大模型速度增加提示

This commit is contained in:
吉鑫升 committed 2026-05-14 19:12:14 +08:00
1 parent 4e764350a4
commit 3ea33e1dee
52 files changed
+4673 -176

No files matched your search

@@ -1,28 +1,98 @@
import type { Message } from "@langchain/langgraph-sdk";
import { CoinsIcon } from "lucide-react";
import { CoinsIcon, GaugeIcon } from "lucide-react";
import { useI18n } from "@/core/i18n/hooks";
import { formatTokenCount, getUsageMetadata } from "@/core/messages/usage";
import { useLocalSettings } from "@/core/settings";
import { cn } from "@/lib/utils";
import {
MessageMetricsContext,
useMessageMetric,
useMessageMetrics,
type MessageMetric,
} from "./use-message-metrics";
const FULL_SCALE_TOKENS_PER_SEC = 80;
function classifySpeed(tps: number): { color: string; label: string } {
if (tps >= 50) return { color: "#16a34a", label: "极快" };
if (tps >= 30) return { color: "#65a30d", label: "流畅" };
if (tps >= 15) return { color: "#eab308", label: "一般" };
if (tps >= 5) return { color: "#f97316", label: "偏慢" };
return { color: "#ef4444", label: "缓慢" };
}
function formatDuration(ms: number): string {
if (ms < 1000) return `${Math.round(ms)} ms`;
if (ms < 60_000) return `${(ms / 1000).toFixed(1)} s`;
const mins = Math.floor(ms / 60_000);
const secs = Math.round((ms % 60_000) / 1000);
return `${mins}m ${secs}s`;
}
function SpeedPill({ metric }: { metric: MessageMetric }) {
const fill = Math.max(0, Math.min(1, metric.tokensPerSec / FULL_SCALE_TOKENS_PER_SEC));
const { color, label } = classifySpeed(metric.tokensPerSec);
const display =
metric.tokensPerSec >= 100
? Math.round(metric.tokensPerSec).toString()
: metric.tokensPerSec.toFixed(1);
const prefix = metric.fromUsage ? "" : "~";
return (
<span
className="inline-flex items-center gap-1.5"
title={`${label} · ${metric.fromUsage ? "按 output_tokens 计算" : "字符估算 ≈ 2 字符 / token"}`}
>
<GaugeIcon className="size-3" style={{ color }} />
<span className="font-mono tabular-nums" style={{ color }}>
{prefix}
{display}
</span>
<span className="text-muted-foreground">tok/s</span>
<span className="relative inline-block h-1 w-12 overflow-hidden rounded-full bg-current/15 align-middle">
<span
className="absolute inset-y-0 left-0 rounded-full transition-[width,background-color] duration-200 ease-out"
style={{ width: `${fill * 100}%`, backgroundColor: color }}
/>
</span>
</span>
);
}
export function MessageTokenUsage({
className,
enabled = false,
isLoading = false,
message,
metric,
}: {
className?: string;
enabled?: boolean;
isLoading?: boolean;
message: Message;
metric?: MessageMetric;
}) {
const { t } = useI18n();
const [settings] = useLocalSettings();
const speedEnabled = settings.display.streamSpeedMeterEnabled;
const contextMetric = useMessageMetric(message.id as string | undefined);
const resolvedMetric = metric ?? contextMetric;
if (!enabled || isLoading || message.type !== "ai") {
if (message.type !== "ai") {
return null;
}
// Show speed-only line while streaming (no usage yet) if the user opted in,
// otherwise wait until usage arrives so the bar doesn't render on its own.
const usage = getUsageMetadata(message);
const showUsage = enabled && !isLoading && Boolean(usage);
const showSpeed = speedEnabled && resolvedMetric !== undefined && resolvedMetric.durationMs > 0;
if (!showUsage && !showSpeed) {
return null;
}
return (
<div
@@ -31,12 +101,12 @@ export function MessageTokenUsage({
className,
)}
>
<span className="inline-flex items-center gap-1 font-medium">
<CoinsIcon className="size-3" />
{t.tokenUsage.label}
</span>
{usage ? (
{showUsage && usage ? (
<>
<span className="inline-flex items-center gap-1 font-medium">
<CoinsIcon className="size-3" />
{t.tokenUsage.label}
</span>
<span>
{t.tokenUsage.input}: {formatTokenCount(usage.inputTokens)}
</span>
@@ -47,9 +117,17 @@ export function MessageTokenUsage({
{t.tokenUsage.total}: {formatTokenCount(usage.totalTokens)}
</span>
</>
) : (
<span>{t.tokenUsage.unavailableShort}</span>
)}
) : null}
{showSpeed && resolvedMetric ? (
<>
<SpeedPill metric={resolvedMetric} />
{resolvedMetric.durationMs > 0 && (
<span className="font-mono tabular-nums">
{formatDuration(resolvedMetric.durationMs)}
</span>
)}
</>
) : null}
</div>
);
}
@@ -65,12 +143,14 @@ export function MessageTokenUsageList({
isLoading?: boolean;
messages: Message[];
}) {
if (!enabled || isLoading) {
const [settings] = useLocalSettings();
const speedEnabled = settings.display.streamSpeedMeterEnabled;
if (!enabled && !speedEnabled) {
return null;
}
const aiMessages = messages.filter((message) => message.type === "ai");
if (aiMessages.length === 0) {
return null;
}
@@ -89,3 +169,25 @@ export function MessageTokenUsageList({
</>
);
}
/**
* Hoist the metrics map computation to the top of the message list so
* descendant {@link MessageTokenUsage} components can read it via context
* instead of each running its own hook.
*/
export function MessageMetricsProvider({
messages,
isLoading,
children,
}: {
messages: Message[];
isLoading: boolean;
children: React.ReactNode;
}) {
const metrics = useMessageMetrics(messages, isLoading);
return (
<MessageMetricsContext.Provider value={metrics}>
{children}
</MessageMetricsContext.Provider>
);
}