功能:接入 prompt_tokens_details.cached_tokens,消息与模型统计展示缓存命中
- engine: assistant 落库与 SSE 事件带 usage {pt,ct,cached}(上游未报 details 时 cached=null 区分未上报与 0 命中);recordModelUsage 写 cached_tokens;modelUsageStats 聚合 today_cached/total_cached
- db: messages 加 usage TEXT、model_usage 加 cached_tokens 列(建表+迁移双路径,幂等)
- server: /api/pool 统计行透传 today_cached/total_cached
- 前端 msg-actions: 时间戳后新增「In: 1.2k/95%, Out: 456」小字(title 显示精确 tok 数),历史/SSE/摘要卡片三处渲染同参
- 模型卡新增「缓存命中」行:累计缓存 tok · 历史命中率(累计缓存÷累计输入)
This commit is contained in:
@@ -70,8 +70,10 @@ app.get('/api/models', (req, res) => {
|
||||
const s = stats[r.id] || {};
|
||||
r.today_prompt = s.today_prompt || 0;
|
||||
r.today_completion = s.today_completion || 0;
|
||||
r.today_cached = s.today_cached || 0;
|
||||
r.total_prompt = s.total_prompt || 0;
|
||||
r.total_completion = s.total_completion || 0;
|
||||
r.total_cached = s.total_cached || 0;
|
||||
}
|
||||
res.json(rows);
|
||||
});
|
||||
|
||||
Reference in New Issue
Block a user