优化: token分项显示(in/cache/out/reasoning)+ 详情面板 + base前置
token分项(各计费不同,不显 total):df-ai 解析 provider cache/reasoning(openai_compat prompt_cache_hit/miss/reasoning_tokens + anthropic cache_read/creation)+ TokenUsage 加字段(全构造点)+ AiMessage/AiCompleted/DB V39(ai_messages 加 cache_hit/miss/reasoning 列)+ message_repo 映射(持久化)+ 前端 MessageList 显 in·cache·out·reason(in=cache_miss 全价,reasoning 有才显)+ 点击 token 弹详情面板(完整 usage+缓存命中率+model)+ df-miniapp 同步 base前置(提升 prompt cache 命中率):chat.rs aug 拼 base 后(4处)+ knowledge_inject 知识拼 base 后(固定 base 前缀,cache 命中) 附修:replace_conversation 原 13 列 INSERT 丢消息级 token → 改 18 列
This commit is contained in:
@@ -100,6 +100,50 @@ function formatTokens(n: number): string {
|
||||
return n >= 1000 ? `${(n / 1000).toFixed(1)}k` : String(n)
|
||||
}
|
||||
|
||||
// ── token 分项显示(2026-08-02):in/cache/out/reason 分计费 + 详情面板 ──
|
||||
//
|
||||
// 各 provider 计费不同(deepseek:cache 命中低价/未命中全价/输出价高/reasoning 隐藏输出),
|
||||
// 故不显 total(三者相加无意义)。in=cache_miss(全价真实),cache=cache_hit(命中),out=completion,
|
||||
// reasoning > 0 才显(reason 后缀)。
|
||||
//
|
||||
// 详情面板(tokenPopoverMsgId 控制显隐):点击 token 区弹出,含完整 usage + 流程(模型/缓存命中率)。
|
||||
// 流程数据从 message 取(model 在 msg.model;cache 命中率 = hit/(hit+miss))。
|
||||
const tokenPopoverMsgId = ref<string | null>(null)
|
||||
|
||||
/** 取消息的 in token(全价输入)= prompt_tokens(总) - cache_hit(命中低价)。
|
||||
* 统一 GLM(无 cache,prompt_tokens 即全价)与 deepseek(prompt=hit+miss,全价=miss=prompt-hit)。
|
||||
* 原 cache_miss ?? prompt 因 agent 设 cache_miss=0(GLM 不报)致 ?? 不触发返 0,改减法。 */
|
||||
function tokenInOf(m: AiMessage): number {
|
||||
const prompt = m.tokenUsage?.prompt ?? m.prompt_tokens ?? 0
|
||||
const hit = m.tokenUsage?.cache_hit ?? m.prompt_cache_hit_tokens ?? 0
|
||||
return Math.max(0, prompt - hit)
|
||||
}
|
||||
/** 取消息的 cache hit token */
|
||||
function tokenCacheOf(m: AiMessage): number {
|
||||
return m.tokenUsage?.cache_hit ?? m.prompt_cache_hit_tokens ?? 0
|
||||
}
|
||||
/** 取消息的 out token */
|
||||
function tokenOutOf(m: AiMessage): number {
|
||||
return m.tokenUsage?.completion ?? m.completion_tokens ?? 0
|
||||
}
|
||||
/** 取消息的 reasoning token */
|
||||
function tokenReasonOf(m: AiMessage): number {
|
||||
return m.tokenUsage?.reasoning ?? m.reasoning_tokens ?? 0
|
||||
}
|
||||
/** 计算 cache 命中率(0-100),hit+miss=0 时返回 null(无 cache 数据) */
|
||||
function cacheHitRate(m: AiMessage): number | null {
|
||||
const hit = tokenCacheOf(m)
|
||||
const miss = tokenInOf(m)
|
||||
const sum = hit + miss
|
||||
return sum > 0 ? Math.round((hit / sum) * 100) : null
|
||||
}
|
||||
/** 切换 token 详情面板显隐(同消息再点关,不同消息切) */
|
||||
function toggleTokenPopover(m: AiMessage, e: Event): void {
|
||||
e.stopPropagation()
|
||||
const id = m.id
|
||||
tokenPopoverMsgId.value = tokenPopoverMsgId.value === id ? null : id
|
||||
}
|
||||
|
||||
// ── 滚动跟随 / 回到底部(B-260618-24 跟随意图锁存,已抽取至 useMessageScroll) ──
|
||||
const {
|
||||
showBackToBottom,
|
||||
@@ -616,8 +660,15 @@ watch(() => store.state.activeConversationId, () => {
|
||||
// streaming 翻 false 的 watch 已清 rafId,此为中途卸载兜底。
|
||||
onBeforeUnmount(() => {
|
||||
cancelPendingRaf()
|
||||
document.removeEventListener('click', closeTokenPopoverOnOutsideClick)
|
||||
})
|
||||
|
||||
// token 详情面板:点击外部关闭(token 区内的 @click.stop 已阻冒泡,故文档级点击必为外部)
|
||||
function closeTokenPopoverOnOutsideClick(): void {
|
||||
tokenPopoverMsgId.value = null
|
||||
}
|
||||
document.addEventListener('click', closeTokenPopoverOnOutsideClick)
|
||||
|
||||
// Markdown 预热(父原 loadMarkdown 在 onMounted 调,子组件同样幂等——useMarkdown 单例,
|
||||
// 多次调用安全,确保子组件挂载即预热)。
|
||||
loadMarkdown()
|
||||
@@ -761,7 +812,9 @@ defineExpose({
|
||||
</div>
|
||||
|
||||
|
||||
<!-- token 用量(显示在每条 assistant 消息底部,有数据时)。
|
||||
<!-- token 分项显示(2026-08-02):in/cache/out/reason 分计费,不显 total(相加无意义)。
|
||||
in=cache_miss(全价真实,非 prompt 总),cache=cache_hit(命中低价),out=completion,
|
||||
reasoning > 0 才显。点击 token 区弹详情面板(完整 usage + 流程:模型/缓存命中率)。
|
||||
tokenUsage(内存,AiCompleted 实时)优先;fallback prompt_tokens/completion_tokens
|
||||
(DB 持久化字段,reload 自动有)—— 不依赖某个 reload 映射点,压缩/切会话都生效。 -->
|
||||
<div
|
||||
@@ -769,7 +822,53 @@ defineExpose({
|
||||
class="ai-token-usage"
|
||||
>
|
||||
<span class="ai-token-usage-icon">🔣</span>
|
||||
<span>{{ formatTokens(item.msg.tokenUsage?.prompt ?? item.msg.prompt_tokens ?? 0) }} in · {{ formatTokens(item.msg.tokenUsage?.completion ?? item.msg.completion_tokens ?? 0) }} out</span>
|
||||
<button
|
||||
class="ai-token-usage-trigger"
|
||||
:aria-expanded="tokenPopoverMsgId === item.msg.id"
|
||||
@click="toggleTokenPopover(item.msg, $event)"
|
||||
>
|
||||
<span>{{ formatTokens(tokenInOf(item.msg)) }} in</span>
|
||||
<span class="ai-token-sep">·</span>
|
||||
<span>{{ formatTokens(tokenCacheOf(item.msg)) }} cache</span>
|
||||
<span class="ai-token-sep">·</span>
|
||||
<span>{{ formatTokens(tokenOutOf(item.msg)) }} out</span>
|
||||
<template v-if="tokenReasonOf(item.msg) > 0">
|
||||
<span class="ai-token-sep">·</span>
|
||||
<span>{{ formatTokens(tokenReasonOf(item.msg)) }} reason</span>
|
||||
</template>
|
||||
</button>
|
||||
<!-- 详情面板(absolute/right 0/popout 样式,复用 TopBar 风格):完整 usage + 流程 -->
|
||||
<div
|
||||
v-if="tokenPopoverMsgId === item.msg.id"
|
||||
class="ai-token-popover"
|
||||
@click.stop
|
||||
>
|
||||
<div class="ai-token-popover-title">Token 用量详情</div>
|
||||
<div class="ai-token-popover-row">
|
||||
<span class="ai-token-popover-label">输入(未命中,全价)</span>
|
||||
<span class="ai-token-popover-val">{{ formatTokens(tokenInOf(item.msg)) }} ({{ tokenInOf(item.msg) }})</span>
|
||||
</div>
|
||||
<div class="ai-token-popover-row">
|
||||
<span class="ai-token-popover-label">缓存命中(低价)</span>
|
||||
<span class="ai-token-popover-val">{{ formatTokens(tokenCacheOf(item.msg)) }} ({{ tokenCacheOf(item.msg) }})</span>
|
||||
</div>
|
||||
<div class="ai-token-popover-row">
|
||||
<span class="ai-token-popover-label">输出</span>
|
||||
<span class="ai-token-popover-val">{{ formatTokens(tokenOutOf(item.msg)) }} ({{ tokenOutOf(item.msg) }})</span>
|
||||
</div>
|
||||
<div v-if="tokenReasonOf(item.msg) > 0" class="ai-token-popover-row">
|
||||
<span class="ai-token-popover-label">思考(reasoning)</span>
|
||||
<span class="ai-token-popover-val">{{ formatTokens(tokenReasonOf(item.msg)) }} ({{ tokenReasonOf(item.msg) }})</span>
|
||||
</div>
|
||||
<div v-if="cacheHitRate(item.msg) != null" class="ai-token-popover-row">
|
||||
<span class="ai-token-popover-label">缓存命中率</span>
|
||||
<span class="ai-token-popover-val">{{ cacheHitRate(item.msg) }}%</span>
|
||||
</div>
|
||||
<div v-if="item.msg.model" class="ai-token-popover-row">
|
||||
<span class="ai-token-popover-label">模型</span>
|
||||
<span class="ai-token-popover-val">{{ item.msg.model }}</span>
|
||||
</div>
|
||||
</div>
|
||||
</div>
|
||||
|
||||
<!-- 工具调用卡片(渲染/折叠/审批全下沉到 ToolCardList+ToolCard 子组件,MessageList 仅转发审批) -->
|
||||
@@ -1128,6 +1227,7 @@ defineExpose({
|
||||
|
||||
/* ── token 用量条(克制:右对齐 / 最小字号 / dim / 分隔线,不抢正文焦点) ── */
|
||||
.ai-token-usage {
|
||||
position: relative;
|
||||
display: flex;
|
||||
justify-content: flex-end;
|
||||
align-items: center;
|
||||
@@ -1141,6 +1241,64 @@ defineExpose({
|
||||
opacity: 0.7;
|
||||
}
|
||||
.ai-token-usage-icon { font-size: 9px; }
|
||||
/* token 分项显示(2026-08-02):trigger 是可点击按钮(无背景,继承 dim 样式) */
|
||||
.ai-token-usage-trigger {
|
||||
display: inline-flex;
|
||||
align-items: center;
|
||||
gap: 3px;
|
||||
padding: 0;
|
||||
border: none;
|
||||
background: transparent;
|
||||
font: inherit;
|
||||
color: inherit;
|
||||
cursor: pointer;
|
||||
opacity: 1;
|
||||
}
|
||||
.ai-token-usage-trigger:hover {
|
||||
color: var(--df-text);
|
||||
opacity: 1;
|
||||
}
|
||||
.ai-token-sep {
|
||||
opacity: 0.5;
|
||||
}
|
||||
/* 详情面板:复用 TopBar popout 风格(absolute / right 0 / bg-card / border / shadow) */
|
||||
.ai-token-popover {
|
||||
position: absolute;
|
||||
top: 100%;
|
||||
right: 0;
|
||||
margin-top: 4px;
|
||||
padding: 8px 10px;
|
||||
min-width: 200px;
|
||||
background: var(--df-bg-card, var(--df-bg));
|
||||
border: 0.5px solid var(--df-border);
|
||||
border-radius: 4px;
|
||||
box-shadow: 0 4px 12px rgba(0, 0, 0, 0.15);
|
||||
font-family: var(--df-font-mono);
|
||||
font-size: 10px;
|
||||
color: var(--df-text);
|
||||
z-index: 10;
|
||||
text-align: left;
|
||||
}
|
||||
.ai-token-popover-title {
|
||||
font-weight: 600;
|
||||
margin-bottom: 6px;
|
||||
padding-bottom: 4px;
|
||||
border-bottom: 0.5px solid var(--df-border);
|
||||
}
|
||||
.ai-token-popover-row {
|
||||
display: flex;
|
||||
justify-content: space-between;
|
||||
align-items: center;
|
||||
gap: 12px;
|
||||
padding: 2px 0;
|
||||
}
|
||||
.ai-token-popover-label {
|
||||
color: var(--df-text-dim);
|
||||
}
|
||||
.ai-token-popover-val {
|
||||
color: var(--df-text);
|
||||
font-weight: 500;
|
||||
}
|
||||
|
||||
/* 用户气泡内的图片(多模态消息渲染) */
|
||||
.ai-msg-images {
|
||||
|
||||
Reference in New Issue
Block a user