264 lines
9.2 KiB
Go
264 lines
9.2 KiB
Go
|
package main
|
|||
|
|
|
|||
|
|
import (
|
|||
|
|
"bytes"
|
|||
|
|
"encoding/json"
|
|||
|
|
"flag"
|
|||
|
|
"fmt"
|
|||
|
|
"io"
|
|||
|
|
"net/http"
|
|||
|
|
"strings"
|
|||
|
|
"time"
|
|||
|
|
)
|
|||
|
|
|
|||
|
|
const moderationSystemPrompt = `你是中文社区的内容审核助手，负责对"帖子标题、正文、配图"做联合审核。目标是平衡社区安全与正常交流：必须拦截高风险违规内容，但不要误伤正常玩梗、二创、吐槽和轻度调侃。请只输出指定JSON。
|
|||
|
|
|
|||
|
|
审核流程：
|
|||
|
|
1) 先判断是否命中硬性违规；
|
|||
|
|
2) 再判断语境（玩笑/自嘲/朋友间互动/作品讨论）；
|
|||
|
|
3) 做文图交叉判断（文本+图片合并理解）；
|
|||
|
|
4) 给出 approved 与简短 reason。
|
|||
|
|
|
|||
|
|
硬性违规（命中任一项必须 approved=false）：
|
|||
|
|
A. 宣传对立与煽动撕裂：
|
|||
|
|
- 明确煽动群体对立、地域对立、性别对立、民族宗教对立，鼓动仇恨、排斥、报复。
|
|||
|
|
B. 严重人身攻击与网暴引导：
|
|||
|
|
- 持续性侮辱贬损、羞辱人格、号召围攻/骚扰/挂人/线下冲突。
|
|||
|
|
C. 开盒/人肉/隐私暴露：
|
|||
|
|
- 故意公开、拼接、索取他人可识别隐私信息（姓名+联系方式、身份证号、住址、学校单位、车牌、定位轨迹等）；
|
|||
|
|
- 图片/截图中出现可识别隐私信息并伴随曝光意图，也按违规处理。
|
|||
|
|
D. 其他高危违规：
|
|||
|
|
- 违法犯罪、暴力威胁、极端仇恨、色情低俗、诈骗引流、恶意广告等。
|
|||
|
|
|
|||
|
|
放行规则（以下通常 approved=true）：
|
|||
|
|
- 正常玩梗、表情包、谐音梗、二次创作、无恶意的吐槽；
|
|||
|
|
- 非定向、轻度口语化吐槽（无明确攻击对象、无网暴号召、无隐私暴露）；
|
|||
|
|
- 对社会事件/作品的理性讨论、观点争论（即使语气尖锐，但未煽动对立或人身攻击）。
|
|||
|
|
|
|||
|
|
边界判定：
|
|||
|
|
- 若只是"梗文化表达"且不指向现实伤害，优先通过；
|
|||
|
|
- 若存在明确伤害意图（煽动、围攻、曝光隐私），必须拒绝；
|
|||
|
|
- 对模糊内容不因个别粗口直接拒绝，需结合对象、意图、号召性和可执行性综合判断。
|
|||
|
|
|
|||
|
|
reason 要求：
|
|||
|
|
- approved=false 时：中文10-30字，说明核心违规点；
|
|||
|
|
- approved=true 时：reason 为空字符串。
|
|||
|
|
|
|||
|
|
输出格式（严格）：
|
|||
|
|
仅输出一行JSON对象，不要Markdown，不要额外解释：
|
|||
|
|
{"approved": true/false, "reason": "..."}`
|
|||
|
|
|
|||
|
|
type chatMessage struct {
|
|||
|
|
Role string `json:"role"`
|
|||
|
|
Content interface{} `json:"content"`
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
type contentPart struct {
|
|||
|
|
Type string `json:"type"`
|
|||
|
|
Text string `json:"text,omitempty"`
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
type chatCompletionsRequest struct {
|
|||
|
|
Model string `json:"model"`
|
|||
|
|
Messages []chatMessage `json:"messages"`
|
|||
|
|
Temperature float64 `json:"temperature,omitempty"`
|
|||
|
|
MaxTokens int `json:"max_tokens,omitempty"`
|
|||
|
|
EnableThinking *bool `json:"enable_thinking,omitempty"` // qwen3.5思考模式控制
|
|||
|
|
ThinkingBudget *int `json:"thinking_budget,omitempty"` // 思考过程最大token数
|
|||
|
|
ResponseFormat *responseFormatConfig `json:"response_format,omitempty"` // 响应格式
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
type responseFormatConfig struct {
|
|||
|
|
Type string `json:"type"` // "text" or "json_object"
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
type chatCompletionsResponse struct {
|
|||
|
|
Choices []struct {
|
|||
|
|
Message struct {
|
|||
|
|
Content string `json:"content"`
|
|||
|
|
} `json:"message"`
|
|||
|
|
FinishReason string `json:"finish_reason"`
|
|||
|
|
} `json:"choices"`
|
|||
|
|
Usage struct {
|
|||
|
|
PromptTokens int `json:"prompt_tokens"`
|
|||
|
|
CompletionTokens int `json:"completion_tokens"`
|
|||
|
|
TotalTokens int `json:"total_tokens"`
|
|||
|
|
} `json:"usage"`
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
func main() {
|
|||
|
|
baseURL := flag.String("url", "https://api.littlelan.cn/", "API base URL")
|
|||
|
|
apiKey := flag.String("key", "", "API key")
|
|||
|
|
model := flag.String("model", "qwen3.5-plus", "Model name")
|
|||
|
|
maxTokens := flag.Int("max-tokens", 220, "Max tokens for completion")
|
|||
|
|
enableThinking := flag.Bool("enable-thinking", false, "Enable thinking mode for qwen3.5")
|
|||
|
|
flag.Parse()
|
|||
|
|
|
|||
|
|
if *apiKey == "" {
|
|||
|
|
fmt.Println("Error: API key is required. Use -key flag")
|
|||
|
|
return
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
// 测试用例
|
|||
|
|
testCases := []struct {
|
|||
|
|
name string
|
|||
|
|
content string
|
|||
|
|
}{
|
|||
|
|
{
|
|||
|
|
name: "简单正常内容",
|
|||
|
|
content: "帖子标题：今天天气真好\n帖子内容：出门散步，心情愉快！",
|
|||
|
|
},
|
|||
|
|
{
|
|||
|
|
name: "中等长度内容",
|
|||
|
|
content: "帖子标题：分享我的学习经验\n帖子内容：最近在学习Go语言，发现这门语言真的很适合后端开发。并发处理特别方便，goroutine和channel的设计非常优雅。有一起学习的小伙伴吗？",
|
|||
|
|
},
|
|||
|
|
{
|
|||
|
|
name: "较长内容",
|
|||
|
|
content: "帖子标题：关于校园生活的一些思考\n帖子内容：大学四年转眼就过去了，回想起来有很多感慨。刚入学的时候什么都不懂，现在感觉自己成长了很多。在这里想分享一些自己的经验，希望能对学弟学妹们有所帮助。首先是学习方面，一定要认真听课，做好笔记。其次是社交方面，多参加社团活动，结交志同道合的朋友。最后是规划方面，早点想清楚自己想做什么，为之努力。",
|
|||
|
|
},
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
client := &http.Client{Timeout: 120 * time.Second}
|
|||
|
|
|
|||
|
|
fmt.Println("============================================")
|
|||
|
|
fmt.Printf("模型: %s\n", *model)
|
|||
|
|
fmt.Printf("API URL: %s\n", *baseURL)
|
|||
|
|
fmt.Printf("MaxTokens 设置: %d\n", *maxTokens)
|
|||
|
|
fmt.Printf("EnableThinking: %v\n", *enableThinking)
|
|||
|
|
fmt.Println("============================================")
|
|||
|
|
|
|||
|
|
for _, tc := range testCases {
|
|||
|
|
fmt.Printf("\n========== 测试: %s ==========\n", tc.name)
|
|||
|
|
fmt.Printf("内容长度: %d 字符\n", len(tc.content))
|
|||
|
|
|
|||
|
|
userPrompt := fmt.Sprintf("%s\n图片批次：1/1（本次仅提供当前批次图片）", tc.content)
|
|||
|
|
|
|||
|
|
reqBody := chatCompletionsRequest{
|
|||
|
|
Model: *model,
|
|||
|
|
Messages: []chatMessage{
|
|||
|
|
{Role: "system", Content: moderationSystemPrompt},
|
|||
|
|
{Role: "user", Content: []contentPart{{Type: "text", Text: userPrompt}}},
|
|||
|
|
},
|
|||
|
|
Temperature: 0.1,
|
|||
|
|
MaxTokens: *maxTokens,
|
|||
|
|
}
|
|||
|
|
// 设置思考模式
|
|||
|
|
if !*enableThinking {
|
|||
|
|
reqBody.EnableThinking = enableThinking
|
|||
|
|
// 设置思考预算为0，完全禁用思考
|
|||
|
|
zero := 0
|
|||
|
|
reqBody.ThinkingBudget = &zero
|
|||
|
|
}
|
|||
|
|
// 使用JSON输出格式
|
|||
|
|
reqBody.ResponseFormat = &responseFormatConfig{Type: "json_object"}
|
|||
|
|
|
|||
|
|
data, err := json.Marshal(reqBody)
|
|||
|
|
if err != nil {
|
|||
|
|
fmt.Printf("Error marshaling request: %v\n", err)
|
|||
|
|
continue
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
endpoint := strings.TrimRight(*baseURL, "/") + "/v1/chat/completions"
|
|||
|
|
if strings.HasSuffix(strings.TrimRight(*baseURL, "/"), "/v1") {
|
|||
|
|
endpoint = strings.TrimRight(*baseURL, "/") + "/chat/completions"
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
req, err := http.NewRequest(http.MethodPost, endpoint, bytes.NewReader(data))
|
|||
|
|
if err != nil {
|
|||
|
|
fmt.Printf("Error creating request: %v\n", err)
|
|||
|
|
continue
|
|||
|
|
}
|
|||
|
|
req.Header.Set("Content-Type", "application/json")
|
|||
|
|
req.Header.Set("Authorization", "Bearer "+*apiKey)
|
|||
|
|
|
|||
|
|
start := time.Now()
|
|||
|
|
resp, err := client.Do(req)
|
|||
|
|
if err != nil {
|
|||
|
|
fmt.Printf("Error sending request: %v\n", err)
|
|||
|
|
continue
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
body, err := io.ReadAll(resp.Body)
|
|||
|
|
resp.Body.Close()
|
|||
|
|
if err != nil {
|
|||
|
|
fmt.Printf("Error reading response: %v\n", err)
|
|||
|
|
continue
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
elapsed := time.Since(start)
|
|||
|
|
|
|||
|
|
if resp.StatusCode >= 400 {
|
|||
|
|
fmt.Printf("API Error: status=%d, body=%s\n", resp.StatusCode, string(body))
|
|||
|
|
continue
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
var parsed chatCompletionsResponse
|
|||
|
|
if err := json.Unmarshal(body, &parsed); err != nil {
|
|||
|
|
fmt.Printf("Error parsing response: %v\n", err)
|
|||
|
|
fmt.Printf("Raw response: %s\n", string(body))
|
|||
|
|
continue
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
if len(parsed.Choices) == 0 {
|
|||
|
|
fmt.Println("No choices in response")
|
|||
|
|
fmt.Printf("Raw response: %s\n", string(body))
|
|||
|
|
continue
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
fmt.Printf("响应时间: %v\n", elapsed)
|
|||
|
|
fmt.Printf("Finish Reason: %s\n", parsed.Choices[0].FinishReason)
|
|||
|
|
fmt.Printf("Token使用情况:\n")
|
|||
|
|
fmt.Printf(" - PromptTokens: %d\n", parsed.Usage.PromptTokens)
|
|||
|
|
fmt.Printf(" - CompletionTokens: %d\n", parsed.Usage.CompletionTokens)
|
|||
|
|
fmt.Printf(" - TotalTokens: %d\n", parsed.Usage.TotalTokens)
|
|||
|
|
|
|||
|
|
output := parsed.Choices[0].Message.Content
|
|||
|
|
fmt.Printf("输出内容长度: %d 字符\n", len(output))
|
|||
|
|
|
|||
|
|
// 检查输出是否符合预期
|
|||
|
|
if parsed.Usage.CompletionTokens > *maxTokens {
|
|||
|
|
fmt.Printf("\n⚠️ 警告: CompletionTokens (%d) 超过了 max_tokens 设置 (%d)!\n",
|
|||
|
|
parsed.Usage.CompletionTokens, *maxTokens)
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
if len(output) > 500 {
|
|||
|
|
fmt.Printf("\n⚠️ 警告: 输出内容过长! 长度=%d\n", len(output))
|
|||
|
|
fmt.Printf("前500字符:\n%s...\n", output[:min(500, len(output))])
|
|||
|
|
} else {
|
|||
|
|
fmt.Printf("输出内容: %s\n", output)
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
// 尝试解析JSON
|
|||
|
|
extractJSONObject := func(raw string) string {
|
|||
|
|
text := strings.TrimSpace(raw)
|
|||
|
|
start := strings.Index(text, "{")
|
|||
|
|
end := strings.LastIndex(text, "}")
|
|||
|
|
if start >= 0 && end > start {
|
|||
|
|
return text[start : end+1]
|
|||
|
|
}
|
|||
|
|
return text
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
jsonStr := extractJSONObject(output)
|
|||
|
|
var result struct {
|
|||
|
|
Approved bool `json:"approved"`
|
|||
|
|
Reason string `json:"reason"`
|
|||
|
|
}
|
|||
|
|
if err := json.Unmarshal([]byte(jsonStr), &result); err != nil {
|
|||
|
|
fmt.Printf("\n⚠️ 警告: 无法解析JSON输出: %v\n", err)
|
|||
|
|
fmt.Printf("提取的JSON: %s\n", jsonStr)
|
|||
|
|
} else {
|
|||
|
|
fmt.Printf("\n✓ 解析成功: approved=%v, reason=\"%s\"\n", result.Approved, result.Reason)
|
|||
|
|
}
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
fmt.Println("\n========== 测试完成 ==========")
|
|||
|
|
}
|
|||
|
|
|
|||
|
|
func min(a, b int) int {
|
|||
|
|
if a < b {
|
|||
|
|
return a
|
|||
|
|
}
|
|||
|
|
return b
|
|||
|
|
}
|