
return usage, nil
|
||
}
|