374 lines
10 KiB
Go
374 lines
10 KiB
Go
package copilot
|
|
|
|
import (
|
|
"bufio"
|
|
"bytes"
|
|
"context"
|
|
"crypto/tls"
|
|
"encoding/json"
|
|
"errors"
|
|
"fmt"
|
|
"io"
|
|
"log"
|
|
"math/rand"
|
|
"net/http"
|
|
"os"
|
|
"strconv"
|
|
"strings"
|
|
"time"
|
|
|
|
"github.com/gin-gonic/gin"
|
|
"github.com/gofrs/uuid"
|
|
"github.com/tidwall/gjson"
|
|
"github.com/tidwall/sjson"
|
|
)
|
|
|
|
// CodeCompletions 代码补全
|
|
func CodeCompletions(c *gin.Context) {
|
|
ctx := c.Request.Context()
|
|
|
|
requestID := uuid.Must(uuid.NewV4()).String()
|
|
c.Header("x-github-request-id", requestID)
|
|
|
|
debounceTime, _ := strconv.Atoi(os.Getenv("COPILOT_DEBOUNCE"))
|
|
time.Sleep(time.Duration(debounceTime) * time.Millisecond)
|
|
|
|
if ctx.Err() != nil {
|
|
abortCodex(c, http.StatusRequestTimeout)
|
|
return
|
|
}
|
|
|
|
body, err := io.ReadAll(c.Request.Body)
|
|
if nil != err {
|
|
abortCodex(c, http.StatusBadRequest)
|
|
return
|
|
}
|
|
|
|
c.Header("Content-Type", "text/event-stream")
|
|
codexServiceType := os.Getenv("CODEX_SERVICE_TYPE")
|
|
body = ConstructRequestBody(body, codexServiceType)
|
|
|
|
req, err := http.NewRequestWithContext(ctx, http.MethodPost, os.Getenv("CODEX_API_BASE"), io.NopCloser(bytes.NewBuffer(body)))
|
|
if nil != err {
|
|
abortCodex(c, http.StatusInternalServerError)
|
|
return
|
|
}
|
|
|
|
req.Header.Set("Content-Type", "application/json")
|
|
|
|
apiKeys := strings.Split(os.Getenv("CODEX_API_KEY"), ",")
|
|
|
|
// 检查 apiKeys 是否有效
|
|
if len(apiKeys) == 0 || (len(apiKeys) == 1 && apiKeys[0] == "") {
|
|
abortCodex(c, http.StatusInternalServerError)
|
|
return
|
|
}
|
|
|
|
randGen := rand.New(rand.NewSource(time.Now().UnixNano()))
|
|
selectedKey := strings.TrimSpace(apiKeys[randGen.Intn(len(apiKeys))])
|
|
|
|
if selectedKey == "" {
|
|
abortCodex(c, http.StatusInternalServerError)
|
|
return
|
|
}
|
|
|
|
req.Header.Set("Authorization", "Bearer "+selectedKey)
|
|
httpClientTimeout, _ := time.ParseDuration(os.Getenv("HTTP_CLIENT_TIMEOUT") + "s")
|
|
client := &http.Client{
|
|
Timeout: httpClientTimeout,
|
|
Transport: &http.Transport{
|
|
TLSClientConfig: &tls.Config{InsecureSkipVerify: true},
|
|
},
|
|
}
|
|
resp, err := client.Do(req)
|
|
if nil != err {
|
|
if errors.Is(err, context.Canceled) {
|
|
abortCodex(c, http.StatusRequestTimeout)
|
|
return
|
|
}
|
|
|
|
log.Println("request completions failed:", err.Error())
|
|
abortCodex(c, http.StatusInternalServerError)
|
|
return
|
|
}
|
|
defer CloseIO(resp.Body)
|
|
|
|
if resp.StatusCode != http.StatusOK {
|
|
body, _ := io.ReadAll(resp.Body)
|
|
log.Println("request completions failed:", string(body))
|
|
|
|
abortCodex(c, resp.StatusCode)
|
|
return
|
|
}
|
|
|
|
c.Status(resp.StatusCode)
|
|
// 处理 Ollama 服务的流式响应
|
|
if codexServiceType == "ollama" {
|
|
reader := bufio.NewReader(resp.Body)
|
|
for {
|
|
line, err := reader.ReadString('\n')
|
|
if err != nil {
|
|
if err == io.EOF {
|
|
break
|
|
}
|
|
break
|
|
}
|
|
|
|
if strings.TrimSpace(line) == "" {
|
|
continue
|
|
}
|
|
|
|
// json解析 line
|
|
lineJson := gjson.Parse(line)
|
|
uuid := uuid.Must(uuid.NewV4()).String()
|
|
done := lineJson.Get("done").Bool()
|
|
doneReason := lineJson.Get("done_reason").Str
|
|
response := lineJson.Get("response").Str
|
|
timestamp := time.Now().Unix()
|
|
choice := map[string]interface{}{
|
|
"text": response,
|
|
"index": 0,
|
|
"logprobs": nil,
|
|
"finish_reason": doneReason,
|
|
}
|
|
choices := []map[string]interface{}{choice}
|
|
constructLineData := map[string]interface{}{
|
|
"id": uuid,
|
|
"choices": choices,
|
|
"created": timestamp,
|
|
"model": lineJson.Get("model").Str,
|
|
"system_fingerprint": "fp_1c141eb703",
|
|
"object": "text_completion",
|
|
}
|
|
|
|
if done && strings.Contains(doneReason, "stop") {
|
|
usage := map[string]interface{}{
|
|
"prompt_tokens": lineJson.Get("prompt_eval_count").Int(),
|
|
"completion_tokens": lineJson.Get("eval_count").Int(),
|
|
"total_tokens": lineJson.Get("prompt_eval_count").Int(),
|
|
"prompt_cache_hit_tokens": lineJson.Get("prompt_eval_count").Int(),
|
|
"prompt_cache_miss_tokens": lineJson.Get("eval_count").Int(),
|
|
}
|
|
constructLineData["usage"] = usage
|
|
}
|
|
|
|
// 将修改后的数据重新编码为 JSON
|
|
modifiedJSON, err := json.Marshal(constructLineData)
|
|
if err != nil {
|
|
continue
|
|
}
|
|
|
|
// 发送修改后的数据
|
|
_, _ = c.Writer.WriteString("data: " + string(modifiedJSON) + "\n\n")
|
|
c.Writer.Flush()
|
|
}
|
|
|
|
_, _ = c.Writer.WriteString("data: [DONE]\n\n")
|
|
c.Writer.Flush()
|
|
return
|
|
}
|
|
|
|
// 处理默认服务的响应
|
|
_, _ = io.Copy(c.Writer, resp.Body)
|
|
}
|
|
|
|
// ConstructRequestBody 重新构建请求体
|
|
func ConstructRequestBody(body []byte, codexServiceType string) []byte {
|
|
envCodexModel := os.Getenv("CODEX_API_MODEL_NAME")
|
|
body, _ = sjson.SetBytes(body, "model", envCodexModel)
|
|
body, _ = sjson.SetBytes(body, "stream", true) // 强制流式输出
|
|
body, _ = sjson.DeleteBytes(body, "extra")
|
|
body, _ = sjson.DeleteBytes(body, "nwo")
|
|
|
|
// 限制 prompt 和 suffix 的长度
|
|
body = applyPromptLengthLimit(body)
|
|
|
|
temperature, _ := strconv.ParseFloat(os.Getenv("CODEX_TEMPERATURE"), 64)
|
|
if temperature != -1 {
|
|
body, _ = sjson.SetBytes(body, "temperature", temperature)
|
|
}
|
|
|
|
codeMaxTokens, _ := strconv.Atoi(os.Getenv("CODEX_MAX_TOKENS"))
|
|
if int(gjson.GetBytes(body, "max_tokens").Int()) > codeMaxTokens {
|
|
body, _ = sjson.SetBytes(body, "max_tokens", codeMaxTokens)
|
|
}
|
|
|
|
if gjson.GetBytes(body, "n").Int() > 1 {
|
|
body, _ = sjson.SetBytes(body, "n", 1)
|
|
}
|
|
|
|
// https://ollama.com/library/stable-code || https://ollama.com/library/codegemma
|
|
if strings.Contains(envCodexModel, "stable-code") || strings.Contains(envCodexModel, "codegemma") {
|
|
return constructWithStableCodeModel(body)
|
|
}
|
|
|
|
// https://ollama.com/library/codellama
|
|
if strings.Contains(envCodexModel, "codellama") {
|
|
return constructWithCodeLlamaModel(body)
|
|
}
|
|
|
|
// https://help.aliyun.com/zh/model-studio/user-guide/qwen-coder?spm=a2c4g.11186623.0.0.a5234823I6LvAG
|
|
if strings.Contains(envCodexModel, "qwen-coder-turbo") {
|
|
return constructWithQwenCoderTurboModel(body)
|
|
}
|
|
|
|
// 支持 Ollama FIM 的模型, 如:https://ollama.com/library/deepseek-coder-v2
|
|
if codexServiceType == "ollama" {
|
|
return constructWithOllamaModel(body, codeMaxTokens)
|
|
}
|
|
|
|
return body
|
|
}
|
|
|
|
// applyPromptLengthLimit 对 prompt 和 suffix 应用长度限制
|
|
func applyPromptLengthLimit(body []byte) []byte {
|
|
envLimitPrompt := os.Getenv("CODEX_LIMIT_PROMPT")
|
|
limitPrompt, err := strconv.Atoi(envLimitPrompt)
|
|
if err != nil || limitPrompt <= 0 {
|
|
return body
|
|
}
|
|
|
|
body = limitPromptLength(body, limitPrompt)
|
|
body = limitSuffixLength(body, limitPrompt)
|
|
|
|
return body
|
|
}
|
|
|
|
// limitPromptLength 限制 prompt 长度
|
|
func limitPromptLength(body []byte, limitRows int) []byte {
|
|
prompt := gjson.GetBytes(body, "prompt")
|
|
if !prompt.Exists() {
|
|
return body
|
|
}
|
|
|
|
rows := strings.Split(prompt.Str, "\n")
|
|
if len(rows) <= limitRows {
|
|
return body
|
|
}
|
|
|
|
// 保留后面的内容
|
|
newPrompt := strings.Join(rows[len(rows)-limitRows:], "\n")
|
|
body, _ = sjson.SetBytes(body, "prompt", newPrompt)
|
|
|
|
return body
|
|
}
|
|
|
|
// limitSuffixLength 限制 suffix 长度
|
|
func limitSuffixLength(body []byte, limitRows int) []byte {
|
|
suffix := gjson.GetBytes(body, "suffix")
|
|
if !suffix.Exists() {
|
|
return body
|
|
}
|
|
|
|
rows := strings.Split(suffix.Str, "\n")
|
|
if len(rows) <= limitRows {
|
|
return body
|
|
}
|
|
|
|
// 保留前面的内容
|
|
newSuffix := strings.Join(rows[:limitRows], "\n")
|
|
body, _ = sjson.SetBytes(body, "suffix", newSuffix)
|
|
|
|
return body
|
|
}
|
|
|
|
// constructWithCodeLlamaModel 重写codeLlama模型要求的请求体
|
|
func constructWithCodeLlamaModel(body []byte) []byte {
|
|
suffix := gjson.GetBytes(body, "suffix")
|
|
prompt := gjson.GetBytes(body, "prompt")
|
|
content := fmt.Sprintf("<PRE> %s <SUF> %s <MID>", prompt, suffix)
|
|
|
|
return constructWithChatModel(body, content)
|
|
}
|
|
|
|
// constructWithStableCodeModel 重写StableCode模型要求的请求体
|
|
func constructWithStableCodeModel(body []byte) []byte {
|
|
suffix := gjson.GetBytes(body, "suffix")
|
|
prompt := gjson.GetBytes(body, "prompt")
|
|
content := fmt.Sprintf("<fim_prefix>%s<fim_suffix>%s<fim_middle>", prompt, suffix)
|
|
|
|
return constructWithChatModel(body, content)
|
|
}
|
|
|
|
// constructWithChatModel 重写Chat请求体
|
|
func constructWithChatModel(body []byte, content string) []byte {
|
|
// 创建新的 JSON 对象并添加到 body 中
|
|
messages := []map[string]string{
|
|
{
|
|
"role": "user",
|
|
"content": content,
|
|
},
|
|
}
|
|
|
|
body, _ = sjson.SetBytes(body, "messages", messages)
|
|
|
|
jsonStr := string(body)
|
|
jsonStr = strings.ReplaceAll(jsonStr, "\\u003c", "<")
|
|
jsonStr = strings.ReplaceAll(jsonStr, "\\u003e", ">")
|
|
return []byte(jsonStr)
|
|
}
|
|
|
|
// constructWithQwenCoderTurboModel 重写QwenCoderTurbo模型要求的请求体
|
|
func constructWithQwenCoderTurboModel(body []byte) []byte {
|
|
if gjson.GetBytes(body, "n").Int() > 1 {
|
|
body, _ = sjson.SetBytes(body, "n", 1)
|
|
}
|
|
suffix := gjson.GetBytes(body, "suffix")
|
|
prompt := gjson.GetBytes(body, "prompt")
|
|
codeLanguage := gjson.GetBytes(body, "extra.language")
|
|
|
|
messages := []map[string]interface{}{
|
|
{
|
|
"role": "system",
|
|
"content": "You are an expert in " + codeLanguage.Str + " programming, highly skilled at understanding and continuing to write code.",
|
|
},
|
|
{
|
|
"role": "user",
|
|
"content": "Combined with subsequent code snippets, help me complete the code:\n\n" +
|
|
"Code subsequent content:\n```" + codeLanguage.Str + "\n" + suffix.Str + "```\n\n" +
|
|
"Remember:\n" +
|
|
"- Do not generate content outside of the code.\n" +
|
|
"- Do not directly fill in all the code content, the maximum number of lines of code should not exceed 5 lines.\n" +
|
|
"- Answer must refer to the code suffix content, do not exceed the boundary, otherwise repeated code will occur.\n" +
|
|
"- If you don't know how to answer, just reply with an empty string.",
|
|
},
|
|
{
|
|
"role": "assistant",
|
|
"content": prompt.Str,
|
|
"partial": true,
|
|
},
|
|
}
|
|
|
|
body, _ = sjson.SetBytes(body, "messages", messages)
|
|
body, _ = sjson.DeleteBytes(body, "prompt")
|
|
return body
|
|
}
|
|
|
|
// constructWithOllamaModel 重写Ollama模型要求的请求体
|
|
func constructWithOllamaModel(body []byte, codeMaxTokens int) []byte {
|
|
body, _ = sjson.SetBytes(body, "options.temperature", 0)
|
|
// stop参数处理
|
|
stopArray := gjson.GetBytes(body, "stop").Array()
|
|
stopSlice := make([]interface{}, len(stopArray))
|
|
for i, v := range stopArray {
|
|
stopSlice[i] = v.String()
|
|
}
|
|
body, _ = sjson.SetBytes(body, "options.stop", stopSlice)
|
|
body, _ = sjson.SetBytes(body, "stream", true)
|
|
|
|
maxTokens := gjson.GetBytes(body, "max_tokens").Int()
|
|
if int(maxTokens) > codeMaxTokens {
|
|
body, _ = sjson.SetBytes(body, "options.num_predict", codeMaxTokens)
|
|
} else {
|
|
body, _ = sjson.SetBytes(body, "options.num_predict", maxTokens)
|
|
}
|
|
return body
|
|
}
|
|
|
|
// abortCodex 中断请求
|
|
func abortCodex(c *gin.Context, status int) {
|
|
c.Header("Content-Type", "text/event-stream")
|
|
c.String(status, "data: [DONE]\n\n")
|
|
c.Abort()
|
|
}
|