Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

Studio: realtime option for speed and context stats, shorter Chat defaults (#16888)

Michael Hanc2ca6b409ed5
diff --git a/studio/backend/core/inference/diffusion_memory.py b/studio/backend/core/inference/diffusion_memory.py
index e4b863353b4..43f19af897a 100644
--- a/studio/backend/core/inference/diffusion_memory.py
+++ b/studio/backend/core/inference/diffusion_memory.py
@@ -2528,7 +2528,7 @@ UNIFIED_OVERSIZE_ENV = "UNSLOTH_DIFFUSION_ALLOW_OVERSIZED_LOAD"
UNIFIED_OVERSIZE_REQUEST_FIELD = "allow_oversized_load"
# Must match the Settings switch label and location in frontend unified-memory-load-refusal.ts.
UNIFIED_OVERSIZE_SETTING_LABEL = "Allow oversized model loads"
-UNIFIED_OVERSIZE_SETTING_LOCATION = "Settings > System > Model memory"
+UNIFIED_OVERSIZE_SETTING_LOCATION = "Settings > System > Model loading"
UNIFIED_OVERSIZE_REFUSAL_CODE = "unified-memory-load"
# Unlike the env var, the per-load flag does NOT lift the MPS allocator cap: that cap is process-wide and set once,
diff --git a/studio/backend/tests/test_diffusion_unified_load_override.py b/studio/backend/tests/test_diffusion_unified_load_override.py
index 77f268dc952..45dd91322b7 100644
--- a/studio/backend/tests/test_diffusion_unified_load_override.py
+++ b/studio/backend/tests/test_diffusion_unified_load_override.py
@@ -202,7 +202,7 @@ def test_default_still_refuses_with_a_coded_runtime_error():
assert excinfo.value.code == UNIFIED_OVERSIZE_REFUSAL_CODE == "unified-memory-load"
message = str(excinfo.value)
assert "Load anyway" in message
- assert "'Allow oversized model loads' in Settings > System > Model memory" in message
+ assert "'Allow oversized model loads' in Settings > System > Model loading" in message
assert "allow_oversized_load" in message
assert f"{UNIFIED_OVERSIZE_ENV}=1" in message
diff --git a/studio/frontend/smoke-composer-settings-main.tsx b/studio/frontend/smoke-composer-settings-main.tsx
index f71632dac04..dcc360525c6 100644
--- a/studio/frontend/smoke-composer-settings-main.tsx
+++ b/studio/frontend/smoke-composer-settings-main.tsx
@@ -10,7 +10,11 @@ import {
useAui,
useAuiState,
} from "@assistant-ui/react";
-import { ComposerSettings } from "@/features/settings/components/composer-settings";
+import {
+ ComposerFormatRows,
+ ComposerSettings,
+ ContextWindowUsageRow,
+} from "@/features/settings/components/composer-settings";
import { ComposerDraftPreview } from "@/components/assistant-ui/composer-draft-preview";
import { TooltipProvider } from "@/components/ui/tooltip";
import { useChatPreferencesStore } from "@/features/chat/stores/chat-preferences-store";
@@ -79,6 +83,8 @@ function App() {
<TooltipProvider>
<main className="mx-auto max-w-4xl px-5 py-12">
<ComposerSettings />
+ <ComposerFormatRows />
+ <ContextWindowUsageRow />
<Draft />
</main>
</TooltipProvider>
diff --git a/studio/frontend/src/components/assistant-ui/live-generation-line.tsx b/studio/frontend/src/components/assistant-ui/live-generation-line.tsx
index 47debf5bb39..0f3453e67b5 100644
--- a/studio/frontend/src/components/assistant-ui/live-generation-line.tsx
+++ b/studio/frontend/src/components/assistant-ui/live-generation-line.tsx
@@ -3,34 +3,51 @@
import {
formatLiveDecode,
+ type LiveDecodeProgress,
useChatPreferencesStore,
useChatRuntimeStore,
+ useThrottledLiveGeneration,
} from "@/features/chat";
import { useLocale, useT } from "@/i18n";
import { useAuiState } from "@assistant-ui/react";
import type { FC } from "react";
-/** Gated on the preference first so a reply mounts no assistant-ui selector while it is off
- * (composer-keystroke-subscription-budget), like AssistantMessageTime. */
+/** Gated on the preferences first so a reply mounts no assistant-ui selector while both are off
+ * (composer-keystroke-subscription-budget). */
export const LiveGenerationLine: FC = () => {
- const enabled = useChatPreferencesStore((state) => state.showGenerationDetails);
- return enabled ? <RunningGate /> : null;
+ const details = useChatPreferencesStore((state) => state.showGenerationDetails);
+ const realtime = useChatPreferencesStore(
+ (state) => state.generationStatsUpdates === "realtime",
+ );
+ return details || realtime ? <RunningGate realtime={realtime} /> : null;
};
-const RunningGate: FC = () => {
+const RunningGate: FC<{ realtime: boolean }> = ({ realtime }) => {
const running = useAuiState(({ message }) => message.status?.type === "running");
- return running ? <LiveDecodeFigures /> : null;
+ if (!running) return null;
+ return realtime ? <ThrottledDecodeFigures /> : <LiveDecodeFigures />;
};
+// Keyed like the prefill indicator: a first turn reads "__default" until the thread persists.
+const useLiveThreadKey = () =>
+ useAuiState(({ threadListItem }) => threadListItem.remoteId) ?? "__default";
+
const LiveDecodeFigures: FC = () => {
- const locale = useLocale();
- const t = useT();
- // Keyed like the prefill indicator: a first turn reads "__default" until the thread persists.
- const threadKey =
- useAuiState(({ threadListItem }) => threadListItem.remoteId) ?? "__default";
+ const threadKey = useLiveThreadKey();
const decode = useChatRuntimeStore(
(s) => s.liveGenerationByThreadId[threadKey]?.decode ?? null,
);
+ return <DecodeFigures decode={decode} />;
+};
+
+const ThrottledDecodeFigures: FC = () => {
+ const decode = useThrottledLiveGeneration(useLiveThreadKey())?.decode ?? null;
+ return <DecodeFigures decode={decode} />;
+};
+
+const DecodeFigures: FC<{ decode: LiveDecodeProgress | null }> = ({ decode }) => {
+ const locale = useLocale();
+ const t = useT();
if (!decode) {
return null;
}
diff --git a/studio/frontend/src/features/chat/api/chat-adapter.ts b/studio/frontend/src/features/chat/api/chat-adapter.ts
index c18c09e8c7b..9e5dced5e2c 100644
--- a/studio/frontend/src/features/chat/api/chat-adapter.ts
+++ b/studio/frontend/src/features/chat/api/chat-adapter.ts
@@ -6441,6 +6441,8 @@ export function createOpenAIStreamAdapter(
if (usageThreadKey !== null) {
store.setThreadContextUsage(usageThreadKey, usage);
}
+ // The realtime meter shows this exact count from here until the next pass starts.
+ store.settleLiveGeneration(usageKey);
if (
store.activeThreadId === (usageThreadKey ?? activeThreadIdAtRunStart) &&
store.params.checkpoint === params.checkpoint
diff --git a/studio/frontend/src/features/chat/chat-page.tsx b/studio/frontend/src/features/chat/chat-page.tsx
index 5cb84a34a10..f32a7ba7734 100644
--- a/studio/frontend/src/features/chat/chat-page.tsx
+++ b/studio/frontend/src/features/chat/chat-page.tsx
@@ -182,7 +182,7 @@ import {
chatModelSwitchMeta,
type ChatModelSwitchTarget,
} from "./components/chat-model-notice-switch";
-import { ContextUsageBar } from "./components/context-usage-bar";
+import { LiveContextUsageBar } from "./components/live-context-usage-bar";
import { contextUsageRows } from "./lib/context-usage-bar-state";
import { ModelLoadInlineStatus } from "./components/model-load-status";
import { ProjectSwitcher } from "./components/project-switcher";
@@ -4860,7 +4860,9 @@ export function ChatPage({
(view.mode === "single" ||
(view.mode === "project" && activeThreadId != null)) &&
(contextUsage || contextWindowKnown) ? (
- <ContextUsageBar
+ <LiveContextUsageBar
+ // Keyed like the adapter's run (resolvedThreadId || "__default"), local ids included.
+ liveThreadKey={activeThreadId ?? "__default"}
compactUnknownAsIcon={view.mode === "single"}
used={contextUsage?.contextTokens ?? contextUsage?.totalTokens ?? null}
// null on external providers; the bar handles that.
diff --git a/studio/frontend/src/features/chat/components/live-context-usage-bar.tsx b/studio/frontend/src/features/chat/components/live-context-usage-bar.tsx
new file mode 100644
index 00000000000..7b19affbae6
--- /dev/null
+++ b/studio/frontend/src/features/chat/components/live-context-usage-bar.tsx
@@ -0,0 +1,46 @@
+// SPDX-License-Identifier: AGPL-3.0-only
+// Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+"use client";
+
+import type { ComponentProps, FC } from "react";
+import { useThrottledLiveGeneration } from "../hooks/use-throttled-live-generation";
+import { liveContextUsage } from "../lib/live-generation-stats";
+import { useChatPreferencesStore } from "../stores/chat-preferences-store";
+import { ContextUsageBar } from "./context-usage-bar";
+
+type BarProps = ComponentProps<typeof ContextUsageBar>;
+
+/** Settled mode subscribes to no live figures, so the header never re-renders mid-stream. */
+export const LiveContextUsageBar: FC<BarProps & { liveThreadKey: string }> = ({
+ liveThreadKey,
+ ...props
+}) => {
+ const realtime = useChatPreferencesStore(
+ (state) => state.generationStatsUpdates === "realtime",
+ );
+ return realtime ? (
+ <RealtimeContextUsageBar liveThreadKey={liveThreadKey} {...props} />
+ ) : (
+ <ContextUsageBar {...props} />
+ );
+};
+
+const RealtimeContextUsageBar: FC<BarProps & { liveThreadKey: string }> = ({
+ liveThreadKey,
+ ...props
+}) => {
+ const usage = liveContextUsage(useThrottledLiveGeneration(liveThreadKey));
+ if (!usage) return <ContextUsageBar {...props} />;
+ return (
+ <ContextUsageBar
+ {...props}
+ used={usage.used}
+ promptTokens={usage.promptTokens}
+ completionTokens={usage.completionTokens}
+ cached={usage.cached}
+ cacheWrites={undefined}
+ estimated={false}
+ />
+ );
+};
diff --git a/studio/frontend/src/features/chat/hooks/use-throttled-live-generation.ts b/studio/frontend/src/features/chat/hooks/use-throttled-live-generation.ts
new file mode 100644
index 00000000000..ae755f07972
--- /dev/null
+++ b/studio/frontend/src/features/chat/hooks/use-throttled-live-generation.ts
@@ -0,0 +1,46 @@
+// SPDX-License-Identifier: AGPL-3.0-only
+// Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+import { useEffect, useState } from "react";
+import {
+ LIVE_STATS_INTERVAL_MS,
+ createTrailingThrottle,
+} from "../lib/live-generation-stats";
+import {
+ type LiveGenerationProgress,
+ useChatRuntimeStore,
+} from "../stores/chat-runtime-store";
+
+/** Subscribes outside React so a burst of frames costs one trailing render per interval. */
+export function useThrottledLiveGeneration(
+ threadKey: string,
+): LiveGenerationProgress | undefined {
+ const read = () =>
+ useChatRuntimeStore.getState().liveGenerationByThreadId[threadKey];
+ // Tagged with its key so the render after a thread switch never shows the last thread's figures.
+ const [shown, setShown] = useState(() => ({ key: threadKey, live: read() }));
+ useEffect(() => {
+ const show = () =>
+ setShown({
+ key: threadKey,
+ live: useChatRuntimeStore.getState().liveGenerationByThreadId[
+ threadKey
+ ],
+ });
+ const throttle = createTrailingThrottle(LIVE_STATS_INTERVAL_MS, show);
+ show();
+ const unsubscribe = useChatRuntimeStore.subscribe((state, previous) => {
+ if (
+ state.liveGenerationByThreadId[threadKey] !==
+ previous.liveGenerationByThreadId[threadKey]
+ ) {
+ throttle.schedule();
+ }
+ });
+ return () => {
+ unsubscribe();
+ throttle.cancel();
+ };
+ }, [threadKey]);
+ return shown.key === threadKey ? shown.live : read();
+}
diff --git a/studio/frontend/src/features/chat/index.ts b/studio/frontend/src/features/chat/index.ts
index 479cd5015e5..1b6fde4ab0b 100644
--- a/studio/frontend/src/features/chat/index.ts
+++ b/studio/frontend/src/features/chat/index.ts
@@ -100,6 +100,7 @@ export {
prefillPercent,
} from "./lib/generation-progress";
export type { LiveDecodeProgress } from "./stores/chat-runtime-store";
+export { useThrottledLiveGeneration } from "./hooks/use-throttled-live-generation";
export {
preferFullToolOutput,
preferSanitizedFullToolOutput,
diff --git a/studio/frontend/src/features/chat/lib/generation-progress.ts b/studio/frontend/src/features/chat/lib/generation-progress.ts
index ae2ba7bc180..73fb70b378d 100644
--- a/studio/frontend/src/features/chat/lib/generation-progress.ts
+++ b/studio/frontend/src/features/chat/lib/generation-progress.ts
@@ -91,8 +91,7 @@ export function readGenerationProgress(
return null;
}
-/** "9,216 / 23,700 tokens · 1,326 tok/s" for the Generation details setting, counted past the
- * cached prefix like the percent. */
+/** Counted past the cached prefix, like the percent. */
export function formatPrefillDetails(
prefill: LivePrefillProgress,
t: Translate,
diff --git a/studio/frontend/src/features/chat/lib/live-generation-stats.ts b/studio/frontend/src/features/chat/lib/live-generation-stats.ts
new file mode 100644
index 00000000000..4d3b8f02794
--- /dev/null
+++ b/studio/frontend/src/features/chat/lib/live-generation-stats.ts
@@ -0,0 +1,95 @@
+// SPDX-License-Identifier: AGPL-3.0-only
+// Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+import type { LiveGenerationProgress } from "../stores/chat-runtime-store";
+
+export type GenerationStatsUpdates = "settled" | "realtime";
+
+export const GENERATION_STATS_UPDATES: readonly GenerationStatsUpdates[] = [
+ "settled",
+ "realtime",
+];
+
+export const DEFAULT_GENERATION_STATS_UPDATES: GenerationStatsUpdates =
+ "settled";
+
+// Under the server's 250 ms decode cadence, so throttling never drops a frame it sends, only bursts.
+export const LIVE_STATS_INTERVAL_MS = 200;
+
+export function normaliseGenerationStatsUpdates(
+ value: unknown,
+): GenerationStatsUpdates {
+ return value === "realtime" ? "realtime" : DEFAULT_GENERATION_STATS_UPDATES;
+}
+
+export type ThrottleClock = {
+ now: () => number;
+ setTimeout: (run: () => void, ms: number) => unknown;
+ clearTimeout: (handle: unknown) => void;
+};
+
+const defaultClock: ThrottleClock = {
+ now: () => Date.now(),
+ setTimeout: (run, ms) => globalThis.setTimeout(run, ms),
+ clearTimeout: (handle) =>
+ globalThis.clearTimeout(handle as ReturnType<typeof setTimeout>),
+};
+
+/** Leading and trailing: the last change always lands. */
+export function createTrailingThrottle(
+ intervalMs: number,
+ run: () => void,
+ clock: ThrottleClock = defaultClock,
+): { schedule: () => void; cancel: () => void } {
+ let lastRunAt = Number.NEGATIVE_INFINITY;
+ let pending: unknown = null;
+ const fire = () => {
+ pending = null;
+ lastRunAt = clock.now();
+ run();
+ };
+ return {
+ schedule: () => {
+ if (pending !== null) return;
+ const wait = lastRunAt + intervalMs - clock.now();
+ if (wait <= 0) fire();
+ else pending = clock.setTimeout(fire, wait);
+ },
+ cancel: () => {
+ if (pending !== null) clock.clearTimeout(pending);
+ pending = null;
+ },
+ };
+}
+
+export type LiveContextUsage = {
+ used: number;
+ promptTokens: number;
+ completionTokens: number;
+ cached: number;
+};
+
+/** Null when the stream has not said, so the meter keeps the last settled count. */
+export function liveContextUsage(
+ live: LiveGenerationProgress | undefined,
+): LiveContextUsage | null {
+ const prefill = live?.prefill;
+ if (!prefill || live?.compacting || live?.settled) return null;
+ const cached = prefill.cached ?? 0;
+ const decode = live.decode;
+ if (decode) {
+ if (!prefill.total) return null;
+ return {
+ used: prefill.total + decode.tokens,
+ promptTokens: prefill.total,
+ completionTokens: decode.tokens,
+ cached,
+ };
+ }
+ return {
+ used: prefill.processed,
+ promptTokens: prefill.processed,
+ completionTokens: 0,
+ cached,
+ };
+}
diff --git a/studio/frontend/src/features/chat/stores/chat-preferences-store.ts b/studio/frontend/src/features/chat/stores/chat-preferences-store.ts
index dd33ffc0dc0..16db6911cb6 100644
--- a/studio/frontend/src/features/chat/stores/chat-preferences-store.ts
+++ b/studio/frontend/src/features/chat/stores/chat-preferences-store.ts
@@ -19,6 +19,11 @@ import {
type DisplayVisibility,
migrateVisibility,
} from "../utils/display-visibility.ts";
+import {
+ DEFAULT_GENERATION_STATS_UPDATES,
+ type GenerationStatsUpdates,
+ normaliseGenerationStatsUpdates,
+} from "../lib/live-generation-stats.ts";
export interface ChatPreferencesState {
plainTextComposer: boolean;
@@ -41,6 +46,8 @@ export interface ChatPreferencesState {
setShowReplyTimes: (value: boolean) => void;
showGenerationDetails: boolean;
setShowGenerationDetails: (value: boolean) => void;
+ generationStatsUpdates: GenerationStatsUpdates;
+ setGenerationStatsUpdates: (value: GenerationStatsUpdates) => void;
showInlineEditResponse: boolean;
setShowInlineEditResponse: (value: boolean) => void;
thinkingVisibility: DisplayVisibility;
@@ -108,6 +115,9 @@ export const useChatPreferencesStore = create<ChatPreferencesState>()(
showGenerationDetails: false,
setShowGenerationDetails: (showGenerationDetails) =>
set({ showGenerationDetails }),
+ generationStatsUpdates: DEFAULT_GENERATION_STATS_UPDATES,
+ setGenerationStatsUpdates: (value) =>
+ set({ generationStatsUpdates: normaliseGenerationStatsUpdates(value) }),
showInlineEditResponse: true,
setShowInlineEditResponse: (showInlineEditResponse) =>
set({ showInlineEditResponse }),
@@ -165,6 +175,9 @@ export const useChatPreferencesStore = create<ChatPreferencesState>()(
showLinkPreviews: saved?.showLinkPreviews !== false,
showReplyTimes: saved?.showReplyTimes === true,
showGenerationDetails: saved?.showGenerationDetails === true,
+ generationStatsUpdates: normaliseGenerationStatsUpdates(
+ saved?.generationStatsUpdates,
+ ),
showInlineEditResponse: saved?.showInlineEditResponse ?? true,
thinkingVisibility: migrateVisibility(
saved?.thinkingVisibility,
diff --git a/studio/frontend/src/features/chat/stores/chat-runtime-store.ts b/studio/frontend/src/features/chat/stores/chat-runtime-store.ts
index 5ddb76e0490..5ff64a9e9a9 100644
--- a/studio/frontend/src/features/chat/stores/chat-runtime-store.ts
+++ b/studio/frontend/src/features/chat/stores/chat-runtime-store.ts
@@ -452,6 +452,8 @@ export type LiveGenerationProgress = {
decode: LiveDecodeProgress | null;
/** The context fit is evicting turns; ends with this pass's context_truncated or its first prefill sample. */
compacting?: boolean;
+ /** This pass's own context_usage has landed: it counts the reply exactly, so live figures step aside. */
+ settled?: boolean;
};
export type PendingImageEditReference = {
threadId: string | null;
@@ -2743,6 +2745,7 @@ type ChatRuntimeStore = {
setLiveDecode: (threadId: string | null, decode: LiveDecodeProgress) => void;
setLiveCompacting: (threadId: string | null, compacting: boolean) => void;
clearLiveGenerationForThread: (threadId: string | null) => void;
+ settleLiveGeneration: (threadId: string | null) => void;
setAutoHealToolCalls: (enabled: boolean) => void;
setNudgeToolCalls: (enabled: boolean) => void;
setDeduplicateToolCalls: (enabled: boolean) => void;
@@ -6004,6 +6007,7 @@ export const useChatRuntimeStore = create<ChatRuntimeStore>((set, get) => ({
prefill: current?.prefill ?? null,
prefillShown: current?.prefillShown ?? false,
decode,
+ ...(current?.settled ? { settled: true } : {}),
},
},
};
@@ -6021,6 +6025,15 @@ export const useChatRuntimeStore = create<ChatRuntimeStore>((set, get) => ({
liveGenerationByThreadId: { ...state.liveGenerationByThreadId, [key]: next },
};
}),
+ settleLiveGeneration: (threadId) =>
+ set((state) => {
+ const key = threadId || "__default";
+ const current = state.liveGenerationByThreadId[key];
+ if (current === undefined || current.settled) return state;
+ return {
+ liveGenerationByThreadId: { ...state.liveGenerationByThreadId, [key]: { ...current, settled: true } },
+ };
+ }),
clearLiveGenerationForThread: (threadId) =>
set((state) => {
const key = threadId || "__default";
diff --git a/studio/frontend/src/features/chat/stores/plus-menu-prefs-store.ts b/studio/frontend/src/features/chat/stores/plus-menu-prefs-store.ts
index e48826528e7..dc5dcf11b7a 100644
--- a/studio/frontend/src/features/chat/stores/plus-menu-prefs-store.ts
+++ b/studio/frontend/src/features/chat/stores/plus-menu-prefs-store.ts
@@ -43,6 +43,17 @@ const DEFAULT_PINS: Record<PlusMenuItemId, boolean> = {
exportChat: false,
};
+export function plusMenuAtDefaults(
+ pins: Record<PlusMenuItemId, boolean>,
+ order: readonly PlusMenuItemId[],
+): boolean {
+ return (
+ order.length === PLUS_MENU_ORDER.length &&
+ order.every((id, index) => id === PLUS_MENU_ORDER[index]) &&
+ PLUS_MENU_ORDER.every((id) => pins[id] === DEFAULT_PINS[id])
+ );
+}
+
export const PLUS_MENU_PINS_STORAGE_KEY = "unsloth_plus_menu_pins";
/** A saved order with unknown ids dropped and missing ones (added in a later release) placed
diff --git a/studio/frontend/src/features/model-picker/components/chat-template-editor-dialog.tsx b/studio/frontend/src/features/model-picker/components/chat-template-editor-dialog.tsx
index 3b6286ea8e3..79ac4e4827f 100644
--- a/studio/frontend/src/features/model-picker/components/chat-template-editor-dialog.tsx
+++ b/studio/frontend/src/features/model-picker/components/chat-template-editor-dialog.tsx
@@ -156,12 +156,11 @@ export function ChatTemplateEditorDialog({
<>
<Button
type="button"
- variant="ghost"
+ variant="outline"
onClick={() => setDraft(defaultTemplate ?? "")}
disabled={
defaultLoading || renderedDraft === (defaultTemplate ?? "")
}
- className="text-muted-foreground"
>
{defaultLoading ? (
<Spinner className="size-3.5" />
diff --git a/studio/frontend/src/features/settings/components/composer-settings.tsx b/studio/frontend/src/features/settings/components/composer-settings.tsx
index ceadb2c8afa..d4b604525ef 100644
--- a/studio/frontend/src/features/settings/components/composer-settings.tsx
+++ b/studio/frontend/src/features/settings/components/composer-settings.tsx
@@ -134,38 +134,55 @@ export function ComposerSettings({ embedded = false }: { embedded?: boolean }) {
))}
</div>
</SettingsRow>
- <SettingsRow
- label={t("composerSettings.plainText")}
- description={t("composerSettings.plainTextDescription")}
- >
- <Switch
- aria-label={t("composerSettings.plainText")}
- checked={prefs.plainTextComposer}
- onCheckedChange={prefs.setPlainTextComposer}
- />
- </SettingsRow>
- <SettingsRow
- label={t("composerSettings.richPaste")}
- badge={<Badge variant="secondary" className="h-4 px-1.5 text-ui-10 text-muted-foreground">{t("shell.beta")}</Badge>}
- description={t("composerSettings.richPasteDescription", {
- shortcut: isMacPlatform() ? "\u21e7\u2318V" : "Ctrl+Shift+V",
- })}
- >
- <Switch
- aria-label={t("composerSettings.richPaste")}
- checked={prefs.richTextPaste}
- onCheckedChange={prefs.setRichTextPaste}
- />
- </SettingsRow>
- <SettingsRow label={t("composerSettings.showContext")}>
- <Switch
- aria-label={t("composerSettings.showContext")}
- checked={prefs.showContextWindowUsage}
- onCheckedChange={prefs.setShowContextWindowUsage}
- />
- </SettingsRow>
</div>
</SettingsSection>
</div>
);
}
+
+export function ComposerFormatRows() {
+ const t = useT();
+ const prefs = useChatPreferencesStore();
+ return (
+ <>
+ <SettingsRow
+ label={t("composerSettings.plainText")}
+ description={t("composerSettings.plainTextDescription")}
+ >
+ <Switch
+ aria-label={t("composerSettings.plainText")}
+ checked={prefs.plainTextComposer}
+ onCheckedChange={prefs.setPlainTextComposer}
+ />
+ </SettingsRow>
+ <SettingsRow
+ label={t("composerSettings.richPaste")}
+ badge={<Badge variant="secondary" className="h-4 px-1.5 text-ui-10 text-muted-foreground">{t("shell.beta")}</Badge>}
+ description={t("composerSettings.richPasteDescription", {
+ shortcut: isMacPlatform() ? "\u21e7\u2318V" : "Ctrl+Shift+V",
+ })}
+ >
+ <Switch
+ aria-label={t("composerSettings.richPaste")}
+ checked={prefs.richTextPaste}
+ onCheckedChange={prefs.setRichTextPaste}
+ />
+ </SettingsRow>
+ </>
+ );
+}
+
+export function ContextWindowUsageRow() {
+ const t = useT();
+ const checked = useChatPreferencesStore((s) => s.showContextWindowUsage);
+ const setChecked = useChatPreferencesStore((s) => s.setShowContextWindowUsage);
+ return (
+ <SettingsRow label={t("composerSettings.showContext")}>
+ <Switch
+ aria-label={t("composerSettings.showContext")}
+ checked={checked}
+ onCheckedChange={setChecked}
+ />
+ </SettingsRow>
+ );
+}
diff --git a/studio/frontend/src/features/settings/components/generation-stats-updates-row.tsx b/studio/frontend/src/features/settings/components/generation-stats-updates-row.tsx
new file mode 100644
index 00000000000..6810cdea6be
--- /dev/null
+++ b/studio/frontend/src/features/settings/components/generation-stats-updates-row.tsx
@@ -0,0 +1,59 @@
+// SPDX-License-Identifier: AGPL-3.0-only
+// Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+import { SegmentedPill } from "@/components/segmented-control";
+import { useChatPreferencesStore } from "@/features/chat";
+import { GENERATION_STATS_UPDATES } from "@/features/chat/lib/live-generation-stats";
+import { useT } from "@/i18n";
+import { cn } from "@/lib/utils";
+import { useId } from "react";
+import { SettingsRow } from "./settings-row";
+
+export function GenerationStatsUpdatesRow() {
+ const t = useT();
+ const pillId = useId();
+ const generationStatsUpdates = useChatPreferencesStore(
+ (state) => state.generationStatsUpdates,
+ );
+ const setGenerationStatsUpdates = useChatPreferencesStore(
+ (state) => state.setGenerationStatsUpdates,
+ );
+ return (
+ <SettingsRow
+ label={t("settings.chat.generationStatsUpdates")}
+ description={t("settings.chat.generationStatsUpdatesDescription")}
+ hint={t("settings.chat.generationStatsUpdatesHint")}
+ >
+ <div
+ className="hub-tab-toggle inline-flex h-8 items-center rounded-full"
+ // biome-ignore lint/a11y/useSemanticElements: matches Scroll while generating; a fieldset's default box would break the pill row.
+ role="group"
+ aria-label={t("settings.chat.generationStatsUpdates")}
+ >
+ {GENERATION_STATS_UPDATES.map((mode) => (
+ <button
+ key={mode}
+ type="button"
+ aria-pressed={generationStatsUpdates === mode}
+ onClick={() => setGenerationStatsUpdates(mode)}
+ className={cn(
+ "relative inline-flex h-8 cursor-pointer items-center rounded-full px-3.5 text-ui-12 font-medium transition-colors focus-visible:outline-none focus-visible:ring-1 focus-visible:ring-ring",
+ generationStatsUpdates === mode
+ ? "text-foreground"
+ : "text-muted-foreground hover:text-foreground",
+ )}
+ >
+ {generationStatsUpdates === mode && (
+ <SegmentedPill group={`${pillId}-generation-stats`} />
+ )}
+ {t(
+ mode === "realtime"
+ ? "settings.chat.generationStatsUpdatesRealtime"
+ : "settings.chat.generationStatsUpdatesSettled",
+ )}
+ </button>
+ ))}
+ </div>
+ </SettingsRow>
+ );
+}
diff --git a/studio/frontend/src/features/settings/settings-search.ts b/studio/frontend/src/features/settings/settings-search.ts
index 7f318c57154..1afc4bcfa61 100644
--- a/studio/frontend/src/features/settings/settings-search.ts
+++ b/studio/frontend/src/features/settings/settings-search.ts
@@ -148,6 +148,8 @@ export const SETTINGS_SEARCH_INDEX: Record<SettingsTab, TranslationKey[]> = {
"settings.chat.pastedTextThreshold",
"settings.chat.turnNavigation",
"settings.chat.groups.conversations.title",
+ "settings.chat.groups.replies.title",
+ "settings.chat.groups.thinkingTools.title",
"settings.chat.library.label",
"settings.chat.groups.display.title",
"settings.chat.modelSelection.title",
@@ -172,6 +174,7 @@ export const SETTINGS_SEARCH_INDEX: Record<SettingsTab, TranslationKey[]> = {
"settings.chat.inlineEditResponse",
"settings.chat.showReplyTimes",
"settings.chat.showGenerationDetails",
+ "settings.chat.generationStatsUpdates",
"settings.chat.groups.menu.title",
],
browser: [
@@ -511,6 +514,7 @@ export const SETTINGS_SEARCH_KEYWORDS: Partial<
"settings.chat.showLinkPreviews": "settings.chat.showLinkPreviewsDescription",
"settings.chat.showReplyTimes": "settings.chat.showReplyTimesKeywords",
"settings.chat.showGenerationDetails": "settings.chat.showGenerationDetailsKeywords",
+ "settings.chat.generationStatsUpdates": "settings.chat.generationStatsUpdatesKeywords",
// match protocol and agent names omitted from the title.
"settings.apiKeys.mcp.title": "settings.apiKeys.mcp.keywords",
// the intro names only some agents; every listed one has to find this page.
diff --git a/studio/frontend/src/features/settings/tabs/browser-tab.tsx b/studio/frontend/src/features/settings/tabs/browser-tab.tsx
index 8b041df020a..b6187490bb9 100644
--- a/studio/frontend/src/features/settings/tabs/browser-tab.tsx
+++ b/studio/frontend/src/features/settings/tabs/browser-tab.tsx
@@ -84,7 +84,7 @@ function DownloadLocationRow() {
>
<div className="flex gap-2">
{folder?.custom ? (
- <Button variant="ghost" size="sm" onClick={reset}>
+ <Button variant="outline" size="sm" onClick={reset}>
{t("browser.downloadLocationReset")}
</Button>
) : null}
diff --git a/studio/frontend/src/features/settings/tabs/chat-tab.tsx b/studio/frontend/src/features/settings/tabs/chat-tab.tsx
index 1d521190f3d..47b26b90ecd 100644
--- a/studio/frontend/src/features/settings/tabs/chat-tab.tsx
+++ b/studio/frontend/src/features/settings/tabs/chat-tab.tsx
@@ -23,6 +23,7 @@ import {
usePlusMenuPrefsStore,
} from "@/features/chat";
import { PASTED_TEXT_THRESHOLD_CHOICES } from "@/features/chat/utils/pasted-text";
+import { plusMenuAtDefaults } from "@/features/chat/stores/plus-menu-prefs-store";
import { refreshContextUsage } from "@/features/chat/utils/refresh-context-usage";
import { formatBindingLabel, isMacPlatform } from "../lib/keyboard-shortcuts";
import { useIsAccountOwner } from "@/features/auth";
@@ -57,8 +58,13 @@ import {
} from "../api/tool-result-limit";
import { ChatMenuOrderList } from "../components/chat-menu-order-list";
import { FirstTokenTimeoutRow } from "../components/first-token-timeout-row";
+import { GenerationStatsUpdatesRow } from "../components/generation-stats-updates-row";
import { SettingsRow } from "../components/settings-row";
-import { ComposerSettings } from "../components/composer-settings";
+import {
+ ComposerFormatRows,
+ ComposerSettings,
+ ContextWindowUsageRow,
+} from "../components/composer-settings";
import { SettingsSection } from "../components/settings-section";
import { useSettingsDialogStore } from "../stores/settings-dialog-store";
@@ -289,6 +295,9 @@ export function ChatTab() {
const pillId = useId();
const navigate = useNavigate();
const resetPlusMenu = usePlusMenuPrefsStore((state) => state.resetMenu);
+ const plusMenuPristine = usePlusMenuPrefsStore((state) =>
+ plusMenuAtDefaults(state.pins, state.order),
+ );
const autoTitle = useChatRuntimeStore((state) => state.autoTitle);
const setAutoTitle = useChatRuntimeStore((state) => state.setAutoTitle);
const projectAttachmentTarget = useChatRuntimeStore(
@@ -502,64 +511,6 @@ export function ChatTab() {
<SettingsSection title={t("settings.general.chatDefaults")}>
<ComposerSettings embedded={true} />
- <SettingsRow
- label={t("settings.chat.pastedTextThreshold")}
- description={
- pastedTextMinChars > 0
- ? t("settings.chat.pastedTextShortDescription", {
- count: pastedTextMinChars.toLocaleString(),
- })
- : t("settings.chat.pastedTextOffDescription")
- }
- hint={t("settings.chat.pastedTextThresholdDescription", {
- shortcut: plainPasteLabel,
- })}
- >
- <Select
- value={String(pastedTextMinChars)}
- onValueChange={(value) => setPastedTextMinChars(Number(value))}
- >
- <SelectTrigger
- className="w-36"
- aria-label={t("settings.chat.pastedTextThreshold")}
- >
- <SelectValue />
- </SelectTrigger>
- <SelectContent>
- {PASTED_TEXT_THRESHOLD_CHOICES.map((choice) => (
- <SelectItem key={choice} value={String(choice)}>
- {choice === 0
- ? t("settings.chat.pastedTextThresholdOff")
- : choice.toLocaleString()}
- </SelectItem>
- ))}
- </SelectContent>
- </Select>
- </SettingsRow>
- <SettingsRow
- label={t("settings.chat.autoCompact")}
- description={t("settings.chat.autoCompactDescription")}
- hint={t("settings.chat.autoCompactHint")}
- >
- <Switch
- aria-label={t("settings.chat.autoCompact")}
- checked={autoCompactEnabled}
- onCheckedChange={setAutoCompactEnabled}
- />
- </SettingsRow>
- <SettingsRow
- label={t("settings.chat.keepRunningWhenTabClosed")}
- description={t("settings.chat.keepRunningWhenTabClosedDescription")}
- hint={t("settings.chat.keepRunningWhenTabClosedHint")}
- >
- <Switch
- aria-label={t("settings.chat.keepRunningWhenTabClosed")}
- checked={keepRunningWhenTabClosed}
- onCheckedChange={setKeepRunningWhenTabClosed}
- />
- </SettingsRow>
- {isOwner ? <ToolResultLimitRow /> : null}
- {isOwner ? <FirstTokenTimeoutRow /> : null}
<SettingsRow
label={t("settings.chat.autoScroll")}
description={t("settings.chat.autoScrollDescription")}
@@ -595,25 +546,48 @@ export function ChatTab() {
</div>
</SettingsRow>
<SettingsRow
- label={t("settings.chat.replyNotification")}
- description={t("settings.chat.replyNotificationDescription")}
+ label={t("settings.chat.pastedTextThreshold")}
+ description={
+ pastedTextMinChars > 0
+ ? t("settings.chat.pastedTextShortDescription", {
+ count: pastedTextMinChars.toLocaleString(),
+ })
+ : t("settings.chat.pastedTextOffDescription")
+ }
+ hint={t("settings.chat.pastedTextThresholdDescription", {
+ shortcut: plainPasteLabel,
+ })}
>
- <Switch
- aria-label={t("settings.chat.replyNotification")}
- checked={notifyWhenReplyFinishes}
- onCheckedChange={(checked) => {
- setNotifyWhenReplyFinishes(checked);
- if (!checked) return;
- void requestReplyFinishedPermission().then((granted) => {
- if (granted) return;
- setNotifyWhenReplyFinishes(false);
- toast.error(t("settings.chat.replyNotificationBlocked"));
- });
- }}
- />
+ <Select
+ value={String(pastedTextMinChars)}
+ onValueChange={(value) => setPastedTextMinChars(Number(value))}
+ >
+ <SelectTrigger
+ className="w-36"
+ aria-label={t("settings.chat.pastedTextThreshold")}
+ >
+ <SelectValue />
+ </SelectTrigger>
+ <SelectContent>
+ {PASTED_TEXT_THRESHOLD_CHOICES.map((choice) => (
+ <SelectItem key={choice} value={String(choice)}>
+ {choice === 0
+ ? t("settings.chat.pastedTextThresholdOff")
+ : choice.toLocaleString()}
+ </SelectItem>
+ ))}
+ </SelectContent>
+ </Select>
</SettingsRow>
+ <GenerationStatsUpdatesRow />
</SettingsSection>
+ <LazyImportBoundary fallback={null}>
+ <Suspense fallback={null}>
+ <ChatMemorySection />
+ </Suspense>
+ </LazyImportBoundary>
+
<SettingsSection title={t("settings.chat.groups.conversations.title")}>
<SettingsRow
label={t("settings.chat.rememberParamsPerModel")}
@@ -670,31 +644,54 @@ export function ChatTab() {
onCheckedChange={setAutoTitle}
/>
</SettingsRow>
+ <SettingsRow
+ label={t("settings.chat.autoCompact")}
+ description={t("settings.chat.autoCompactDescription")}
+ hint={t("settings.chat.autoCompactHint")}
+ >
+ <Switch
+ aria-label={t("settings.chat.autoCompact")}
+ checked={autoCompactEnabled}
+ onCheckedChange={setAutoCompactEnabled}
+ />
+ </SettingsRow>
</SettingsSection>
- <LazyImportBoundary fallback={null}>
- <Suspense fallback={null}>
- <ChatMemorySection />
- </Suspense>
- </LazyImportBoundary>
-
- <SettingsSection title={t("settings.chat.groups.display.title")}>
+ <SettingsSection title={t("settings.chat.groups.replies.title")}>
<SettingsRow
- label={t("settings.chat.showLinkPreviews")}
- description={t("settings.chat.showLinkPreviewsDescription")}
+ label={t("settings.chat.keepRunningWhenTabClosed")}
+ description={t("settings.chat.keepRunningWhenTabClosedDescription")}
+ hint={t("settings.chat.keepRunningWhenTabClosedHint")}
>
<Switch
- aria-label={t("settings.chat.showLinkPreviews")}
- checked={showLinkPreviews}
- onCheckedChange={setShowLinkPreviews}
+ aria-label={t("settings.chat.keepRunningWhenTabClosed")}
+ checked={keepRunningWhenTabClosed}
+ onCheckedChange={setKeepRunningWhenTabClosed}
/>
</SettingsRow>
<SettingsRow
- label={t("settings.chat.webSearch.images")}
- description={t("settings.chat.webSearch.imagesDescription")}
+ label={t("settings.chat.replyNotification")}
+ description={t("settings.chat.replyNotificationDescription")}
>
- <Switch checked={searchImages} onCheckedChange={setSearchImages} />
+ <Switch
+ aria-label={t("settings.chat.replyNotification")}
+ checked={notifyWhenReplyFinishes}
+ onCheckedChange={(checked) => {
+ setNotifyWhenReplyFinishes(checked);
+ if (!checked) return;
+ void requestReplyFinishedPermission().then((granted) => {
+ if (granted) return;
+ setNotifyWhenReplyFinishes(false);
+ toast.error(t("settings.chat.replyNotificationBlocked"));
+ });
+ }}
+ />
</SettingsRow>
+ {isOwner ? <ToolResultLimitRow /> : null}
+ {isOwner ? <FirstTokenTimeoutRow /> : null}
+ </SettingsSection>
+
+ <SettingsSection title={t("settings.chat.groups.thinkingTools.title")}>
<SettingsRow
label={t("settings.chat.thinking.visibility")}
description={t("settings.chat.thinking.visibilityDescription")}
@@ -772,6 +769,25 @@ export function ChatTab() {
onCheckedChange={setFoldToolActivityIntoThinking}
/>
</SettingsRow>
+ <SettingsRow
+ label={t("settings.chat.webSearch.images")}
+ description={t("settings.chat.webSearch.imagesDescription")}
+ >
+ <Switch checked={searchImages} onCheckedChange={setSearchImages} />
+ </SettingsRow>
+ </SettingsSection>
+
+ <SettingsSection title={t("settings.chat.groups.display.title")}>
+ <SettingsRow
+ label={t("settings.chat.showLinkPreviews")}
+ description={t("settings.chat.showLinkPreviewsDescription")}
+ >
+ <Switch
+ aria-label={t("settings.chat.showLinkPreviews")}
+ checked={showLinkPreviews}
+ onCheckedChange={setShowLinkPreviews}
+ />
+ </SettingsRow>
<SettingsRow
label={t("settings.chat.showResponseModel")}
description={t("settings.chat.showResponseModelDescription")}
@@ -802,6 +818,7 @@ export function ChatTab() {
onCheckedChange={setShowGenerationDetails}
/>
</SettingsRow>
+ <ContextWindowUsageRow />
<SettingsRow
label={t("settings.chat.inlineEditResponse")}
description={t("settings.chat.inlineEditResponseDescription")}
@@ -822,6 +839,7 @@ export function ChatTab() {
onCheckedChange={setShowTurnNavigation}
/>
</SettingsRow>
+ <ComposerFormatRows />
</SettingsSection>
<SettingsSection title={t("settings.chat.modelSelection.title")}>
@@ -884,12 +902,20 @@ export function ChatTab() {
) as Record<PlusMenuItemId, { label: string; icon: ReactNode }>}
/>
</div>
- <div className="flex justify-end pt-1">
- <Button variant="ghost" size="sm" onClick={resetPlusMenu}>
- {t("settings.chat.menu.reset")}
- </Button>
- </div>
</SettingsSection>
+
+ {/* Same footer as Appearance: a rule, then the reset on the right. */}
+ <div className="flex justify-end border-t border-border/60 pt-4">
+ <Button
+ type="button"
+ variant="outline"
+ size="sm"
+ disabled={plusMenuPristine}
+ onClick={resetPlusMenu}
+ >
+ {t("settings.chat.menu.reset")}
+ </Button>
+ </div>
</div>
);
}
diff --git a/studio/frontend/src/features/settings/tabs/keyboard-shortcuts-tab.tsx b/studio/frontend/src/features/settings/tabs/keyboard-shortcuts-tab.tsx
index accca1efb74..da3a93c3c8f 100644
--- a/studio/frontend/src/features/settings/tabs/keyboard-shortcuts-tab.tsx
+++ b/studio/frontend/src/features/settings/tabs/keyboard-shortcuts-tab.tsx
@@ -544,13 +544,12 @@ export function KeyboardShortcutsTab() {
);
})}
- {/* Click-through strip; -bottom-6 undoes the scroller padding sticky would inset by. */}
- <div className="pointer-events-none sticky -bottom-6 z-10 -mb-6 mt-auto flex justify-end py-4">
+ {/* Same footer as Appearance: Reset sits after the list, not floating over it. */}
+ <div className="mt-auto flex justify-end border-t border-border/60 pt-4">
<Button
type="button"
- variant="dark"
- // Opaque disabled fill so rows never show through; dark: outranks the variant's own dark fill.
- className="pointer-events-auto text-ui-13 disabled:bg-[color-mix(in_oklab,var(--foreground)_34%,var(--background))] disabled:text-background disabled:opacity-100 dark:disabled:bg-[color-mix(in_oklab,var(--foreground)_42%,var(--background))] dark:disabled:text-background"
+ variant="outline"
+ size="sm"
disabled={Object.keys(overrides).length === 0}
onClick={() => {
stopRecording();
diff --git a/studio/frontend/src/features/settings/tabs/resources-tab.tsx b/studio/frontend/src/features/settings/tabs/resources-tab.tsx
index 16b6b469158..8b54dbe46aa 100644
--- a/studio/frontend/src/features/settings/tabs/resources-tab.tsx
+++ b/studio/frontend/src/features/settings/tabs/resources-tab.tsx
@@ -1206,7 +1206,7 @@ export function ResourcesTab() {
>
{hfCache?.isCustom ? (
<Button
- variant="ghost"
+ variant="outline"
size="sm"
disabled={cacheSaving}
onClick={() => void saveCacheFolder(null)}
diff --git a/studio/frontend/src/i18n/locales/ar.ts b/studio/frontend/src/i18n/locales/ar.ts
index 4cc30bd27f5..4ed52fc3a58 100644
--- a/studio/frontend/src/i18n/locales/ar.ts
+++ b/studio/frontend/src/i18n/locales/ar.ts
@@ -2595,7 +2595,7 @@ export const ar = {
unavailable: "غير متاح",
detecting: "جارٍ البحث عن وحدات GPU...",
unreadable: "تعذّرت قراءة عتاد هذا الخادم.",
- noGpu: "لم يُكتشف أي GPU مرئي. تُعرض موارد CPU فقط أعلاه.",
+ noGpu: "لم يُكتشف معالج رسومات مخصص.",
noUsableGpu: "لا يوجد على هذا الجهاز أي GPU يمكن لـ PyTorch استخدامه.",
mismatchCpuBuild:
"PyTorch إصدار للـ CPU فقط ({version})، لذا لا يمكن استخدام وحدات GPU أدناه. أصلح التثبيت لاستعادة دعم GPU.",
@@ -2671,7 +2671,7 @@ export const ar = {
"llama.cpp backend gguf استدلال cuda rocm hip vulkan metal cpu gpu مسرّع prebuilt تبديل محرك",
},
modelMemory: {
- title: "ذاكرة النموذج",
+ title: "تحميل النموذج",
keepResident: "إبقاء النموذج في ذاكرة كرت الرسوميات",
keepResidentDescription: "يبقى في ذاكرة كرت الرسوميات بين الرسائل.",
keepResidentHint: "لا تُعاد الأوزان إلى ذاكرة النظام ما دام النموذج محمَّلاً. يعطّل التفريغ التلقائي عند الخمول، وإذا كانت الأوزان موجودة فعلاً في ذاكرة المضيف (ذاكرة موحّدة أو تفريغ جزئي إلى المعالج الرسومي) فإنه يمرّر ‎--mlock‎ أيضاً حتى لا يُخرجها النظام من الذاكرة ويعيد رفعها عند رسالتك التالية.",
@@ -2927,6 +2927,8 @@ export const ar = {
},
chat: {
groups: {
+ replies: { title: "الردود" },
+ thinkingTools: { title: "التفكير والأدوات" },
conversations: { title: "المحادثات" },
files: { title: "الملفات واللصق" },
display: { title: "العرض" },
@@ -3011,11 +3013,20 @@ export const ar = {
showReplyTimesDescription: "إظهار وقت انتهاء كل رد أسفله.",
showReplyTimesKeywords:
"طابع زمني وقت ساعة تاريخ متى انتهى رد رسالة timestamp time clock date reply",
- showGenerationDetails: "تفاصيل التوليد",
+ showGenerationDetails: "معلومات مفصّلة عن السرعة والتحميل",
showGenerationDetailsDescription:
- "عرض سرعتَي معالجة الطلب والرد مباشرةً مع عدد الرموز، ومكان تحميل طبقات النموذج.",
+ "يضيف سرعة قراءة الموجّه وعدد الرموز وقبول المسودات إلى القراءات المباشرة، ويعرض مكان تحميل طبقات النموذج.",
showGenerationDetailsKeywords:
"تفاصيل سرعة رموز إحصاءات طبقات ذاكرة verbose developer stats speed tok/s prefill decode layers offload vram experts",
+ generationStatsUpdates: "إحصاءات السرعة والسياق",
+ generationStatsUpdatesDescription:
+ "يحسب الوضع الفوري الرموز في الثانية واستخدام السياق أثناء بث الرد. أما عند الانتهاء فيحدّثها عند اكتمال كل رد.",
+ generationStatsUpdatesHint:
+ "يحدّث الوضع الفوري مؤشر الرموز في الثانية أسفل الرد وشريط السياق أعلى المحادثة عدة مرات في الثانية أثناء كتابة الرد. أما عند الانتهاء فيحدّثهما مرة واحدة بعد انتهاء كل رد. قد يسبب الوضع الفوري بطئًا أو تراجعًا في الأداء، خاصةً على الأجهزة القديمة أو المشغولة.",
+ generationStatsUpdatesSettled: "عند الانتهاء",
+ generationStatsUpdatesRealtime: "فوري",
+ generationStatsUpdatesKeywords:
+ "فوري مباشر تحديث سرعة سياق إحصاءات realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "تعديل الرد في الردود",
inlineEditResponseDescription: "إبقاء تعديل الرد على كل رد بدلًا من قائمة المزيد.",
turnNavigation: "التنقل بين الجولات",
diff --git a/studio/frontend/src/i18n/locales/cs.ts b/studio/frontend/src/i18n/locales/cs.ts
index b7963f0af2c..af0af89f266 100644
--- a/studio/frontend/src/i18n/locales/cs.ts
+++ b/studio/frontend/src/i18n/locales/cs.ts
@@ -2814,7 +2814,7 @@ export const cs = {
detecting: "Hledání GPU...",
unreadable: "Hardware tohoto serveru se nepodařilo načíst.",
noGpu:
- "Nebylo rozpoznáno žádné viditelné GPU. Výše jsou zobrazeny pouze prostředky CPU.",
+ "Nebyla zjištěna samostatná GPU.",
noUsableGpu:
"Na tomto počítači není žádné GPU, které by PyTorch mohl použít.",
mismatchCpuBuild:
@@ -2904,7 +2904,7 @@ export const cs = {
"llama.cpp backend gguf inference cuda rocm hip vulkan metal cpu gpu akcelerátor accelerator předsestavené prebuilt přepnout switch modul engine vlastní custom cesta path složka folder adresář directory llama-server spustitelný soubor executable binary",
},
modelMemory: {
- title: "Paměť modelu",
+ title: "Načítání modelu",
keepResident: "Ponechat model v paměti GPU",
keepResidentDescription: "Zůstat ve VRAM mezi prompty.",
keepResidentHint:
@@ -3178,6 +3178,8 @@ export const cs = {
},
chat: {
groups: {
+ replies: { title: "Odpovědi" },
+ thinkingTools: { title: "Přemýšlení a nástroje" },
conversations: {
title: "Konverzace",
},
@@ -3276,11 +3278,20 @@ export const cs = {
"Pod každou odpovědí zobrazovat, kdy byla dokončena.",
showReplyTimesKeywords:
"časové razítko čas hodiny datum kdy dokončeno odpověď zpráva timestamp time clock date hour when finished arrived reply response message",
- showGenerationDetails: "Podrobnosti generování",
+ showGenerationDetails: "Podrobné informace o rychlosti a načítání",
showGenerationDetailsDescription:
- "Zobrazovat živé rychlosti promptu a odpovědi s počty tokenů a kam se načetly vrstvy modelu.",
+ "Přidá do živých údajů rychlost čtení výzvy, počty tokenů a přijetí návrhů a ukáže, kam se načetly vrstvy modelu.",
showGenerationDetailsKeywords:
"podrobné vývojář statistiky rychlost vrstvy verbose developer stats speed tok/s prefill decode layers offload vram experts",
+ generationStatsUpdates: "Statistiky rychlosti a kontextu",
+ generationStatsUpdatesDescription:
+ "Režim V reálném čase počítá tokeny za sekundu a využití kontextu během streamování odpovědi. Po dokončení je aktualizuje, až každá odpověď skončí.",
+ generationStatsUpdatesHint:
+ "V reálném čase obnovuje ukazatel tokenů za sekundu pod odpovědí a lištu kontextu nahoře v chatu několikrát za sekundu, zatímco se odpověď píše. Po dokončení je aktualizuje jednou, až každá odpověď skončí. Režim v reálném čase může způsobit zasekávání nebo nižší výkon, hlavně na starších či vytížených počítačích.",
+ generationStatsUpdatesSettled: "Po dokončení",
+ generationStatsUpdatesRealtime: "V reálném čase",
+ generationStatsUpdatesKeywords:
+ "reálný čas živě aktualizace rychlost kontext statistiky realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "Upravit odpověď přímo u odpovědí",
inlineEditResponseDescription:
"Zobrazovat Upravit odpověď u každé odpovědi místo v nabídce Další.",
diff --git a/studio/frontend/src/i18n/locales/de.ts b/studio/frontend/src/i18n/locales/de.ts
index 51b0a208253..3f09a683402 100644
--- a/studio/frontend/src/i18n/locales/de.ts
+++ b/studio/frontend/src/i18n/locales/de.ts
@@ -2627,7 +2627,7 @@ export const de = {
detecting: "Suche nach GPUs...",
unreadable: "Die Hardware dieses Servers konnte nicht gelesen werden.",
noGpu:
- "Keine sichtbare GPU erkannt. Oben werden nur die CPU-Ressourcen angezeigt.",
+ "Keine dedizierte GPU erkannt.",
noUsableGpu: "Auf diesem Rechner ist keine GPU für PyTorch nutzbar.",
mismatchCpuBuild:
"PyTorch ist ein reiner CPU-Build ({version}), daher können die GPUs unten nicht genutzt werden. Reparieren Sie die Installation, um die GPU-Unterstützung wiederherzustellen.",
@@ -2703,7 +2703,7 @@ export const de = {
"llama.cpp backend gguf inferenz cuda rocm hip vulkan metal cpu gpu beschleuniger prebuilt wechseln engine",
},
modelMemory: {
- title: "Modellspeicher",
+ title: "Modell laden",
keepResident: "Modell im GPU-Speicher behalten",
keepResidentDescription: "Zwischen Prompts im VRAM bleiben.",
keepResidentHint: "Die Gewichte werden nicht an den System-RAM zurückgegeben, solange das Modell geladen bleibt. Deaktiviert das automatische Entladen im Leerlauf und übergibt zusätzlich --mlock, wenn die Gewichte tatsächlich im Host-RAM liegen (Unified Memory oder teilweises GPU-Offload), damit das Betriebssystem sie nicht auslagert und beim nächsten Prompt neu hochlädt.",
@@ -2968,6 +2968,8 @@ export const de = {
},
chat: {
groups: {
+ replies: { title: "Antworten" },
+ thinkingTools: { title: "Denken und Werkzeuge" },
conversations: { title: "Gespräche" },
files: { title: "Dateien und Einfügen" },
display: { title: "Anzeige" },
@@ -3052,11 +3054,20 @@ export const de = {
showReplyTimesDescription: "Unter jeder Antwort anzeigen, wann sie fertig war.",
showReplyTimesKeywords:
"Zeitstempel Uhrzeit Datum wann fertig Antwort Nachricht timestamp time clock date reply",
- showGenerationDetails: "Generierungsdetails",
+ showGenerationDetails: "Detaillierte Geschwindigkeits- und Ladeinfos",
showGenerationDetailsDescription:
- "Prompt- und Antwortgeschwindigkeit live mit Tokenzahlen anzeigen, und wo die Schichten eines Modells geladen wurden.",
+ "Ergänzt die Live-Anzeigen um Prompt-Lesegeschwindigkeit, Tokenzahlen und Entwurfsannahme und zeigt, wo die Schichten eines Modells geladen wurden.",
showGenerationDetailsKeywords:
"Details Geschwindigkeit Token Statistik Schichten Auslagerung verbose developer stats speed tok/s prefill decode layers offload vram experts",
+ generationStatsUpdates: "Geschwindigkeits- und Kontextstatistik",
+ generationStatsUpdatesDescription:
+ "Echtzeit zählt Token pro Sekunde und Kontextnutzung, während eine Antwort gestreamt wird. Nach Abschluss aktualisiert sie, sobald jede Antwort fertig ist.",
+ generationStatsUpdatesHint:
+ "Echtzeit aktualisiert die Anzeige der Token pro Sekunde unter einer Antwort und die Kontextleiste oben im Chat mehrmals pro Sekunde, während die Antwort geschrieben wird. Nach Abschluss aktualisiert sie einmal, wenn jede Antwort fertig ist. Echtzeit kann zu Verzögerungen oder geringerer Leistung führen, besonders auf älteren oder ausgelasteten Geräten.",
+ generationStatsUpdatesSettled: "Nach Abschluss",
+ generationStatsUpdatesRealtime: "Echtzeit",
+ generationStatsUpdatesKeywords:
+ "Echtzeit live aktualisieren Geschwindigkeit Kontext Statistik realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "Antwort bearbeiten bei Antworten",
inlineEditResponseDescription: "Antwort bearbeiten bei jeder Antwort anzeigen statt im Menü „Mehr“.",
turnNavigation: "Rundennavigation",
diff --git a/studio/frontend/src/i18n/locales/en.ts b/studio/frontend/src/i18n/locales/en.ts
index 6f7e7bba387..d24bc86eadd 100644
--- a/studio/frontend/src/i18n/locales/en.ts
+++ b/studio/frontend/src/i18n/locales/en.ts
@@ -2612,7 +2612,7 @@ export const en = {
unavailable: "unavailable",
detecting: "Checking for GPUs...",
unreadable: "Could not read this server's hardware.",
- noGpu: "No visible GPU detected. CPU-only resources are shown above.",
+ noGpu: "No dedicated GPU detected.",
noUsableGpu: "No GPU on this machine is usable by PyTorch.",
mismatchCpuBuild:
"PyTorch is a CPU-only build ({version}), so the GPUs below cannot be used. Repair the installation to restore GPU support.",
@@ -2689,7 +2689,7 @@ export const en = {
"llama.cpp backend gguf inference cuda rocm hip vulkan metal cpu gpu accelerator prebuilt switch engine custom path folder directory llama-server executable binary",
},
modelMemory: {
- title: "Model memory",
+ title: "Model loading",
keepResident: "Keep model in GPU memory",
keepResidentDescription: "Stay in VRAM between prompts.",
keepResidentHint: "Don't hand the weights back to system RAM while the model stays loaded. Turns off idle auto-unload, and when the weights do sit in host RAM (unified memory, or a partial GPU offload) it also passes --mlock so the OS can't page them out and re-upload them on your next prompt.",
@@ -2946,6 +2946,8 @@ export const en = {
},
chat: {
groups: {
+ replies: { title: "Replies" },
+ thinkingTools: { title: "Thinking and tools" },
conversations: { title: "Conversations" },
files: { title: "Files & pasting" },
display: { title: "Display" },
@@ -3027,11 +3029,20 @@ export const en = {
showReplyTimesDescription: "Show when each reply finished, below the reply.",
showReplyTimesKeywords:
"timestamp time clock date hour when finished arrived reply response message",
- showGenerationDetails: "Generation details",
+ showGenerationDetails: "Detailed speed and loading info",
showGenerationDetailsDescription:
- "Show live prompt and reply speeds with token counts, and where a model's layers loaded.",
+ "Adds prompt-reading speed, token counts and draft acceptance to the live readouts, and shows where a model's layers loaded.",
showGenerationDetailsKeywords:
- "verbose developer stats speed tok/s prefill decode layers offload vram experts",
+ "generation details verbose developer stats speed tok/s prefill decode layers offload vram experts",
+ generationStatsUpdates: "Speed and context stats",
+ generationStatsUpdatesDescription:
+ "Realtime counts tokens per second and context usage while a reply streams. When done updates them as each reply finishes.",
+ generationStatsUpdatesHint:
+ "Realtime refreshes the tokens per second readout under a reply and the context bar at the top of the chat a few times a second while the reply is written. When done updates them once, after each reply finishes. Realtime may cause lag or slower performance, especially on older or busy machines.",
+ generationStatsUpdatesSettled: "When done",
+ generationStatsUpdatesRealtime: "Realtime",
+ generationStatsUpdatesKeywords:
+ "realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "Edit response on responses",
inlineEditResponseDescription:
"Keep Edit response on every response, instead of in the More menu.",
diff --git a/studio/frontend/src/i18n/locales/es-419.ts b/studio/frontend/src/i18n/locales/es-419.ts
index 5bdad86c9cc..1ce20b00990 100644
--- a/studio/frontend/src/i18n/locales/es-419.ts
+++ b/studio/frontend/src/i18n/locales/es-419.ts
@@ -2840,7 +2840,7 @@ export const es419 = {
detecting: "Buscando GPU...",
unreadable: "No se pudo leer el hardware de este servidor.",
noGpu:
- "No se detectó ninguna GPU visible. Arriba se muestran los recursos solo de CPU.",
+ "No se detectó una GPU dedicada.",
noUsableGpu: "PyTorch no puede usar ninguna GPU de esta computadora.",
mismatchCpuBuild:
"PyTorch es una compilación solo para CPU ({version}), así que las GPU de abajo no se pueden usar. Repara la instalación para restaurar la compatibilidad con GPU.",
@@ -2928,7 +2928,7 @@ export const es419 = {
"llama.cpp backend gguf inferencia inference cuda rocm hip vulkan metal cpu gpu acelerador accelerator prebuilt precompilado cambiar switch motor engine personalizado custom ruta path carpeta folder directorio directory llama-server ejecutable executable binario binary",
},
modelMemory: {
- title: "Memoria del modelo",
+ title: "Carga del modelo",
keepResident: "Mantener el modelo en la memoria de GPU",
keepResidentDescription: "Permanece en la VRAM entre prompts.",
keepResidentHint:
@@ -3207,6 +3207,8 @@ export const es419 = {
},
chat: {
groups: {
+ replies: { title: "Respuestas" },
+ thinkingTools: { title: "Razonamiento y herramientas" },
conversations: {
title: "Conversaciones",
},
@@ -3305,11 +3307,20 @@ export const es419 = {
"Muestra debajo de cada respuesta cuándo terminó.",
showReplyTimesKeywords:
"marca de tiempo hora reloj fecha cuándo terminó llegó respuesta mensaje timestamp time clock date hour when finished arrived reply response message",
- showGenerationDetails: "Detalles de generación",
+ showGenerationDetails: "Información detallada de velocidad y carga",
showGenerationDetailsDescription:
- "Muestra en vivo las velocidades del prompt y de la respuesta con el recuento de tokens, y dónde se cargaron las capas de un modelo.",
+ "Agrega a las lecturas en vivo la velocidad de lectura del prompt, el conteo de tokens y la aceptación de borradores, y muestra dónde se cargaron las capas del modelo.",
showGenerationDetailsKeywords:
"detallado desarrollador estadísticas velocidad capas trasladar expertos verbose developer stats speed tok/s prefill decode layers offload vram experts",
+ generationStatsUpdates: "Estadísticas de velocidad y contexto",
+ generationStatsUpdatesDescription:
+ "En tiempo real cuenta los tokens por segundo y el uso del contexto mientras se transmite una respuesta. Al terminar los actualiza cuando termina cada respuesta.",
+ generationStatsUpdatesHint:
+ "En tiempo real actualiza el indicador de tokens por segundo bajo una respuesta y la barra de contexto en la parte superior del chat varias veces por segundo mientras se escribe la respuesta. Al terminar los actualiza una vez, cuando termina cada respuesta. En tiempo real puede causar lentitud o menor rendimiento, sobre todo en equipos antiguos u ocupados.",
+ generationStatsUpdatesSettled: "Al terminar",
+ generationStatsUpdatesRealtime: "En tiempo real",
+ generationStatsUpdatesKeywords:
+ "tiempo real en vivo actualizar velocidad contexto estadísticas realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "Editar respuesta en las respuestas",
inlineEditResponseDescription:
"Mantén Editar respuesta en cada respuesta, en lugar de en el menú Más.",
diff --git a/studio/frontend/src/i18n/locales/es.ts b/studio/frontend/src/i18n/locales/es.ts
index 1c5bdf2edf9..7b0bab3806d 100644
--- a/studio/frontend/src/i18n/locales/es.ts
+++ b/studio/frontend/src/i18n/locales/es.ts
@@ -2622,7 +2622,7 @@ export const es = {
detecting: "Buscando GPU...",
unreadable: "No se pudo leer el hardware de este servidor.",
noGpu:
- "No se detectó ninguna GPU visible. Arriba se muestran los recursos solo de CPU.",
+ "No se ha detectado una GPU dedicada.",
noUsableGpu: "Ninguna GPU de esta máquina se puede usar con PyTorch.",
mismatchCpuBuild:
"PyTorch es una compilación solo para CPU ({version}), por lo que las GPU de abajo no se pueden usar. Repare la instalación para restaurar la compatibilidad con GPU.",
@@ -2698,7 +2698,7 @@ export const es = {
"llama.cpp backend gguf inferencia cuda rocm hip vulkan metal cpu gpu acelerador prebuilt cambiar motor",
},
modelMemory: {
- title: "Memoria del modelo",
+ title: "Carga del modelo",
keepResident: "Mantener el modelo en la memoria de la GPU",
keepResidentDescription: "Permanece en la VRAM entre mensajes.",
keepResidentHint: "No devuelve los pesos a la RAM del sistema mientras el modelo siga cargado. Desactiva la descarga automática por inactividad y, cuando los pesos sí residen en la RAM del host (memoria unificada o descarga parcial a la GPU), también pasa --mlock para que el sistema operativo no los pagine ni los vuelva a subir en tu siguiente mensaje.",
@@ -2959,6 +2959,8 @@ export const es = {
},
chat: {
groups: {
+ replies: { title: "Respuestas" },
+ thinkingTools: { title: "Razonamiento y herramientas" },
conversations: { title: "Conversaciones" },
files: { title: "Archivos y pegado" },
display: { title: "Visualización" },
@@ -3043,11 +3045,20 @@ export const es = {
showReplyTimesDescription: "Mostrar debajo de cada respuesta cuándo terminó.",
showReplyTimesKeywords:
"marca de tiempo hora fecha cuándo terminó respuesta mensaje timestamp time clock date reply",
- showGenerationDetails: "Detalles de generación",
+ showGenerationDetails: "Información detallada de velocidad y carga",
showGenerationDetailsDescription:
- "Muestra en vivo las velocidades del prompt y de la respuesta con el recuento de tokens, y dónde se cargaron las capas de un modelo.",
+ "Añade a las lecturas en directo la velocidad de lectura del prompt, el recuento de tokens y la aceptación de borradores, y muestra dónde se cargaron las capas del modelo.",
showGenerationDetailsKeywords:
"detalles velocidad tokens estadísticas capas descarga verbose developer stats speed tok/s prefill decode layers offload vram experts",
+ generationStatsUpdates: "Estadísticas de velocidad y contexto",
+ generationStatsUpdatesDescription:
+ "En tiempo real cuenta los tokens por segundo y el uso del contexto mientras se transmite una respuesta. Al terminar los actualiza cuando acaba cada respuesta.",
+ generationStatsUpdatesHint:
+ "En tiempo real actualiza el indicador de tokens por segundo bajo una respuesta y la barra de contexto en la parte superior del chat varias veces por segundo mientras se escribe la respuesta. Al terminar los actualiza una vez, cuando acaba cada respuesta. En tiempo real puede causar retrasos o menor rendimiento, sobre todo en equipos antiguos u ocupados.",
+ generationStatsUpdatesSettled: "Al terminar",
+ generationStatsUpdatesRealtime: "En tiempo real",
+ generationStatsUpdatesKeywords:
+ "tiempo real en directo actualizar velocidad contexto estadísticas realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "Editar respuesta en las respuestas",
inlineEditResponseDescription: "Mostrar Editar respuesta en cada respuesta, en lugar de en el menú Más.",
turnNavigation: "Navegación por turnos",
diff --git a/studio/frontend/src/i18n/locales/fa.ts b/studio/frontend/src/i18n/locales/fa.ts
index fbdf2f09b29..8fdb9e18549 100644
--- a/studio/frontend/src/i18n/locales/fa.ts
+++ b/studio/frontend/src/i18n/locales/fa.ts
@@ -2785,7 +2785,7 @@ export const fa = {
detecting: "در حال بررسی GPUها...",
unreadable: "خواندن سخت‌افزار این سرور ممکن نشد.",
noGpu:
- "هیچ GPU قابل مشاهده‌ای شناسایی نشد. منابع فقط CPU در بالا نمایش داده شده‌اند.",
+ "پردازنده گرافیکی اختصاصی پیدا نشد.",
noUsableGpu: "هیچ GPU در این دستگاه برای PyTorch قابل استفاده نیست.",
mismatchCpuBuild:
"PyTorch یک بیلد فقط CPU است ({version})، بنابراین نمی‌توان از GPUهای زیر استفاده کرد. برای بازگرداندن پشتیبانی GPU، نصب را تعمیر کنید.",
@@ -2873,7 +2873,7 @@ export const fa = {
"بک‌اند استنتاج شتاب‌دهنده موتور مسیر سفارشی پوشه تغییر llama.cpp backend gguf inference cuda rocm hip vulkan metal cpu gpu accelerator prebuilt switch engine custom path folder directory llama-server executable binary",
},
modelMemory: {
- title: "حافظهٔ مدل",
+ title: "بارگذاری مدل",
keepResident: "نگه داشتن مدل در حافظهٔ GPU",
keepResidentDescription: "میان پرامپت‌ها در VRAM بماند.",
keepResidentHint:
@@ -3147,6 +3147,8 @@ export const fa = {
},
chat: {
groups: {
+ replies: { title: "پاسخ‌ها" },
+ thinkingTools: { title: "تفکر و ابزارها" },
conversations: {
title: "گفتگوها",
},
@@ -3246,11 +3248,20 @@ export const fa = {
"زمان پایان هر پاسخ، زیر همان پاسخ نمایش داده شود.",
showReplyTimesKeywords:
"مهر زمانی زمان ساعت تاریخ پایان رسیدن پاسخ پیام timestamp",
- showGenerationDetails: "جزئیات تولید",
+ showGenerationDetails: "اطلاعات دقیق سرعت و بارگذاری",
showGenerationDetailsDescription:
- "سرعت زندهٔ پرامپت و پاسخ همراه با تعداد توکن‌ها، و محل بارگذاری لایه‌های مدل نمایش داده شود.",
+ "سرعت خواندن پرامپت، شمار توکن‌ها و پذیرش پیش‌نویس را به نمایش زنده می‌افزاید و نشان می‌دهد لایه‌های مدل کجا بارگذاری شده‌اند.",
showGenerationDetailsKeywords:
"مفصل توسعه‌دهنده آمار سرعت tok/s prefill decode لایه‌ها offload vram experts",
+ generationStatsUpdates: "آمار سرعت و زمینه",
+ generationStatsUpdatesDescription:
+ "حالت بی‌درنگ توکن در ثانیه و مصرف زمینه را هنگام پخش پاسخ می‌شمارد. حالت پس از پایان آن‌ها را با تمام شدن هر پاسخ به‌روز می‌کند.",
+ generationStatsUpdatesHint:
+ "حالت بی‌درنگ نمایشگر توکن در ثانیه زیر پاسخ و نوار زمینه در بالای گفتگو را هنگام نوشتن پاسخ چند بار در ثانیه تازه می‌کند. حالت پس از پایان آن‌ها را یک بار، پس از پایان هر پاسخ، به‌روز می‌کند. حالت بی‌درنگ ممکن است باعث کندی یا افت کارایی شود، به‌ویژه روی دستگاه‌های قدیمی یا پرمشغله.",
+ generationStatsUpdatesSettled: "پس از پایان",
+ generationStatsUpdatesRealtime: "بی‌درنگ",
+ generationStatsUpdatesKeywords:
+ "بی‌درنگ زنده به‌روزرسانی سرعت زمینه آمار realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "«ویرایش پاسخ» روی پاسخ‌ها",
inlineEditResponseDescription:
"«ویرایش پاسخ» روی هر پاسخ باقی بماند، به‌جای منوی «بیشتر».",
diff --git a/studio/frontend/src/i18n/locales/fr.ts b/studio/frontend/src/i18n/locales/fr.ts
index 1b11fe1d842..7fb0071752d 100644
--- a/studio/frontend/src/i18n/locales/fr.ts
+++ b/studio/frontend/src/i18n/locales/fr.ts
@@ -2625,7 +2625,7 @@ export const fr = {
detecting: "Recherche de GPU...",
unreadable: "Impossible de lire le matériel de ce serveur.",
noGpu:
- "Aucun GPU visible n'a été détecté. Seules les ressources du CPU sont affichées ci-dessus.",
+ "Aucun GPU dédié détecté.",
noUsableGpu: "Aucun GPU de cette machine n'est utilisable par PyTorch.",
mismatchCpuBuild:
"PyTorch est une version CPU uniquement ({version}), les GPU ci-dessous ne peuvent donc pas être utilisés. Réparez l'installation pour rétablir la prise en charge du GPU.",
@@ -2701,7 +2701,7 @@ export const fr = {
"llama.cpp backend gguf inférence cuda rocm hip vulkan metal cpu gpu accélérateur prebuilt changer moteur",
},
modelMemory: {
- title: "Mémoire du modèle",
+ title: "Chargement du modèle",
keepResident: "Conserver le modèle dans la mémoire du GPU",
keepResidentDescription: "Reste en VRAM entre les messages.",
keepResidentHint: "Ne rend pas les poids à la RAM système tant que le modèle reste chargé. Désactive le déchargement automatique en veille et, lorsque les poids résident réellement en RAM hôte (mémoire unifiée ou déchargement GPU partiel), passe aussi --mlock afin que le système ne les décharge pas pour les retransférer au prochain message.",
@@ -2964,6 +2964,8 @@ export const fr = {
},
chat: {
groups: {
+ replies: { title: "Réponses" },
+ thinkingTools: { title: "Réflexion et outils" },
conversations: { title: "Conversations" },
files: { title: "Fichiers et collage" },
display: { title: "Affichage" },
@@ -3048,11 +3050,20 @@ export const fr = {
showReplyTimesDescription: "Afficher sous chaque réponse l'heure à laquelle elle s'est terminée.",
showReplyTimesKeywords:
"horodatage heure date quand terminé réponse message timestamp time clock date reply",
- showGenerationDetails: "Détails de génération",
+ showGenerationDetails: "Infos détaillées de vitesse et de chargement",
showGenerationDetailsDescription:
- "Affiche en direct les vitesses du prompt et de la réponse avec le nombre de jetons, et où les couches d'un modèle ont été chargées.",
+ "Ajoute aux indicateurs en direct la vitesse de lecture du prompt, le nombre de tokens et l'acceptation des brouillons, et indique où les couches du modèle ont été chargées.",
showGenerationDetailsKeywords:
"détails vitesse jetons statistiques couches déchargement verbose developer stats speed tok/s prefill decode layers offload vram experts",
+ generationStatsUpdates: "Statistiques de vitesse et de contexte",
+ generationStatsUpdatesDescription:
+ "En temps réel compte les tokens par seconde et l'utilisation du contexte pendant la diffusion d'une réponse. À la fin les met à jour quand chaque réponse se termine.",
+ generationStatsUpdatesHint:
+ "En temps réel actualise l'indicateur de tokens par seconde sous une réponse et la barre de contexte en haut du chat plusieurs fois par seconde pendant l'écriture de la réponse. À la fin les met à jour une fois, quand chaque réponse se termine. Le mode en temps réel peut provoquer des ralentissements ou baisser les performances, surtout sur les machines anciennes ou chargées.",
+ generationStatsUpdatesSettled: "À la fin",
+ generationStatsUpdatesRealtime: "En temps réel",
+ generationStatsUpdatesKeywords:
+ "temps réel direct mise à jour vitesse contexte statistiques realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "Modifier la réponse sur les réponses",
inlineEditResponseDescription: "Afficher Modifier la réponse sur chaque réponse, plutôt que dans le menu Plus.",
turnNavigation: "Navigation par tours",
diff --git a/studio/frontend/src/i18n/locales/he.ts b/studio/frontend/src/i18n/locales/he.ts
index 16400d45b63..94d9577f31f 100644
--- a/studio/frontend/src/i18n/locales/he.ts
+++ b/studio/frontend/src/i18n/locales/he.ts
@@ -2649,7 +2649,7 @@ export const he = {
unavailable: "לא זמין",
detecting: "בודק קיום מעבדים גרפיים (GPUs)...",
unreadable: "לא ניתן לקרוא את חומרת השרת.",
- noGpu: "לא זוהה GPU גלוי. משאבי CPU בלבד מוצגים למעלה.",
+ noGpu: "לא זוהה GPU ייעודי.",
noUsableGpu: "אף GPU במחשב זה אינו שמיש על ידי PyTorch.",
mismatchCpuBuild:
"גרסת PyTorch מותאמת ל-CPU בלבד ({version}), ולכן לא ניתן להשתמש ב-GPUs שלמטה. תקן את ההתקנה כדי לשחזר תמיכה ב-GPU.",
@@ -2735,7 +2735,7 @@ export const he = {
"llama.cpp backend gguf inference cuda rocm hip vulkan metal cpu gpu accelerator prebuilt switch engine custom path folder directory llama-server executable binary הסקה מאיץ החלפה מנוע מותאם בנייה",
},
modelMemory: {
- title: "זיכרון מודל",
+ title: "טעינת מודל",
keepResident: "שמור מודל בזיכרון ה-GPU",
keepResidentDescription: "הישאר ב-VRAM בין פרומפטים.",
keepResidentHint:
@@ -2998,6 +2998,8 @@ export const he = {
},
chat: {
groups: {
+ replies: { title: "תשובות" },
+ thinkingTools: { title: "חשיבה וכלים" },
conversations: {
title: "שיחות",
},
@@ -3091,11 +3093,20 @@ export const he = {
showReplyTimesDescription: "הצג מתחת לכל תשובה מתי היא הסתיימה.",
showReplyTimesKeywords:
"חותמת זמן שעה תאריך מתי הסתיים תשובה הודעה timestamp time clock date reply",
- showGenerationDetails: "פרטי יצירה",
+ showGenerationDetails: "מידע מפורט על מהירות וטעינה",
showGenerationDetailsDescription:
- "הצגת מהירויות הפרומפט והתשובה בזמן אמת עם מספר האסימונים, והיכן נטענו השכבות של המודל.",
+ "מוסיף למדדים החיים את מהירות קריאת ההנחיה, ספירת טוקנים וקבלת טיוטות, ומציג היכן נטענו שכבות המודל.",
showGenerationDetailsKeywords:
"פרטים מהירות אסימונים סטטיסטיקה שכבות verbose developer stats speed tok/s prefill decode layers offload vram experts",
+ generationStatsUpdates: "נתוני מהירות והקשר",
+ generationStatsUpdatesDescription:
+ "מצב זמן אמת סופר טוקנים לשנייה ושימוש בהקשר בזמן שתשובה מוזרמת. מצב בסיום מעדכן אותם כשכל תשובה מסתיימת.",
+ generationStatsUpdatesHint:
+ "זמן אמת מרענן את מד הטוקנים לשנייה מתחת לתשובה ואת סרגל ההקשר בראש הצ'אט כמה פעמים בשנייה בזמן שהתשובה נכתבת. בסיום מעדכן אותם פעם אחת, אחרי שכל תשובה מסתיימת. זמן אמת עלול לגרום להאטה או לביצועים נמוכים יותר, במיוחד במחשבים ישנים או עמוסים.",
+ generationStatsUpdatesSettled: "בסיום",
+ generationStatsUpdatesRealtime: "זמן אמת",
+ generationStatsUpdatesKeywords:
+ "זמן אמת חי עדכון מהירות הקשר נתונים realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "עריכת תגובה על גבי התגובות",
inlineEditResponseDescription:
"הצג את 'עריכת תגובה' בכל תגובה, במקום בתפריט 'עוד'.",
diff --git a/studio/frontend/src/i18n/locales/hi.ts b/studio/frontend/src/i18n/locales/hi.ts
index 2059fe0fe25..bb83b02e585 100644
--- a/studio/frontend/src/i18n/locales/hi.ts
+++ b/studio/frontend/src/i18n/locales/hi.ts
@@ -2601,7 +2601,7 @@ export const hi = {
unavailable: "उपलब्ध नहीं",
detecting: "GPU खोजे जा रहे हैं...",
unreadable: "इस सर्वर का हार्डवेयर नहीं पढ़ा जा सका।",
- noGpu: "कोई दृश्यमान GPU नहीं मिला। केवल-CPU संसाधन ऊपर दिखाए गए हैं।",
+ noGpu: "कोई अलग GPU नहीं मिला।",
noUsableGpu: "इस मशीन का कोई भी GPU PyTorch के लिए उपयोग योग्य नहीं है।",
mismatchCpuBuild:
"PyTorch केवल-CPU बिल्ड ({version}) है, इसलिए नीचे दिए गए GPU उपयोग नहीं किए जा सकते। GPU समर्थन बहाल करने के लिए इंस्टॉलेशन की मरम्मत करें।",
@@ -2677,7 +2677,7 @@ export const hi = {
"llama.cpp backend gguf इनफ़रेंस cuda rocm hip vulkan metal cpu gpu एक्सेलेरेटर prebuilt बदलें इंजन",
},
modelMemory: {
- title: "मॉडल मेमोरी",
+ title: "मॉडल लोडिंग",
keepResident: "मॉडल को GPU मेमोरी में रखें",
keepResidentDescription: "प्रॉम्प्ट के बीच VRAM में बना रहता है।",
keepResidentHint: "जब तक मॉडल लोड रहता है, वेट्स सिस्टम RAM को वापस नहीं दिए जाते। निष्क्रिय होने पर स्वतः अनलोड बंद कर देता है, और जब वेट्स वास्तव में होस्ट RAM में रहते हैं (यूनिफाइड मेमोरी, या आंशिक GPU ऑफ़लोड) तब --mlock भी भेजता है, ताकि OS उन्हें पेज आउट करके आपके अगले प्रॉम्प्ट पर दोबारा अपलोड न करे।",
@@ -2936,6 +2936,8 @@ export const hi = {
},
chat: {
groups: {
+ replies: { title: "उत्तर" },
+ thinkingTools: { title: "सोच और टूल" },
conversations: { title: "बातचीत" },
files: { title: "फ़ाइलें और पेस्ट" },
display: { title: "दिखावट" },
@@ -3020,11 +3022,20 @@ export const hi = {
showReplyTimesDescription: "हर जवाब के नीचे दिखाएँ कि वह कब पूरा हुआ।",
showReplyTimesKeywords:
"टाइमस्टैम्प समय घड़ी तारीख कब पूरा जवाब संदेश timestamp time clock date reply",
- showGenerationDetails: "जनरेशन विवरण",
+ showGenerationDetails: "गति और लोडिंग की विस्तृत जानकारी",
showGenerationDetailsDescription:
- "टोकन गिनती के साथ प्रॉम्प्ट और जवाब की लाइव गति दिखाएँ, और यह भी कि मॉडल की परतें कहाँ लोड हुईं।",
+ "लाइव रीडआउट में प्रॉम्प्ट पढ़ने की गति, टोकन गिनती और ड्राफ़्ट स्वीकृति जोड़ता है, और दिखाता है कि मॉडल की लेयर कहाँ लोड हुईं।",
showGenerationDetailsKeywords:
"विवरण गति टोकन आँकड़े परतें verbose developer stats speed tok/s prefill decode layers offload vram experts",
+ generationStatsUpdates: "गति और संदर्भ आँकड़े",
+ generationStatsUpdatesDescription:
+ "रीयल-टाइम उत्तर स्ट्रीम होते समय प्रति सेकंड टोकन और संदर्भ उपयोग गिनता है। पूरा होने पर हर उत्तर खत्म होने पर इन्हें अपडेट करता है।",
+ generationStatsUpdatesHint:
+ "रीयल-टाइम उत्तर लिखे जाते समय उत्तर के नीचे प्रति सेकंड टोकन संकेतक और चैट के ऊपर संदर्भ बार को प्रति सेकंड कुछ बार ताज़ा करता है। पूरा होने पर हर उत्तर खत्म होने के बाद इन्हें एक बार अपडेट करता है। रीयल-टाइम से धीमापन या कम प्रदर्शन हो सकता है, खासकर पुरानी या व्यस्त मशीनों पर।",
+ generationStatsUpdatesSettled: "पूरा होने पर",
+ generationStatsUpdatesRealtime: "रीयल-टाइम",
+ generationStatsUpdatesKeywords:
+ "रीयल-टाइम लाइव अपडेट गति संदर्भ आँकड़े realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "जवाबों पर जवाब संपादित करें",
inlineEditResponseDescription: "जवाब संपादित करें को “और” मेन्यू के बजाय हर जवाब पर दिखाएँ।",
turnNavigation: "टर्न नेविगेशन",
diff --git a/studio/frontend/src/i18n/locales/id.ts b/studio/frontend/src/i18n/locales/id.ts
index bd017f1c194..7baba52aedb 100644
--- a/studio/frontend/src/i18n/locales/id.ts
+++ b/studio/frontend/src/i18n/locales/id.ts
@@ -2800,7 +2800,7 @@ export const id = {
detecting: "Memeriksa GPU...",
unreadable: "Tidak dapat membaca perangkat keras server ini.",
noGpu:
- "Tidak ada GPU yang terdeteksi. Sumber daya khusus CPU ditampilkan di atas.",
+ "Tidak ada GPU khusus yang terdeteksi.",
noUsableGpu:
"Tidak ada GPU di mesin ini yang dapat digunakan oleh PyTorch.",
mismatchCpuBuild:
@@ -2889,7 +2889,7 @@ export const id = {
"llama.cpp backend gguf inference cuda rocm hip vulkan metal cpu gpu accelerator prebuilt switch engine custom path folder directory llama-server executable binary inferensi akselerator alihkan kustom jalur direktori",
},
modelMemory: {
- title: "Memori model",
+ title: "Pemuatan model",
keepResident: "Simpan model di memori GPU",
keepResidentDescription: "Tetap di VRAM di antara prompt.",
keepResidentHint:
@@ -3162,6 +3162,8 @@ export const id = {
},
chat: {
groups: {
+ replies: { title: "Balasan" },
+ thinkingTools: { title: "Penalaran dan alat" },
conversations: {
title: "Percakapan",
},
@@ -3260,11 +3262,20 @@ export const id = {
"Tampilkan kapan setiap balasan selesai, di bawah balasan.",
showReplyTimesKeywords:
"timestamp time clock date hour when finished arrived reply response message waktu jam tanggal balasan pesan",
- showGenerationDetails: "Detail pembuatan",
+ showGenerationDetails: "Info kecepatan dan pemuatan terperinci",
showGenerationDetailsDescription:
- "Tampilkan kecepatan prompt dan balasan secara langsung beserta jumlah token, serta tempat lapisan model dimuat.",
+ "Menambahkan kecepatan membaca prompt, jumlah token, dan penerimaan draf ke tampilan langsung, serta menunjukkan tempat lapisan model dimuat.",
showGenerationDetailsKeywords:
"verbose developer stats speed tok/s prefill decode layers offload vram experts statistik kecepatan lapisan",
+ generationStatsUpdates: "Statistik kecepatan dan konteks",
+ generationStatsUpdatesDescription:
+ "Waktu nyata menghitung token per detik dan penggunaan konteks saat balasan sedang dialirkan. Setelah selesai memperbaruinya saat setiap balasan selesai.",
+ generationStatsUpdatesHint:
+ "Waktu nyata menyegarkan penunjuk token per detik di bawah balasan dan bilah konteks di atas chat beberapa kali per detik saat balasan ditulis. Setelah selesai memperbaruinya sekali, setelah setiap balasan selesai. Waktu nyata dapat menyebabkan lag atau kinerja lebih lambat, terutama pada perangkat lama atau sibuk.",
+ generationStatsUpdatesSettled: "Setelah selesai",
+ generationStatsUpdatesRealtime: "Waktu nyata",
+ generationStatsUpdatesKeywords:
+ "waktu nyata langsung perbarui kecepatan konteks statistik realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "Edit respons pada respons",
inlineEditResponseDescription:
"Tampilkan Edit respons di setiap respons, alih-alih di menu Lainnya.",
diff --git a/studio/frontend/src/i18n/locales/it.ts b/studio/frontend/src/i18n/locales/it.ts
index 1d4cce6332e..3022e712b4f 100644
--- a/studio/frontend/src/i18n/locales/it.ts
+++ b/studio/frontend/src/i18n/locales/it.ts
@@ -2527,7 +2527,7 @@ export const it = {
detecting: "Ricerca di GPU...",
unreadable: "Impossibile leggere l'hardware di questo server.",
noGpu:
- "Nessuna GPU visibile rilevata. Sopra sono mostrate le risorse della sola CPU.",
+ "Nessuna GPU dedicata rilevata.",
noUsableGpu: "Nessuna GPU di questa macchina è utilizzabile da PyTorch.",
mismatchCpuBuild:
"PyTorch è una build solo CPU ({version}), quindi le GPU sottostanti non possono essere usate. Ripara l'installazione per ripristinare il supporto GPU.",
@@ -2603,7 +2603,7 @@ export const it = {
"llama.cpp backend gguf inferenza cuda rocm hip vulkan metal cpu gpu acceleratore prebuilt cambiare motore",
},
modelMemory: {
- title: "Memoria del modello",
+ title: "Caricamento del modello",
keepResident: "Mantieni il modello nella memoria della GPU",
keepResidentDescription: "Resta nella VRAM tra un prompt e l'altro.",
keepResidentHint: "Non restituisce i pesi alla RAM di sistema finché il modello resta caricato. Disattiva lo scaricamento automatico in inattività e, quando i pesi risiedono davvero nella RAM host (memoria unificata o offload parziale sulla GPU), passa anche --mlock, così il sistema operativo non li pagina per ricaricarli al prompt successivo.",
@@ -2863,6 +2863,8 @@ export const it = {
},
chat: {
groups: {
+ replies: { title: "Risposte" },
+ thinkingTools: { title: "Ragionamento e strumenti" },
conversations: { title: "Conversazioni" },
files: { title: "File e incolla" },
display: { title: "Visualizzazione" },
@@ -2946,11 +2948,20 @@ export const it = {
showReplyTimesDescription: "Mostra sotto ogni risposta quando è terminata.",
showReplyTimesKeywords:
"orario ora data quando terminata risposta messaggio timestamp time clock date reply",
- showGenerationDetails: "Dettagli generazione",
+ showGenerationDetails: "Info dettagliate su velocità e caricamento",
showGenerationDetailsDescription:
- "Mostra in tempo reale la velocità del prompt e della risposta con il numero di token, e dove sono stati caricati i livelli di un modello.",
+ "Aggiunge alle letture dal vivo la velocità di lettura del prompt, il conteggio dei token e l'accettazione delle bozze, e mostra dove sono stati caricati i livelli del modello.",
showGenerationDetailsKeywords:
"dettagli velocità token statistiche livelli scaricamento verbose developer stats speed tok/s prefill decode layers offload vram experts",
+ generationStatsUpdates: "Statistiche di velocità e contesto",
+ generationStatsUpdatesDescription:
+ "In tempo reale conta i token al secondo e l'uso del contesto mentre una risposta è in streaming. Al termine li aggiorna quando ogni risposta finisce.",
+ generationStatsUpdatesHint:
+ "In tempo reale aggiorna l'indicatore dei token al secondo sotto una risposta e la barra del contesto in cima alla chat più volte al secondo mentre la risposta viene scritta. Al termine li aggiorna una volta, quando ogni risposta finisce. La modalità in tempo reale può causare rallentamenti o prestazioni inferiori, soprattutto su computer datati o occupati.",
+ generationStatsUpdatesSettled: "Al termine",
+ generationStatsUpdatesRealtime: "In tempo reale",
+ generationStatsUpdatesKeywords:
+ "tempo reale dal vivo aggiorna velocità contesto statistiche realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "Modifica risposta nelle risposte",
inlineEditResponseDescription: "Mostra Modifica risposta su ogni risposta, invece che nel menu Altro.",
turnNavigation: "Navigazione dei turni",
diff --git a/studio/frontend/src/i18n/locales/ja.ts b/studio/frontend/src/i18n/locales/ja.ts
index 47cd00f02c5..e4a0691851b 100644
--- a/studio/frontend/src/i18n/locales/ja.ts
+++ b/studio/frontend/src/i18n/locales/ja.ts
@@ -2577,7 +2577,7 @@ export const ja = {
detecting: "GPU を確認しています...",
unreadable: "このサーバーのハードウェアを読み取れませんでした。",
noGpu:
- "利用可能な GPU が検出されませんでした。CPU のみの環境向けのリソース情報は上に表示されています。",
+ "専用 GPU は検出されませんでした。",
noUsableGpu: "このマシンの GPU はいずれも PyTorch から利用できません。",
mismatchCpuBuild:
"PyTorch は CPU 専用ビルド ({version}) のため、下の GPU は利用できません。インストールを修復すると GPU サポートが復元されます。",
@@ -2653,7 +2653,7 @@ export const ja = {
"llama.cpp backend gguf 推論 cuda rocm hip vulkan metal cpu gpu アクセラレータ prebuilt 切り替え エンジン",
},
modelMemory: {
- title: "モデルメモリ",
+ title: "モデルの読み込み",
keepResident: "モデルを GPU メモリに保持",
keepResidentDescription: "プロンプト間も VRAM に常駐します。",
keepResidentHint: "モデルがロードされている間、重みをシステム RAM に戻しません。アイドル時の自動アンロードを無効にし、重みが実際にホスト RAM 上にある場合(ユニファイドメモリ、または GPU への部分オフロード)は --mlock も渡すため、OS が重みをページアウトして次のプロンプトで再アップロードすることがなくなります。",
@@ -2915,6 +2915,8 @@ export const ja = {
},
chat: {
groups: {
+ replies: { title: "応答" },
+ thinkingTools: { title: "思考とツール" },
conversations: { title: "会話" },
files: { title: "ファイルと貼り付け" },
display: { title: "表示" },
@@ -2997,11 +2999,20 @@ export const ja = {
showReplyTimesDescription: "各応答の下に、完了した時刻を表示します。",
showReplyTimesKeywords:
"タイムスタンプ 時刻 時間 日付 完了 応答 メッセージ timestamp time clock date reply",
- showGenerationDetails: "生成の詳細",
+ showGenerationDetails: "速度と読み込みの詳細情報",
showGenerationDetailsDescription:
- "プロンプトと応答の速度をトークン数とともにリアルタイムで表示し、モデルのレイヤーの読み込み先も表示します。",
+ "ライブ表示にプロンプトの読み取り速度、トークン数、ドラフトの採用率を追加し、モデルのレイヤーの読み込み先を表示します。",
showGenerationDetailsKeywords:
"詳細 速度 トークン 統計 レイヤー オフロード verbose developer stats speed tok/s prefill decode layers offload vram experts",
+ generationStatsUpdates: "速度とコンテキストの統計",
+ generationStatsUpdatesDescription:
+ "リアルタイムでは、応答のストリーミング中に毎秒トークン数とコンテキスト使用量を数えます。完了時は各応答の終了時に更新します。",
+ generationStatsUpdatesHint:
+ "リアルタイムでは、応答の生成中に応答の下の毎秒トークン数表示とチャット上部のコンテキストバーを毎秒数回更新します。完了時は各応答が終わった後に一度だけ更新します。リアルタイムは、特に古いマシンや負荷の高いマシンで動作の遅れやパフォーマンス低下を招くことがあります。",
+ generationStatsUpdatesSettled: "完了時",
+ generationStatsUpdatesRealtime: "リアルタイム",
+ generationStatsUpdatesKeywords:
+ "リアルタイム ライブ 更新 速度 コンテキスト 統計 realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "応答に応答の編集を表示",
inlineEditResponseDescription: "応答の編集を「その他」メニューではなく、各応答に表示します。",
turnNavigation: "ターンナビゲーション",
diff --git a/studio/frontend/src/i18n/locales/ko.ts b/studio/frontend/src/i18n/locales/ko.ts
index 7c592ed7fdb..0690b95f35b 100644
--- a/studio/frontend/src/i18n/locales/ko.ts
+++ b/studio/frontend/src/i18n/locales/ko.ts
@@ -2589,7 +2589,7 @@ export const ko = {
unavailable: "사용할 수 없음",
detecting: "GPU를 확인하는 중...",
unreadable: "이 서버의 하드웨어를 읽을 수 없습니다.",
- noGpu: "인식되는 GPU가 없습니다. 위에는 CPU 관련 리소스만 표시됩니다.",
+ noGpu: "전용 GPU가 감지되지 않았습니다.",
noUsableGpu: "이 컴퓨터의 GPU 중 PyTorch에서 사용할 수 있는 것이 없습니다.",
mismatchCpuBuild:
"PyTorch가 CPU 전용 빌드({version})이므로 아래 GPU를 사용할 수 없습니다. 설치를 복구하면 GPU 지원이 복원됩니다.",
@@ -2665,7 +2665,7 @@ export const ko = {
"llama.cpp backend gguf 추론 cuda rocm hip vulkan metal cpu gpu 가속기 prebuilt 전환 엔진",
},
modelMemory: {
- title: "모델 메모리",
+ title: "모델 로딩",
keepResident: "모델을 GPU 메모리에 유지",
keepResidentDescription: "프롬프트 사이에도 VRAM에 상주합니다.",
keepResidentHint: "모델이 로드되어 있는 동안 가중치를 시스템 RAM으로 되돌리지 않습니다. 유휴 시 자동 언로드를 끄고, 가중치가 실제로 호스트 RAM에 있는 경우(통합 메모리 또는 부분 GPU 오프로드)에는 --mlock도 전달하여 운영체제가 가중치를 페이지 아웃했다가 다음 프롬프트에서 다시 업로드하지 않도록 합니다.",
@@ -2922,6 +2922,8 @@ export const ko = {
},
chat: {
groups: {
+ replies: { title: "응답" },
+ thinkingTools: { title: "생각과 도구" },
conversations: { title: "대화" },
files: { title: "파일 및 붙여넣기" },
display: { title: "표시" },
@@ -3004,11 +3006,20 @@ export const ko = {
showReplyTimesDescription: "각 응답 아래에 완료된 시간을 표시합니다.",
showReplyTimesKeywords:
"타임스탬프 시간 시각 날짜 완료 응답 메시지 timestamp time clock date reply",
- showGenerationDetails: "생성 세부 정보",
+ showGenerationDetails: "자세한 속도 및 로딩 정보",
showGenerationDetailsDescription:
- "프롬프트 및 응답 속도와 토큰 수를 실시간으로 표시하고, 모델의 레이어가 어디에 로드되었는지 보여줍니다.",
+ "실시간 표시에 프롬프트 읽기 속도, 토큰 수, 초안 채택률을 더하고 모델 레이어가 로드된 위치를 보여 줍니다.",
showGenerationDetailsKeywords:
"세부 속도 토큰 통계 레이어 오프로드 verbose developer stats speed tok/s prefill decode layers offload vram experts",
+ generationStatsUpdates: "속도 및 컨텍스트 통계",
+ generationStatsUpdatesDescription:
+ "실시간은 응답이 스트리밍되는 동안 초당 토큰과 컨텍스트 사용량을 셉니다. 완료 시는 각 응답이 끝날 때 업데이트합니다.",
+ generationStatsUpdatesHint:
+ "실시간은 응답이 작성되는 동안 응답 아래의 초당 토큰 표시와 채팅 상단의 컨텍스트 막대를 초당 몇 번씩 새로 고칩니다. 완료 시는 각 응답이 끝난 뒤 한 번만 업데이트합니다. 실시간은 특히 오래되었거나 바쁜 기기에서 지연이나 성능 저하를 일으킬 수 있습니다.",
+ generationStatsUpdatesSettled: "완료 시",
+ generationStatsUpdatesRealtime: "실시간",
+ generationStatsUpdatesKeywords:
+ "실시간 라이브 업데이트 속도 컨텍스트 통계 realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "응답에 응답 편집 표시",
inlineEditResponseDescription: "응답 편집을 더보기 메뉴 대신 모든 응답에 표시합니다.",
turnNavigation: "턴 탐색",
diff --git a/studio/frontend/src/i18n/locales/nl.ts b/studio/frontend/src/i18n/locales/nl.ts
index 7de8e9e202b..7ec8c91d815 100644
--- a/studio/frontend/src/i18n/locales/nl.ts
+++ b/studio/frontend/src/i18n/locales/nl.ts
@@ -2836,7 +2836,7 @@ export const nl = {
detecting: "Zoeken naar GPU's...",
unreadable: "Kan de hardware van deze server niet uitlezen.",
noGpu:
- "Geen zichtbare GPU gedetecteerd. Hierboven worden alleen CPU-bronnen getoond.",
+ "Geen aparte GPU gedetecteerd.",
noUsableGpu:
"Geen enkele GPU op deze machine is bruikbaar voor PyTorch.",
mismatchCpuBuild:
@@ -2927,7 +2927,7 @@ export const nl = {
"llama.cpp backend gguf inferentie inference cuda rocm hip vulkan metal cpu gpu versneller accelerator prebuilt wisselen engine aangepast pad map directory llama-server uitvoerbaar bestand binary",
},
modelMemory: {
- title: "Modelgeheugen",
+ title: "Model laden",
keepResident: "Model in GPU-geheugen houden",
keepResidentDescription: "Tussen prompts in VRAM blijven.",
keepResidentHint:
@@ -3203,6 +3203,8 @@ export const nl = {
},
chat: {
groups: {
+ replies: { title: "Antwoorden" },
+ thinkingTools: { title: "Denken en tools" },
conversations: {
title: "Gesprekken",
},
@@ -3301,11 +3303,20 @@ export const nl = {
"Onder elk antwoord tonen wanneer het klaar was.",
showReplyTimesKeywords:
"tijdstempel tijd klok datum uur wanneer voltooid aangekomen antwoord reactie bericht timestamp time",
- showGenerationDetails: "Generatiedetails",
+ showGenerationDetails: "Gedetailleerde snelheids- en laadinfo",
showGenerationDetailsDescription:
- "Live prompt- en antwoordsnelheden tonen met aantallen tokens, en waar de lagen van een model zijn geladen.",
+ "Voegt promptleessnelheid, tokenaantallen en conceptacceptatie toe aan de live weergaven en toont waar de lagen van een model zijn geladen.",
showGenerationDetailsKeywords:
"uitgebreid ontwikkelaar statistieken snelheid tok/s prefill decode lagen offload vram experts verbose developer stats speed layers",
+ generationStatsUpdates: "Snelheids- en contextstatistieken",
+ generationStatsUpdatesDescription:
+ "Realtime telt tokens per seconde en contextgebruik terwijl een antwoord wordt gestreamd. Na afloop werkt ze bij zodra elk antwoord klaar is.",
+ generationStatsUpdatesHint:
+ "Realtime ververst de weergave van tokens per seconde onder een antwoord en de contextbalk boven in de chat een paar keer per seconde terwijl het antwoord wordt geschreven. Na afloop werkt ze één keer bij, als elk antwoord klaar is. Realtime kan vertraging of tragere prestaties veroorzaken, vooral op oudere of drukke computers.",
+ generationStatsUpdatesSettled: "Na afloop",
+ generationStatsUpdatesRealtime: "Realtime",
+ generationStatsUpdatesKeywords:
+ "realtime live bijwerken snelheid context statistieken realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "Antwoord bewerken bij antwoorden",
inlineEditResponseDescription:
"Antwoord bewerken bij elk antwoord tonen, in plaats van in het menu Meer.",
diff --git a/studio/frontend/src/i18n/locales/pl.ts b/studio/frontend/src/i18n/locales/pl.ts
index 6b759186373..5cf3afa0ca3 100644
--- a/studio/frontend/src/i18n/locales/pl.ts
+++ b/studio/frontend/src/i18n/locales/pl.ts
@@ -2831,7 +2831,7 @@ export const pl = {
detecting: "Wyszukiwanie procesorów GPU...",
unreadable: "Nie udało się odczytać sprzętu tego serwera.",
noGpu:
- "Nie wykryto widocznego GPU. Powyżej pokazano zasoby samego CPU.",
+ "Nie wykryto osobnej karty GPU.",
noUsableGpu:
"Żaden GPU na tym komputerze nie może być używany przez PyTorch.",
mismatchCpuBuild:
@@ -2922,7 +2922,7 @@ export const pl = {
"llama.cpp backend gguf inference cuda rocm hip vulkan metal cpu gpu accelerator prebuilt switch engine custom path folder directory llama-server executable binary wnioskowanie akcelerator przełącz silnik ścieżka katalog plik wykonywalny",
},
modelMemory: {
- title: "Pamięć modelu",
+ title: "Ładowanie modelu",
keepResident: "Utrzymuj model w pamięci GPU",
keepResidentDescription: "Pozostaje w VRAM między promptami.",
keepResidentHint:
@@ -3198,6 +3198,8 @@ export const pl = {
},
chat: {
groups: {
+ replies: { title: "Odpowiedzi" },
+ thinkingTools: { title: "Myślenie i narzędzia" },
conversations: {
title: "Rozmowy",
},
@@ -3296,11 +3298,20 @@ export const pl = {
"Pokazuj pod odpowiedzią, kiedy się zakończyła.",
showReplyTimesKeywords:
"timestamp time clock date hour when finished arrived reply response message znacznik czasu godzina data odpowiedź wiadomość",
- showGenerationDetails: "Szczegóły generowania",
+ showGenerationDetails: "Szczegółowe informacje o szybkości i ładowaniu",
showGenerationDetailsDescription:
- "Pokazuj na żywo szybkość przetwarzania promptu i odpowiedzi z liczbą tokenów oraz miejsce wczytania warstw modelu.",
+ "Dodaje do odczytów na żywo szybkość czytania promptu, liczbę tokenów i akceptację szkiców oraz pokazuje, gdzie załadowano warstwy modelu.",
showGenerationDetailsKeywords:
"verbose developer stats speed tok/s prefill decode layers offload vram experts szczegóły statystyki szybkość warstwy eksperci",
+ generationStatsUpdates: "Statystyki szybkości i kontekstu",
+ generationStatsUpdatesDescription:
+ "Tryb Na bieżąco liczy tokeny na sekundę i użycie kontekstu podczas strumieniowania odpowiedzi. Po zakończeniu aktualizuje je, gdy każda odpowiedź się skończy.",
+ generationStatsUpdatesHint:
+ "Na bieżąco odświeża wskaźnik tokenów na sekundę pod odpowiedzią i pasek kontekstu u góry czatu kilka razy na sekundę, gdy odpowiedź jest pisana. Po zakończeniu aktualizuje je raz, gdy każda odpowiedź się skończy. Tryb na bieżąco może powodować opóźnienia lub spadek wydajności, zwłaszcza na starszych lub obciążonych komputerach.",
+ generationStatsUpdatesSettled: "Po zakończeniu",
+ generationStatsUpdatesRealtime: "Na bieżąco",
+ generationStatsUpdatesKeywords:
+ "na bieżąco na żywo aktualizacja szybkość kontekst statystyki realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "Edycja odpowiedzi przy odpowiedziach",
inlineEditResponseDescription:
"Pokazuj opcję Edytuj odpowiedź przy każdej odpowiedzi zamiast w menu Więcej.",
diff --git a/studio/frontend/src/i18n/locales/pt-br.ts b/studio/frontend/src/i18n/locales/pt-br.ts
index b23d866a50e..59dbbe3a863 100644
--- a/studio/frontend/src/i18n/locales/pt-br.ts
+++ b/studio/frontend/src/i18n/locales/pt-br.ts
@@ -2609,7 +2609,7 @@ export const ptBR = {
unavailable: "indisponível",
detecting: "Procurando GPUs...",
unreadable: "Não foi possível ler o hardware deste servidor.",
- noGpu: "Nenhuma GPU visível detectada. Os recursos somente CPU aparecem acima.",
+ noGpu: "Nenhuma GPU dedicada detectada.",
noUsableGpu: "Nenhuma GPU desta máquina pode ser usada pelo PyTorch.",
mismatchCpuBuild:
"O PyTorch é uma compilação somente CPU ({version}), então as GPUs abaixo não podem ser usadas. Repare a instalação para restaurar o suporte a GPU.",
@@ -2685,7 +2685,7 @@ export const ptBR = {
"llama.cpp backend gguf inference cuda rocm hip vulkan metal cpu gpu accelerator prebuilt switch engine",
},
modelMemory: {
- title: "Memória do modelo",
+ title: "Carregamento do modelo",
keepResident: "Manter o modelo na memória da GPU",
keepResidentDescription: "Permanece na VRAM entre os prompts.",
keepResidentHint: "Não devolve os pesos à RAM do sistema enquanto o modelo continuar carregado. Desativa a descarga automática por inatividade e, quando os pesos de fato ficam na RAM do host (memória unificada ou offload parcial para a GPU), também passa --mlock, para que o sistema operacional não os pagine e os reenvie no próximo prompt.",
@@ -2944,6 +2944,8 @@ export const ptBR = {
},
chat: {
groups: {
+ replies: { title: "Respostas" },
+ thinkingTools: { title: "Raciocínio e ferramentas" },
conversations: { title: "Conversas" },
files: { title: "Arquivos e colagem" },
display: { title: "Exibição" },
@@ -3025,11 +3027,20 @@ export const ptBR = {
showReplyTimesDescription: "Mostrar abaixo de cada resposta quando ela terminou.",
showReplyTimesKeywords:
"carimbo de data hora data quando terminou resposta mensagem timestamp time clock date reply",
- showGenerationDetails: "Detalhes da geração",
+ showGenerationDetails: "Informações detalhadas de velocidade e carregamento",
showGenerationDetailsDescription:
- "Mostrar ao vivo a velocidade do prompt e da resposta com a contagem de tokens, e onde as camadas de um modelo foram carregadas.",
+ "Adiciona às leituras ao vivo a velocidade de leitura do prompt, a contagem de tokens e a aceitação de rascunhos, e mostra onde as camadas do modelo foram carregadas.",
showGenerationDetailsKeywords:
"detalhes geração velocidade tokens estatísticas camadas especialistas desenvolvedor verbose developer stats speed tok/s prefill decode layers offload vram experts",
+ generationStatsUpdates: "Estatísticas de velocidade e contexto",
+ generationStatsUpdatesDescription:
+ "Em tempo real conta tokens por segundo e uso do contexto enquanto uma resposta é transmitida. Ao terminar atualiza quando cada resposta termina.",
+ generationStatsUpdatesHint:
+ "Em tempo real atualiza o indicador de tokens por segundo abaixo de uma resposta e a barra de contexto no topo do chat algumas vezes por segundo enquanto a resposta é escrita. Ao terminar atualiza uma vez, quando cada resposta termina. O modo em tempo real pode causar lentidão ou queda de desempenho, principalmente em máquinas antigas ou ocupadas.",
+ generationStatsUpdatesSettled: "Ao terminar",
+ generationStatsUpdatesRealtime: "Em tempo real",
+ generationStatsUpdatesKeywords:
+ "tempo real ao vivo atualizar velocidade contexto estatísticas realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "Editar resposta nas respostas",
inlineEditResponseDescription: "Mostrar Editar resposta em cada resposta, em vez de no menu Mais.",
turnNavigation: "Navegação por turnos",
diff --git a/studio/frontend/src/i18n/locales/ru.ts b/studio/frontend/src/i18n/locales/ru.ts
index 04366c839d2..aae7772338b 100644
--- a/studio/frontend/src/i18n/locales/ru.ts
+++ b/studio/frontend/src/i18n/locales/ru.ts
@@ -2607,7 +2607,7 @@ export const ru = {
detecting: "Поиск GPU...",
unreadable: "Не удалось прочитать оборудование этого сервера.",
noGpu:
- "Доступные GPU не обнаружены. Выше показаны ресурсы только для CPU.",
+ "Отдельный GPU не обнаружен.",
noUsableGpu: "Ни один GPU на этой машине недоступен для PyTorch.",
mismatchCpuBuild:
"PyTorch собран только для CPU ({version}), поэтому GPU ниже использовать нельзя. Восстановите установку, чтобы вернуть поддержку GPU.",
@@ -2683,7 +2683,7 @@ export const ru = {
"llama.cpp backend gguf инференс cuda rocm hip vulkan metal cpu gpu ускоритель prebuilt переключить движок",
},
modelMemory: {
- title: "Память модели",
+ title: "Загрузка модели",
keepResident: "Держать модель в памяти GPU",
keepResidentDescription: "Остаётся в видеопамяти между запросами.",
keepResidentHint: "Веса не возвращаются в системную ОЗУ, пока модель загружена. Отключает автовыгрузку по простою, а если веса действительно находятся в ОЗУ хоста (единая память или частичная выгрузка на GPU), дополнительно передаёт --mlock, чтобы ОС не вытесняла их и не загружала заново при следующем запросе.",
@@ -2943,6 +2943,8 @@ export const ru = {
},
chat: {
groups: {
+ replies: { title: "Ответы" },
+ thinkingTools: { title: "Размышления и инструменты" },
conversations: { title: "Беседы" },
files: { title: "Файлы и вставка" },
display: { title: "Отображение" },
@@ -3024,11 +3026,20 @@ export const ru = {
showReplyTimesDescription: "Показывать под каждым ответом, когда он завершился.",
showReplyTimesKeywords:
"время метка дата когда завершено ответ сообщение timestamp time clock date reply",
- showGenerationDetails: "Подробности генерации",
+ showGenerationDetails: "Подробности о скорости и загрузке",
showGenerationDetailsDescription:
- "Показывать в реальном времени скорость обработки промпта и ответа с числом токенов, а также где загружены слои модели.",
+ "Добавляет к живым показателям скорость чтения запроса, число токенов и принятие черновиков, а также показывает, куда загружены слои модели.",
showGenerationDetailsKeywords:
"подробности скорость токены статистика слои выгрузка verbose developer stats speed tok/s prefill decode layers offload vram experts",
+ generationStatsUpdates: "Статистика скорости и контекста",
+ generationStatsUpdatesDescription:
+ "В реальном времени считает токены в секунду и использование контекста, пока ответ передаётся. По завершении обновляет их, когда каждый ответ закончен.",
+ generationStatsUpdatesHint:
+ "В реальном времени обновляет показатель токенов в секунду под ответом и полосу контекста вверху чата несколько раз в секунду, пока ответ пишется. По завершении обновляет их один раз, когда каждый ответ закончен. Режим реального времени может вызывать подтормаживания или снижать производительность, особенно на старых или загруженных компьютерах.",
+ generationStatsUpdatesSettled: "По завершении",
+ generationStatsUpdatesRealtime: "В реальном времени",
+ generationStatsUpdatesKeywords:
+ "реальное время вживую обновление скорость контекст статистика realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "Изменение ответа в ответах",
inlineEditResponseDescription: "Показывать «Изменить ответ» у каждого ответа, а не в меню «Ещё».",
turnNavigation: "Навигация по ходам",
diff --git a/studio/frontend/src/i18n/locales/sv.ts b/studio/frontend/src/i18n/locales/sv.ts
index 0d55896b0e6..8371ec76bc5 100644
--- a/studio/frontend/src/i18n/locales/sv.ts
+++ b/studio/frontend/src/i18n/locales/sv.ts
@@ -2722,7 +2722,7 @@ export const sv = {
detecting: "Kontrollerar GPU:er ...",
unreadable: "Det gick inte att läsa serverns maskinvara.",
noGpu:
- "Ingen synlig GPU upptäcktes. Resurser endast för CPU visas ovan.",
+ "Ingen separat GPU hittades.",
noUsableGpu: "Ingen GPU på den här datorn kan användas av PyTorch.",
mismatchCpuBuild:
"PyTorch är en version endast för CPU ({version}), så GPU:erna nedan kan inte användas. Reparera installationen för att återställa GPU-stödet.",
@@ -2810,7 +2810,7 @@ export const sv = {
"llama.cpp bakände gguf inferens cuda rocm hip vulkan metal cpu gpu accelerator förbyggd byta motor anpassad sökväg katalog mapp llama-server körbar fil binär llama.cpp backend gguf inference cuda rocm hip vulkan metal cpu gpu accelerator prebuilt switch engine custom path folder directory llama-server executable binary",
},
modelMemory: {
- title: "Modellminne",
+ title: "Modellinläsning",
keepResident: "Behåll modellen i GPU-minnet",
keepResidentDescription: "Ligger kvar i VRAM mellan uppmaningar.",
keepResidentHint:
@@ -3077,6 +3077,8 @@ export const sv = {
},
chat: {
groups: {
+ replies: { title: "Svar" },
+ thinkingTools: { title: "Tänkande och verktyg" },
conversations: {
title: "Konversationer",
},
@@ -3173,11 +3175,20 @@ export const sv = {
showReplyTimesDescription: "Visa under varje svar när det blev klart.",
showReplyTimesKeywords:
"tidsstämpel tid klocka datum när klar svar meddelande timestamp time clock date reply",
- showGenerationDetails: "Genereringsdetaljer",
+ showGenerationDetails: "Detaljerad hastighets- och laddningsinfo",
showGenerationDetailsDescription:
- "Visa live hastighet för prompt och svar med antal token, och var en modells lager lästes in.",
+ "Lägger till läshastighet för prompten, tokenantal och utkastgodkännande i livevisningen och visar var modellens lager laddades.",
showGenerationDetailsKeywords:
"detaljer hastighet token statistik lager avlastning verbose developer stats speed tok/s prefill decode layers offload vram experts",
+ generationStatsUpdates: "Hastighets- och kontextstatistik",
+ generationStatsUpdatesDescription:
+ "Realtid räknar token per sekund och kontextanvändning medan ett svar strömmas. När klart uppdaterar dem när varje svar är färdigt.",
+ generationStatsUpdatesHint:
+ "Realtid uppdaterar visningen av token per sekund under ett svar och kontextfältet högst upp i chatten några gånger per sekund medan svaret skrivs. När klart uppdaterar dem en gång, när varje svar är färdigt. Realtid kan orsaka fördröjning eller sämre prestanda, särskilt på äldre eller belastade datorer.",
+ generationStatsUpdatesSettled: "När klart",
+ generationStatsUpdatesRealtime: "Realtid",
+ generationStatsUpdatesKeywords:
+ "realtid live uppdatera hastighet kontext statistik realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "Redigera svar på svaren",
inlineEditResponseDescription:
"Visa Redigera svar på varje svar i stället för i menyn Mer.",
diff --git a/studio/frontend/src/i18n/locales/th.ts b/studio/frontend/src/i18n/locales/th.ts
index 585d72f42a3..6900fe45756 100644
--- a/studio/frontend/src/i18n/locales/th.ts
+++ b/studio/frontend/src/i18n/locales/th.ts
@@ -2685,7 +2685,7 @@ export const th = {
unavailable: "ไม่พร้อมใช้งาน",
detecting: "กำลังตรวจหา GPU...",
unreadable: "ไม่สามารถอ่านฮาร์ดแวร์ของเซิร์ฟเวอร์นี้ได้",
- noGpu: "ไม่พบ GPU ทรัพยากรที่แสดงด้านบนเป็นของ CPU เท่านั้น",
+ noGpu: "ไม่พบ GPU แยก",
noUsableGpu: "ไม่มี GPU ในเครื่องนี้ที่ PyTorch ใช้งานได้",
mismatchCpuBuild:
"PyTorch เป็นบิลด์สำหรับ CPU เท่านั้น ({version}) จึงใช้ GPU ด้านล่างไม่ได้ ซ่อมแซมการติดตั้งเพื่อกู้คืนการรองรับ GPU",
@@ -2768,7 +2768,7 @@ export const th = {
"แบ็กเอนด์ การอนุมาน เอนจิน โฟลเดอร์ พาธ llama.cpp backend gguf inference cuda rocm hip vulkan metal cpu gpu accelerator prebuilt switch engine custom path folder directory llama-server executable binary",
},
modelMemory: {
- title: "หน่วยความจำโมเดล",
+ title: "การโหลดโมเดล",
keepResident: "เก็บโมเดลไว้ในหน่วยความจำ GPU",
keepResidentDescription: "คงอยู่ใน VRAM ระหว่างพรอมต์",
keepResidentHint:
@@ -3033,6 +3033,8 @@ export const th = {
},
chat: {
groups: {
+ replies: { title: "คำตอบ" },
+ thinkingTools: { title: "การคิดและเครื่องมือ" },
conversations: {
title: "การสนทนา",
},
@@ -3124,11 +3126,20 @@ export const th = {
showReplyTimesDescription: "แสดงเวลาที่การตอบกลับแต่ละครั้งเสร็จสิ้นไว้ใต้การตอบกลับ",
showReplyTimesKeywords:
"เวลา นาฬิกา วันที่ ชั่วโมง เมื่อใด เสร็จ มาถึง ตอบกลับ คำตอบ ข้อความ timestamp time clock date reply response message",
- showGenerationDetails: "รายละเอียดการสร้าง",
+ showGenerationDetails: "ข้อมูลความเร็วและการโหลดแบบละเอียด",
showGenerationDetailsDescription:
- "แสดงความเร็วของพรอมต์และการตอบกลับแบบเรียลไทม์พร้อมจำนวนโทเค็น และตำแหน่งที่โหลดเลเยอร์ของโมเดล",
+ "เพิ่มความเร็วการอ่านพรอมต์ จำนวนโทเค็น และอัตราการยอมรับร่างลงในการแสดงผลสด และแสดงว่าเลเยอร์ของโมเดลโหลดไว้ที่ใด",
showGenerationDetailsKeywords:
"ละเอียด นักพัฒนา สถิติ ความเร็ว เลเยอร์ verbose developer stats speed tok/s prefill decode layers offload vram experts",
+ generationStatsUpdates: "สถิติความเร็วและบริบท",
+ generationStatsUpdatesDescription:
+ "แบบเรียลไทม์จะนับโทเค็นต่อวินาทีและการใช้บริบทระหว่างที่คำตอบกำลังสตรีม แบบเมื่อเสร็จจะอัปเดตเมื่อแต่ละคำตอบจบ",
+ generationStatsUpdatesHint:
+ "แบบเรียลไทม์จะรีเฟรชตัวแสดงโทเค็นต่อวินาทีใต้คำตอบและแถบบริบทด้านบนแชทหลายครั้งต่อวินาทีขณะเขียนคำตอบ แบบเมื่อเสร็จจะอัปเดตครั้งเดียวหลังแต่ละคำตอบจบ แบบเรียลไทม์อาจทำให้หน่วงหรือประสิทธิภาพลดลง โดยเฉพาะบนเครื่องรุ่นเก่าหรือเครื่องที่ทำงานหนัก",
+ generationStatsUpdatesSettled: "เมื่อเสร็จ",
+ generationStatsUpdatesRealtime: "เรียลไทม์",
+ generationStatsUpdatesKeywords:
+ "เรียลไทม์ สด อัปเดต ความเร็ว บริบท สถิติ realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "แก้ไขคำตอบบนคำตอบ",
inlineEditResponseDescription:
"แสดง “แก้ไขคำตอบ” บนทุกคำตอบ แทนที่จะอยู่ในเมนูเพิ่มเติม",
diff --git a/studio/frontend/src/i18n/locales/tr.ts b/studio/frontend/src/i18n/locales/tr.ts
index 005dc3a59ed..ecf917192c9 100644
--- a/studio/frontend/src/i18n/locales/tr.ts
+++ b/studio/frontend/src/i18n/locales/tr.ts
@@ -2793,7 +2793,7 @@ export const tr = {
detecting: "GPU'lar denetleniyor...",
unreadable: "Bu sunucunun donanımı okunamadı.",
noGpu:
- "Görünür GPU algılanmadı. Yukarıda yalnızca CPU kaynakları gösteriliyor.",
+ "Ayrı bir GPU algılanmadı.",
noUsableGpu: "Bu makinede PyTorch'un kullanabileceği bir GPU yok.",
mismatchCpuBuild:
"PyTorch yalnızca CPU derlemesi ({version}), bu nedenle aşağıdaki GPU'lar kullanılamaz. GPU desteğini geri getirmek için kurulumu onarın.",
@@ -2883,7 +2883,7 @@ export const tr = {
"llama.cpp arka uç gguf çıkarım cuda rocm hip vulkan metal cpu gpu hızlandırıcı hazır derleme değiştir geçiş motor özel yol klasör dizin llama-server yürütülebilir ikili dosya backend inference engine",
},
modelMemory: {
- title: "Model belleği",
+ title: "Model yükleme",
keepResident: "Modeli GPU belleğinde tut",
keepResidentDescription: "İstemler arasında VRAM'de kalır.",
keepResidentHint:
@@ -3159,6 +3159,8 @@ export const tr = {
},
chat: {
groups: {
+ replies: { title: "Yanıtlar" },
+ thinkingTools: { title: "Düşünme ve araçlar" },
conversations: {
title: "Sohbetler",
},
@@ -3258,11 +3260,20 @@ export const tr = {
"Her yanıtın ne zaman tamamlandığını yanıtın altında gösterin.",
showReplyTimesKeywords:
"zaman damgası zaman saat tarih ne zaman bitti geldi yanıt cevap mesaj timestamp",
- showGenerationDetails: "Oluşturma ayrıntıları",
+ showGenerationDetails: "Ayrıntılı hız ve yükleme bilgisi",
showGenerationDetailsDescription:
- "Canlı istem ve yanıt hızlarını token sayılarıyla birlikte ve bir modelin katmanlarının nereye yüklendiğini gösterin.",
+ "Canlı göstergelere istem okuma hızını, token sayılarını ve taslak kabulünü ekler ve modelin katmanlarının nereye yüklendiğini gösterir.",
showGenerationDetailsKeywords:
"ayrıntılı geliştirici istatistik hız tok/s prefill decode katmanlar aktarım offload vram uzmanlar experts",
+ generationStatsUpdates: "Hız ve bağlam istatistikleri",
+ generationStatsUpdatesDescription:
+ "Gerçek zamanlı, bir yanıt akarken saniye başına token ve bağlam kullanımını sayar. Bitince, her yanıt tamamlandığında bunları günceller.",
+ generationStatsUpdatesHint:
+ "Gerçek zamanlı, yanıt yazılırken yanıtın altındaki saniye başına token göstergesini ve sohbetin üstündeki bağlam çubuğunu saniyede birkaç kez yeniler. Bitince, her yanıt tamamlandığında bunları bir kez günceller. Gerçek zamanlı mod, özellikle eski veya yoğun makinelerde takılmaya ya da daha düşük performansa yol açabilir.",
+ generationStatsUpdatesSettled: "Bitince",
+ generationStatsUpdatesRealtime: "Gerçek zamanlı",
+ generationStatsUpdatesKeywords:
+ "gerçek zamanlı canlı güncelle hız bağlam istatistik realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "Yanıtlarda Yanıtı düzenle",
inlineEditResponseDescription:
"Yanıtı düzenle seçeneğini Diğer menüsü yerine her yanıtta tutun.",
diff --git a/studio/frontend/src/i18n/locales/uk.ts b/studio/frontend/src/i18n/locales/uk.ts
index cb61d60ae0f..84279df76f7 100644
--- a/studio/frontend/src/i18n/locales/uk.ts
+++ b/studio/frontend/src/i18n/locales/uk.ts
@@ -2822,7 +2822,7 @@ export const uk = {
unavailable: "недоступно",
detecting: "Пошук GPU...",
unreadable: "Не вдалося прочитати дані про обладнання цього сервера.",
- noGpu: "Видимих GPU не виявлено. Вище показано лише ресурси CPU.",
+ noGpu: "Окремий GPU не виявлено.",
noUsableGpu: "На цьому комп’ютері немає GPU, придатного для PyTorch.",
mismatchCpuBuild:
"PyTorch — це збірка лише для CPU ({version}), тому наведені нижче GPU не можна використовувати. Відновіть інсталяцію, щоб повернути підтримку GPU.",
@@ -2911,7 +2911,7 @@ export const uk = {
"бекенд інференс прискорювач готова збірка перемкнути рушій власний шлях папка каталог виконуваний файл llama.cpp backend gguf inference cuda rocm hip vulkan metal cpu gpu accelerator prebuilt switch engine custom path folder directory llama-server executable binary",
},
modelMemory: {
- title: "Пам’ять моделей",
+ title: "Завантаження моделей",
keepResident: "Тримати модель у пам’яті GPU",
keepResidentDescription: "Залишатися у VRAM між запитами.",
keepResidentHint:
@@ -3187,6 +3187,8 @@ export const uk = {
},
chat: {
groups: {
+ replies: { title: "Відповіді" },
+ thinkingTools: { title: "Міркування та інструменти" },
conversations: {
title: "Розмови",
},
@@ -3286,11 +3288,20 @@ export const uk = {
"Показувати під кожною відповіддю, коли її завершено.",
showReplyTimesKeywords:
"позначка часу час годинник дата година коли завершено надійшло відповідь повідомлення timestamp",
- showGenerationDetails: "Відомості про генерування",
+ showGenerationDetails: "Докладна інформація про швидкість і завантаження",
showGenerationDetailsDescription:
- "Показувати поточну швидкість обробки запиту й відповіді з кількістю токенів, а також де завантажено шари моделі.",
+ "Додає до живих показників швидкість читання запиту, кількість токенів і прийняття чернеток, а також показує, куди завантажено шари моделі.",
showGenerationDetailsKeywords:
"докладно розробник статистика швидкість tok/s prefill decode шари вивантаження offload vram експерти",
+ generationStatsUpdates: "Статистика швидкості й контексту",
+ generationStatsUpdatesDescription:
+ "У реальному часі рахує токени за секунду й використання контексту, поки відповідь передається. Після завершення оновлює їх, коли кожна відповідь закінчена.",
+ generationStatsUpdatesHint:
+ "У реальному часі оновлює показник токенів за секунду під відповіддю та смугу контексту вгорі чату кілька разів на секунду, поки відповідь пишеться. Після завершення оновлює їх один раз, коли кожна відповідь закінчена. Режим реального часу може спричиняти підвисання або знижувати продуктивність, особливо на старих чи завантажених комп'ютерах.",
+ generationStatsUpdatesSettled: "Після завершення",
+ generationStatsUpdatesRealtime: "У реальному часі",
+ generationStatsUpdatesKeywords:
+ "реальний час наживо оновлення швидкість контекст статистика realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "Редагування відповіді на відповідях",
inlineEditResponseDescription:
"Показувати «Редагувати відповідь» на кожній відповіді, а не в меню «Більше».",
diff --git a/studio/frontend/src/i18n/locales/vi.ts b/studio/frontend/src/i18n/locales/vi.ts
index 887cca76c97..36a54e91b08 100644
--- a/studio/frontend/src/i18n/locales/vi.ts
+++ b/studio/frontend/src/i18n/locales/vi.ts
@@ -2801,7 +2801,7 @@ export const vi = {
detecting: "Đang kiểm tra GPU...",
unreadable: "Không thể đọc thông tin phần cứng của máy chủ này.",
noGpu:
- "Không phát hiện GPU nào. Tài nguyên chỉ dành cho CPU được hiển thị ở trên.",
+ "Không phát hiện GPU rời.",
noUsableGpu: "Không có GPU nào trên máy này mà PyTorch có thể sử dụng.",
mismatchCpuBuild:
"PyTorch là bản dựng chỉ dành cho CPU ({version}), nên không thể dùng các GPU bên dưới. Hãy sửa chữa bản cài đặt để khôi phục hỗ trợ GPU.",
@@ -2888,7 +2888,7 @@ export const vi = {
"llama.cpp backend gguf inference cuda rocm hip vulkan metal cpu gpu accelerator prebuilt switch engine custom path folder directory llama-server executable binary suy luận tăng tốc chuyển đổi công cụ tùy chỉnh đường dẫn thư mục tệp thực thi",
},
modelMemory: {
- title: "Bộ nhớ mô hình",
+ title: "Tải mô hình",
keepResident: "Giữ mô hình trong bộ nhớ GPU",
keepResidentDescription: "Giữ trong VRAM giữa các lời nhắc.",
keepResidentHint:
@@ -3160,6 +3160,8 @@ export const vi = {
},
chat: {
groups: {
+ replies: { title: "Phản hồi" },
+ thinkingTools: { title: "Suy nghĩ và công cụ" },
conversations: {
title: "Cuộc trò chuyện",
},
@@ -3259,11 +3261,20 @@ export const vi = {
"Hiển thị thời điểm mỗi câu trả lời hoàn tất, bên dưới câu trả lời.",
showReplyTimesKeywords:
"timestamp time clock date hour when finished arrived reply response message dấu thời gian giờ ngày hoàn tất trả lời tin nhắn",
- showGenerationDetails: "Chi tiết quá trình tạo",
+ showGenerationDetails: "Thông tin chi tiết về tốc độ và tải",
showGenerationDetailsDescription:
- "Hiển thị trực tiếp tốc độ xử lý lời nhắc và tốc độ trả lời kèm số token, cùng vị trí nạp các lớp của mô hình.",
+ "Thêm tốc độ đọc lời nhắc, số token và tỷ lệ chấp nhận bản nháp vào chỉ số trực tiếp, và cho biết các lớp của mô hình được tải ở đâu.",
showGenerationDetailsKeywords:
"verbose developer stats speed tok/s prefill decode layers offload vram experts chi tiết nhà phát triển thống kê tốc độ lớp",
+ generationStatsUpdates: "Thống kê tốc độ và ngữ cảnh",
+ generationStatsUpdatesDescription:
+ "Thời gian thực đếm token mỗi giây và mức dùng ngữ cảnh trong khi phản hồi đang truyền. Khi xong cập nhật chúng khi mỗi phản hồi kết thúc.",
+ generationStatsUpdatesHint:
+ "Thời gian thực làm mới chỉ số token mỗi giây dưới phản hồi và thanh ngữ cảnh ở đầu cuộc trò chuyện vài lần mỗi giây trong khi phản hồi đang được viết. Khi xong cập nhật chúng một lần, sau khi mỗi phản hồi kết thúc. Thời gian thực có thể gây giật hoặc giảm hiệu năng, nhất là trên máy cũ hoặc đang bận.",
+ generationStatsUpdatesSettled: "Khi xong",
+ generationStatsUpdatesRealtime: "Thời gian thực",
+ generationStatsUpdatesKeywords:
+ "thời gian thực trực tiếp cập nhật tốc độ ngữ cảnh thống kê realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "Nút Sửa câu trả lời trên câu trả lời",
inlineEditResponseDescription:
"Giữ nút Sửa câu trả lời trên mọi câu trả lời, thay vì trong menu Thêm.",
diff --git a/studio/frontend/src/i18n/locales/zh-CN.ts b/studio/frontend/src/i18n/locales/zh-CN.ts
index 6652842c3e5..ef70882a430 100644
--- a/studio/frontend/src/i18n/locales/zh-CN.ts
+++ b/studio/frontend/src/i18n/locales/zh-CN.ts
@@ -2567,7 +2567,7 @@ export const zhCN = {
unavailable: "不可用",
detecting: "正在检查 GPU...",
unreadable: "无法读取此服务器的硬件信息。",
- noGpu: "未检测到可见 GPU。上方仅显示 CPU 资源。",
+ noGpu: "未检测到独立 GPU。",
noUsableGpu: "本机没有 PyTorch 可用的 GPU。",
mismatchCpuBuild: "PyTorch 是仅 CPU 版本({version}),因此无法使用下方的 GPU。修复安装即可恢复 GPU 支持。",
mismatchUnavailable: "PyTorch({version})无法初始化下方的 GPU,因此无法使用。请检查显卡驱动,或修复安装。",
@@ -2641,7 +2641,7 @@ export const zhCN = {
"llama.cpp backend gguf 推理 cuda rocm hip vulkan metal cpu gpu 加速器 prebuilt 切换 引擎",
},
modelMemory: {
- title: "模型内存",
+ title: "模型加载",
keepResident: "将模型保留在显存中",
keepResidentDescription: "在两次提问之间保持驻留显存。",
keepResidentHint: "模型保持加载期间,不把权重交还给系统内存。关闭空闲自动卸载;当权重确实位于主机内存中时(统一内存,或部分卸载到 GPU),还会传入 --mlock,使操作系统无法将其换出、并在你下次提问时重新上传。",
@@ -2888,6 +2888,8 @@ export const zhCN = {
},
chat: {
groups: {
+ replies: { title: "回复" },
+ thinkingTools: { title: "思考与工具" },
conversations: { title: "对话" },
files: { title: "文件与粘贴" },
display: { title: "显示" },
@@ -2968,11 +2970,20 @@ export const zhCN = {
showReplyTimesDescription: "在每条回复下方显示其完成的时间。",
showReplyTimesKeywords:
"时间戳 时间 时刻 日期 完成 回复 消息 timestamp time clock date reply",
- showGenerationDetails: "生成详情",
+ showGenerationDetails: "详细速度与加载信息",
showGenerationDetailsDescription:
- "实时显示提示词和回复速度及 token 数,并显示模型各层的加载位置。",
+ "在实时读数中加入提示词读取速度、token 数和草稿采纳率,并显示模型各层加载在哪里。",
showGenerationDetailsKeywords:
"详情 速度 统计 层 卸载 verbose developer stats speed tok/s prefill decode layers offload vram experts",
+ generationStatsUpdates: "速度与上下文统计",
+ generationStatsUpdatesDescription:
+ "实时模式在回复流式输出时统计每秒 token 数和上下文用量。完成后模式在每条回复结束时更新。",
+ generationStatsUpdatesHint:
+ "实时模式会在回复生成期间,每秒数次刷新回复下方的每秒 token 数读数和聊天顶部的上下文条。完成后模式会在每条回复结束后更新一次。实时模式可能导致卡顿或性能下降,在较旧或繁忙的机器上尤其明显。",
+ generationStatsUpdatesSettled: "完成后",
+ generationStatsUpdatesRealtime: "实时",
+ generationStatsUpdatesKeywords:
+ "实时 动态 更新 速度 上下文 统计 realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "在回复上显示编辑回复",
inlineEditResponseDescription: "在每条回复上显示编辑回复,而不是放在“更多”菜单中。",
turnNavigation: "轮次导航",
diff --git a/studio/frontend/src/i18n/locales/zh-TW.ts b/studio/frontend/src/i18n/locales/zh-TW.ts
index 271eb5da8ff..9bbc77f5bbd 100644
--- a/studio/frontend/src/i18n/locales/zh-TW.ts
+++ b/studio/frontend/src/i18n/locales/zh-TW.ts
@@ -2672,7 +2672,7 @@ export const zhTW = {
unavailable: "無法使用",
detecting: "正在檢查 GPU...",
unreadable: "無法讀取此伺服器的硬體資訊。",
- noGpu: "未偵測到可見的 GPU。上方顯示的是僅 CPU 的資源。",
+ noGpu: "未偵測到獨立 GPU。",
noUsableGpu: "這台電腦上沒有 PyTorch 可使用的 GPU。",
mismatchCpuBuild:
"PyTorch 是僅支援 CPU 的版本({version}),因此無法使用下列 GPU。請修復安裝以恢復 GPU 支援。",
@@ -2752,7 +2752,7 @@ export const zhTW = {
"llama.cpp 後端 推論 引擎 加速器 預先建置 切換 自訂 路徑 資料夾 目錄 執行檔 backend gguf inference cuda rocm hip vulkan metal cpu gpu accelerator prebuilt switch engine custom path folder directory llama-server executable binary",
},
modelMemory: {
- title: "模型記憶體",
+ title: "模型載入",
keepResident: "將模型保留在 GPU 記憶體中",
keepResidentDescription: "在提示詞之間持續保留在 VRAM 中。",
keepResidentHint:
@@ -3013,6 +3013,8 @@ export const zhTW = {
},
chat: {
groups: {
+ replies: { title: "回覆" },
+ thinkingTools: { title: "思考與工具" },
conversations: {
title: "對話",
},
@@ -3106,11 +3108,20 @@ export const zhTW = {
showReplyTimesDescription: "在每則回覆下方顯示其完成時間。",
showReplyTimesKeywords:
"時間戳記 時間 時鐘 日期 小時 完成 抵達 回覆 回應 訊息 timestamp time clock date hour when finished arrived reply response message",
- showGenerationDetails: "生成詳細資料",
+ showGenerationDetails: "詳細速度與載入資訊",
showGenerationDetailsDescription:
- "即時顯示提示詞和回覆的速度及 token 數,以及模型各層的載入位置。",
+ "在即時讀數中加入提示詞讀取速度、token 數和草稿採用率,並顯示模型各層載入在哪裡。",
showGenerationDetailsKeywords:
"詳細 開發人員 統計 速度 預填 解碼 層 卸載 專家 verbose developer stats speed tok/s prefill decode layers offload vram experts",
+ generationStatsUpdates: "速度與上下文統計",
+ generationStatsUpdatesDescription:
+ "即時模式在回覆串流輸出時統計每秒 token 數和上下文用量。完成後模式在每則回覆結束時更新。",
+ generationStatsUpdatesHint:
+ "即時模式會在回覆產生期間,每秒數次重新整理回覆下方的每秒 token 數讀數和聊天頂端的上下文列。完成後模式會在每則回覆結束後更新一次。即時模式可能造成卡頓或效能下降,在較舊或忙碌的電腦上尤其明顯。",
+ generationStatsUpdatesSettled: "完成後",
+ generationStatsUpdatesRealtime: "即時",
+ generationStatsUpdatesKeywords:
+ "即時 動態 更新 速度 上下文 統計 realtime live delayed settled update refresh streaming tok/s tokens per second speed context usage meter stats",
inlineEditResponse: "在回應上顯示「編輯回應」",
inlineEditResponseDescription:
"在每則回應上保留「編輯回應」,而不是放在「更多」選單中。",
diff --git a/studio/frontend/tests/chat-menu-order.test.ts b/studio/frontend/tests/chat-menu-order.test.ts
index 656e9a75772..ac1b4591121 100644
--- a/studio/frontend/tests/chat-menu-order.test.ts
+++ b/studio/frontend/tests/chat-menu-order.test.ts
@@ -13,6 +13,7 @@ const {
PLUS_MENU_PINS_STORAGE_KEY,
keepSavedSlot,
normalizePlusMenuOrder,
+ plusMenuAtDefaults,
splitPlusMenuItems,
usePlusMenuPrefsStore,
} = await import("../src/features/chat/stores/plus-menu-prefs-store.ts");
@@ -122,3 +123,10 @@ test("moving another row keeps Deep research where it was in More, not at the to
assert.deepEqual(keepSavedSlot(["mcp"], ["mcp", "skills", "deepResearch"], "deepResearch"), ["mcp", "deepResearch"]);
assert.deepEqual(keepSavedSlot(["mcp"], ["mcp"], "deepResearch"), ["deepResearch", "mcp"]);
});
+
+test("Reset to default greys out only while the menu is at its defaults", () => {
+ const initial = usePlusMenuPrefsStore.getInitialState();
+ assert.equal(plusMenuAtDefaults(initial.pins, initial.order), true);
+ assert.equal(plusMenuAtDefaults({ ...initial.pins, mcp: !initial.pins.mcp }, initial.order), false);
+ assert.equal(plusMenuAtDefaults(initial.pins, [...initial.order].reverse()), false);
+});
diff --git a/studio/frontend/tests/chat-settings-sections.test.ts b/studio/frontend/tests/chat-settings-sections.test.ts
new file mode 100644
index 00000000000..f1638ab954d
--- /dev/null
+++ b/studio/frontend/tests/chat-settings-sections.test.ts
@@ -0,0 +1,216 @@
+// SPDX-License-Identifier: AGPL-3.0-only
+// Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+// Settings > Chat: Chat defaults holds the everyday rows; the rest is split so no section runs long.
+
+import assert from "node:assert/strict";
+import test from "node:test";
+
+import { readSrc } from "./helpers/kit.ts";
+
+const tab = readSrc("features/settings/tabs/chat-tab.tsx");
+
+function section(titleKey: string): string {
+ const start = tab.indexOf(`<SettingsSection title={t("${titleKey}")}>`);
+ assert.ok(start !== -1, `${titleKey} section exists`);
+ return tab.slice(start, tab.indexOf("</SettingsSection>", start));
+}
+
+const labels = (body: string): string[] =>
+ [...body.matchAll(/^\s+label=\{t\("([^"]+)"\)\}/gm)].map((match) => match[1]);
+
+const sectionOrder = (keys: string[]): number[] =>
+ keys.map((key) => tab.indexOf(`<SettingsSection title={t("${key}")}>`));
+
+const composer = readSrc("features/settings/components/composer-settings.tsx");
+const composerBody = (name: string): string => {
+ const start = composer.indexOf(`export function ${name}(`);
+ assert.ok(start !== -1, `${name} exists`);
+ const next = composer.indexOf("\nexport function ", start + 1);
+ return composer.slice(start, next === -1 ? undefined : next);
+};
+const composerLabels = labels;
+
+test("Chat defaults keeps the everyday rows: send and follow-up, scrolling, pastes, reply stats", () => {
+ const defaults = section("settings.general.chatDefaults");
+ assert.ok(defaults.indexOf("<ComposerSettings embedded={true} />") !== -1);
+ assert.deepEqual(labels(defaults), [
+ "settings.chat.autoScroll",
+ "settings.chat.pastedTextThreshold",
+ ]);
+ assert.ok(
+ defaults.indexOf("<ComposerSettings embedded={true} />") <
+ defaults.indexOf('label={t("settings.chat.autoScroll")}'),
+ "send shortcut and follow-up stay on top",
+ );
+ assert.match(defaults, /<GenerationStatsUpdatesRow \/>\s+$/);
+ assert.doesNotMatch(defaults, /isOwner/, "the owner limits sit in Replies");
+ assert.deepEqual(composerLabels(composerBody("ComposerSettings")), [
+ "composerSettings.sendShortcut",
+ "composerSettings.followUp",
+ ]);
+});
+
+test("Plain text composer, Rich text paste and Context window usage sit in Display; no Message box section", () => {
+ assert.equal(tab.indexOf('t("settings.chat.groups.composer.title")'), -1);
+ assert.doesNotMatch(
+ readSrc("features/settings/settings-search.ts"),
+ /"settings\.chat\.groups\.composer\.title"/,
+ );
+ assert.deepEqual(composerLabels(composerBody("ComposerFormatRows")), [
+ "composerSettings.plainText",
+ "composerSettings.richPaste",
+ ]);
+ assert.match(
+ composerBody("ContextWindowUsageRow"),
+ /<SettingsRow label=\{t\("composerSettings\.showContext"\)\}>/,
+ );
+ const display = section("settings.chat.groups.display.title");
+ assert.match(display, /<ContextWindowUsageRow \/>/);
+ assert.match(
+ display,
+ /<ComposerFormatRows \/>\s+$/,
+ "the two message-box rows close Display",
+ );
+ assert.equal(
+ tab.match(/<ContextWindowUsageRow \/>|<ComposerFormatRows \/>/g)?.length,
+ 2,
+ );
+});
+
+test("Conversations keeps per-chat setup and compaction", () => {
+ assert.deepEqual(
+ labels(section("settings.chat.groups.conversations.title")),
+ [
+ "settings.chat.rememberParamsPerModel",
+ "settings.chat.currentDate.label",
+ "settings.chat.projectAttachments",
+ "settings.general.autoTitleNewChats",
+ "settings.chat.autoCompact",
+ ],
+ );
+});
+
+test("Replies holds background runs and notifications, owner limits last", () => {
+ const replies = section("settings.chat.groups.replies.title");
+ assert.deepEqual(labels(replies), [
+ "settings.chat.keepRunningWhenTabClosed",
+ "settings.chat.replyNotification",
+ ]);
+ assert.match(
+ replies,
+ /\{isOwner \? <ToolResultLimitRow \/> : null\}\s+\{isOwner \? <FirstTokenTimeoutRow \/> : null\}\s+$/,
+ );
+});
+
+test("Thinking and tools groups the visibility rows out of Display", () => {
+ assert.deepEqual(
+ labels(section("settings.chat.groups.thinkingTools.title")),
+ [
+ "settings.chat.thinking.visibility",
+ "settings.chat.tools.visibility",
+ "settings.chat.tools.foldIntoThinking",
+ "settings.chat.webSearch.images",
+ ],
+ );
+ assert.deepEqual(labels(section("settings.chat.groups.display.title")), [
+ "settings.chat.showLinkPreviews",
+ "settings.chat.showResponseModel",
+ "settings.chat.showReplyTimes",
+ "settings.chat.showGenerationDetails",
+ "settings.chat.inlineEditResponse",
+ "settings.chat.turnNavigation",
+ ]);
+});
+
+test("sections run defaults, memory, conversations, replies, thinking and tools, display", () => {
+ const [defaults, conversations, replies, thinkingTools, display, models] =
+ sectionOrder([
+ "settings.general.chatDefaults",
+ "settings.chat.groups.conversations.title",
+ "settings.chat.groups.replies.title",
+ "settings.chat.groups.thinkingTools.title",
+ "settings.chat.groups.display.title",
+ "settings.chat.modelSelection.title",
+ ]);
+ const memory = tab.indexOf("<ChatMemorySection />");
+ assert.ok(
+ defaults < memory &&
+ memory < conversations &&
+ conversations < replies &&
+ replies < thinkingTools &&
+ thinkingTools < display &&
+ display < models,
+ );
+ const search = readSrc("features/settings/settings-search.ts");
+ for (const key of ["replies", "thinkingTools"]) {
+ assert.match(
+ search,
+ new RegExp(`"settings\\.chat\\.groups\\.${key}\\.title"`),
+ );
+ }
+});
+
+test("section-level resets use the Appearance reset: an outline small button, greyed out with nothing to undo", () => {
+ const appearance = readSrc(
+ "features/settings/components/appearance-custom-controls.tsx",
+ );
+ const resetAll = appearance.slice(
+ appearance.indexOf("export function ResetCustomizationButton"),
+ );
+ assert.match(
+ resetAll,
+ /variant="outline"\s+size="sm"\s+disabled=\{pristine\}/,
+ );
+
+ const menu = tab.slice(tab.indexOf('{t("settings.chat.groups.menu.title")}'));
+ assert.match(
+ menu,
+ /<Button\s+type="button"\s+variant="outline"\s+size="sm"\s+disabled=\{plusMenuPristine\}\s+onClick=\{resetPlusMenu\}\s*>\s*\{t\("settings\.chat\.menu\.reset"\)\}/,
+ );
+ assert.match(tab, /plusMenuAtDefaults\(state\.pins, state\.order\)/);
+
+ const shortcuts = readSrc(
+ "features/settings/tabs/keyboard-shortcuts-tab.tsx",
+ );
+ // Appearance's footer: a full-width rule above, the button on the right, after the last section.
+ const RULED_FOOTER =
+ /className="(?:mt-auto )?flex justify-end border-t border-border\/60 pt-4">\s+<(?:Button|ResetCustomizationButton)\b/;
+ assert.match(
+ readSrc("features/settings/tabs/appearance-tab.tsx"),
+ RULED_FOOTER,
+ );
+ assert.match(tab, RULED_FOOTER);
+ assert.match(shortcuts, RULED_FOOTER);
+ assert.ok(
+ tab.lastIndexOf("</SettingsSection>") <
+ tab.indexOf('t("settings.chat.menu.reset")'),
+ "the Chat menu reset sits below the last section, like Appearance's",
+ );
+ const footer = shortcuts.slice(
+ shortcuts.lastIndexOf(
+ "<Button",
+ shortcuts.indexOf('t("settings.keyboardShortcuts.resetAll")'),
+ ),
+ );
+ assert.match(
+ footer,
+ /^<Button\s+type="button"\s+variant="outline"\s+size="sm"\s+disabled=/,
+ );
+ assert.doesNotMatch(shortcuts, /variant="dark"/);
+
+ for (const [file, key] of [
+ ["features/settings/tabs/browser-tab.tsx", "browser.downloadLocationReset"],
+ [
+ "features/settings/tabs/resources-tab.tsx",
+ "settings.resources.storage.resetAction",
+ ],
+ ] as const) {
+ const source = readSrc(file);
+ const button = source.slice(
+ source.lastIndexOf("<Button", source.indexOf(`t("${key}")`)),
+ source.indexOf(`t("${key}")`),
+ );
+ assert.match(button, /variant="outline"/, `${key} is an outline button`);
+ }
+});
diff --git a/studio/frontend/tests/generation-details.test.ts b/studio/frontend/tests/generation-details.test.ts
index 364a66f992a..79713b7ff26 100644
--- a/studio/frontend/tests/generation-details.test.ts
+++ b/studio/frontend/tests/generation-details.test.ts
@@ -413,7 +413,7 @@ test("the live line follows the branch picker, so its growing text never moves t
test("the live line gates on the preference before any assistant-ui selector", () => {
const line = readSrc("components/assistant-ui/live-generation-line.tsx");
const gate = line.slice(line.indexOf("export const LiveGenerationLine"), line.indexOf("const RunningGate"));
- assert.match(gate, /return enabled \? <RunningGate \/> : null;/);
+ assert.match(gate, /return details \|\| realtime \? <RunningGate realtime=\{realtime\} \/> : null;/);
assert.doesNotMatch(gate, /useAuiState/);
});
diff --git a/studio/frontend/tests/gpu-torch-mismatch.test.ts b/studio/frontend/tests/gpu-torch-mismatch.test.ts
index 638c41e23fb..ef9789e6fd1 100644
--- a/studio/frontend/tests/gpu-torch-mismatch.test.ts
+++ b/studio/frontend/tests/gpu-torch-mismatch.test.ts
@@ -310,7 +310,9 @@ test("every string the banner reaches for exists", () => {
assert.match(gpu.mismatchCpuBuild, /\{version\}/);
assert.match(gpu.mismatchUnavailable, /\{version\}/);
assert.equal(t(NO_GPU), NO_GPU);
- assert.match(gpu.noGpu, /No visible GPU detected/);
+ assert.match(gpu.noGpu, /No dedicated GPU detected/);
+ // Unified-memory hosts share the RAM shown above with a built-in GPU, so it is not "CPU-only".
+ assert.doesNotMatch(gpu.noGpu, /CPU-only/);
});
// start_managed_repair rejects external backends only after the shell already swapped screens.
diff --git a/studio/frontend/tests/realtime-generation-stats.test.ts b/studio/frontend/tests/realtime-generation-stats.test.ts
new file mode 100644
index 00000000000..76ecb2fdd6f
--- /dev/null
+++ b/studio/frontend/tests/realtime-generation-stats.test.ts
@@ -0,0 +1,371 @@
+// SPDX-License-Identifier: AGPL-3.0-only
+// Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+// The Speed and context stats setting: settled (default) vs realtime, and the throttle realtime rides on.
+
+import assert from "node:assert/strict";
+import test from "node:test";
+
+import type { LiveGenerationProgress } from "../src/features/chat/stores/chat-runtime-store.ts";
+import {
+ installLocalStorageFake,
+ readSrc,
+ registerBundlerResolver,
+} from "./helpers/kit.ts";
+
+registerBundlerResolver();
+installLocalStorageFake();
+const {
+ createTrailingThrottle,
+ GENERATION_STATS_UPDATES,
+ LIVE_STATS_INTERVAL_MS,
+ liveContextUsage,
+ normaliseGenerationStatsUpdates,
+} = await import("../src/features/chat/lib/live-generation-stats.ts");
+const { useChatPreferencesStore } = await import(
+ "../src/features/chat/stores/chat-preferences-store.ts"
+);
+
+const LOCALES = [
+ "ar",
+ "cs",
+ "de",
+ "en",
+ "es-419",
+ "es",
+ "fa",
+ "fr",
+ "he",
+ "hi",
+ "id",
+ "it",
+ "ja",
+ "ko",
+ "nl",
+ "pl",
+ "pt-br",
+ "ru",
+ "sv",
+ "th",
+ "tr",
+ "uk",
+ "vi",
+ "zh-CN",
+ "zh-TW",
+];
+
+function fakeClock() {
+ let now = 0;
+ let timers: { at: number; run: () => void; id: number }[] = [];
+ let nextId = 1;
+ return {
+ clock: {
+ now: () => now,
+ setTimeout: (run: () => void, ms: number) => {
+ const id = nextId++;
+ timers.push({ at: now + ms, run, id });
+ return id;
+ },
+ clearTimeout: (id: unknown) => {
+ timers = timers.filter((timer) => timer.id !== id);
+ },
+ },
+ advance(ms: number) {
+ now += ms;
+ for (const timer of timers.filter((t) => t.at <= now)) {
+ timers = timers.filter((t) => t !== timer);
+ timer.run();
+ }
+ },
+ pending: () => timers.length,
+ };
+}
+
+const live = (
+ fields: Partial<LiveGenerationProgress>,
+): LiveGenerationProgress => ({
+ prefill: null,
+ prefillShown: false,
+ decode: null,
+ ...fields,
+});
+const prefill = (
+ processed: number,
+ total: number | null,
+ cached: number | null = null,
+) => ({
+ processed,
+ total,
+ cached,
+ tokPerS: null,
+ elapsedMs: null,
+ fit: null,
+});
+const decode = (tokens: number) => ({
+ tokens,
+ tokPerS: 42,
+ draftN: null,
+ draftAccepted: null,
+});
+
+test("the setting defaults to settled, and only an exact realtime survives a reload", () => {
+ assert.deepEqual(GENERATION_STATS_UPDATES, ["settled", "realtime"]);
+ assert.equal(
+ useChatPreferencesStore.getInitialState().generationStatsUpdates,
+ "settled",
+ );
+ const merge = useChatPreferencesStore.persist.getOptions().merge;
+ assert.ok(merge);
+ const current = useChatPreferencesStore.getInitialState();
+ assert.equal(
+ merge({ generationStatsUpdates: "realtime" }, current)
+ .generationStatsUpdates,
+ "realtime",
+ );
+ assert.equal(
+ merge({ generationStatsUpdates: "settled" }, current)
+ .generationStatsUpdates,
+ "settled",
+ );
+ // Records written before the setting, or by a future value, fall back to today's behaviour.
+ assert.equal(merge({}, current).generationStatsUpdates, "settled");
+ assert.equal(
+ merge({ generationStatsUpdates: "fast" }, current).generationStatsUpdates,
+ "settled",
+ );
+ assert.equal(normaliseGenerationStatsUpdates(undefined), "settled");
+
+ useChatPreferencesStore.getState().setGenerationStatsUpdates("realtime");
+ assert.equal(
+ useChatPreferencesStore.getState().generationStatsUpdates,
+ "realtime",
+ );
+ useChatPreferencesStore.getState().setGenerationStatsUpdates("settled");
+ assert.equal(
+ useChatPreferencesStore.getState().generationStatsUpdates,
+ "settled",
+ );
+});
+
+test("realtime renders at most once per interval however fast frames arrive, and the last frame lands", () => {
+ assert.ok(LIVE_STATS_INTERVAL_MS >= 100 && LIVE_STATS_INTERVAL_MS <= 250);
+ const { clock, advance, pending } = fakeClock();
+ let runs = 0;
+ const throttle = createTrailingThrottle(200, () => runs++, clock);
+ throttle.schedule();
+ assert.equal(runs, 1, "the first frame shows at once");
+ // One frame per token at 1,000 tok/s for 199 ms: one trailing render, not 199.
+ for (let i = 0; i < 199; i++) {
+ advance(1);
+ throttle.schedule();
+ }
+ assert.equal(runs, 1);
+ advance(1);
+ assert.equal(runs, 2, "the burst lands once the interval is up");
+ assert.equal(pending(), 0);
+ advance(500);
+ throttle.schedule();
+ assert.equal(runs, 3);
+ throttle.schedule();
+ throttle.cancel();
+ advance(1000);
+ assert.equal(runs, 3, "an unmounted meter schedules no render");
+});
+
+test("realtime context counts the prompt as it is read, then the prompt plus the reply so far", () => {
+ assert.equal(liveContextUsage(undefined), null);
+ assert.deepEqual(
+ liveContextUsage(live({ prefill: prefill(2048, 5316, 1024) })),
+ {
+ used: 2048,
+ promptTokens: 2048,
+ completionTokens: 0,
+ cached: 1024,
+ },
+ );
+ assert.deepEqual(
+ liveContextUsage(
+ live({ prefill: prefill(5316, 5316), decode: decode(300) }),
+ ),
+ {
+ used: 5616,
+ promptTokens: 5316,
+ completionTokens: 300,
+ cached: 0,
+ },
+ );
+ // No prompt total, a compaction in progress, or the pass already settled: keep the settled count.
+ assert.equal(
+ liveContextUsage(
+ live({ prefill: prefill(10, null), decode: decode(3) }),
+ ),
+ null,
+ );
+ assert.equal(liveContextUsage(live({ decode: decode(3) })), null);
+ assert.equal(
+ liveContextUsage(
+ live({ prefill: prefill(10, 20), compacting: true }),
+ ),
+ null,
+ );
+ assert.equal(
+ liveContextUsage(
+ live({ prefill: prefill(20, 20), decode: decode(3), settled: true }),
+ ),
+ null,
+ );
+});
+
+test("settled mode keeps today's paths: no live subscription in the meter, the line only with details", () => {
+ const bar = readSrc("features/chat/components/live-context-usage-bar.tsx");
+ const gate = bar.slice(
+ bar.indexOf("export const LiveContextUsageBar"),
+ bar.indexOf("const RealtimeContextUsageBar"),
+ );
+ assert.match(
+ gate,
+ /realtime \? \(\s*<RealtimeContextUsageBar [^>]*\/>\s*\) : \(\s*<ContextUsageBar \{\.\.\.props\} \/>/,
+ );
+ assert.doesNotMatch(
+ gate,
+ /useThrottledLiveGeneration|liveGenerationByThreadId|useChatRuntimeStore/,
+ );
+ const realtime = bar.slice(bar.indexOf("const RealtimeContextUsageBar"));
+ assert.match(realtime, /useThrottledLiveGeneration\(liveThreadKey\)/);
+ assert.doesNotMatch(
+ realtime,
+ /liveGenerationByThreadId/,
+ "realtime reads live figures only through the throttle",
+ );
+
+ const page = readSrc("features/chat/chat-page.tsx");
+ assert.match(
+ page,
+ /<LiveContextUsageBar\s+(?:\/\/[^\n]*\s+)?liveThreadKey=\{activeThreadId \?\? "__default"\}/,
+ );
+ assert.doesNotMatch(page, /<ContextUsageBar\b/);
+
+ const line = readSrc("components/assistant-ui/live-generation-line.tsx");
+ assert.match(
+ line,
+ /return realtime \? <ThrottledDecodeFigures \/> : <LiveDecodeFigures \/>;/,
+ );
+ const direct = line.slice(
+ line.indexOf("const LiveDecodeFigures"),
+ line.indexOf("const ThrottledDecodeFigures"),
+ );
+ assert.match(
+ direct,
+ /useChatRuntimeStore\(/,
+ "settled keeps the direct per-frame subscription",
+ );
+ const throttled = line.slice(
+ line.indexOf("const ThrottledDecodeFigures"),
+ line.indexOf("const DecodeFigures"),
+ );
+ assert.match(throttled, /useThrottledLiveGeneration\(/);
+ assert.doesNotMatch(throttled, /useChatRuntimeStore\(/);
+
+ const hook = readSrc("features/chat/hooks/use-throttled-live-generation.ts");
+ assert.match(hook, /createTrailingThrottle\(LIVE_STATS_INTERVAL_MS/);
+ assert.match(hook, /useChatRuntimeStore\.subscribe\(/);
+ assert.match(hook, /throttle\.cancel\(\)/);
+});
+
+test("the toggle row sits in Chat defaults only, is searchable, and is translated in every locale", async () => {
+ const row = readSrc(
+ "features/settings/components/generation-stats-updates-row.tsx",
+ );
+ assert.match(row, /GENERATION_STATS_UPDATES\.map\(\(mode\) =>/);
+ assert.match(row, /onClick=\{\(\) => setGenerationStatsUpdates\(mode\)\}/);
+ const chat = readSrc("features/settings/tabs/chat-tab.tsx");
+ const defaults = chat.slice(
+ chat.indexOf(
+ '<SettingsSection title={t("settings.general.chatDefaults")}>',
+ ),
+ chat.indexOf(
+ '<SettingsSection title={t("settings.chat.groups.conversations.title")}>',
+ ),
+ );
+ assert.match(defaults, /<GenerationStatsUpdatesRow \/>/);
+ assert.equal(
+ chat.match(/<GenerationStatsUpdatesRow \/>/g)?.length,
+ 1,
+ "only once on the Chat tab",
+ );
+ assert.doesNotMatch(
+ chat,
+ /generationStatsUpdates/,
+ "the Chat tab renders the shared row, not its own copy",
+ );
+ const system = readSrc("features/settings/tabs/resources-tab.tsx");
+ assert.doesNotMatch(system, /GenerationStatsUpdatesRow/, "not on System");
+ const search = readSrc("features/settings/settings-search.ts");
+ assert.match(
+ search,
+ /"settings\.chat\.generationStatsUpdates": "settings\.chat\.generationStatsUpdatesKeywords"/,
+ );
+ const index = (tab: string) =>
+ search.slice(
+ search.indexOf(` ${tab}: [`),
+ search.indexOf("],", search.indexOf(` ${tab}: [`)),
+ );
+ assert.match(index("chat"), /"settings\.chat\.generationStatsUpdates"/);
+ assert.doesNotMatch(
+ index("resources"),
+ /"settings\.chat\.generationStatsUpdates"/,
+ );
+
+ const keys = [
+ "generationStatsUpdates",
+ "generationStatsUpdatesDescription",
+ "generationStatsUpdatesSettled",
+ "generationStatsUpdatesRealtime",
+ "generationStatsUpdatesKeywords",
+ ];
+ for (const locale of LOCALES) {
+ const module = (await import(`../src/i18n/locales/${locale}.ts`)) as Record<
+ string,
+ unknown
+ >;
+ const catalog = Object.values(module)[0] as {
+ settings: { chat: Record<string, unknown> };
+ };
+ for (const key of keys) {
+ const value = catalog.settings.chat[key];
+ assert.ok(
+ typeof value === "string" && value.length > 0,
+ `${locale} settings.chat.${key}`,
+ );
+ }
+ }
+});
+
+test("the toggle's tooltip explains both modes and warns that Realtime may cost performance", async () => {
+ const row = readSrc(
+ "features/settings/components/generation-stats-updates-row.tsx",
+ );
+ assert.match(
+ row,
+ /hint=\{t\("settings\.chat\.generationStatsUpdatesHint"\)\}/,
+ );
+ const { en } = (await import("../src/i18n/locales/en.ts")) as {
+ en: { settings: { chat: { generationStatsUpdatesHint: string } } };
+ };
+ const hint = en.settings.chat.generationStatsUpdatesHint;
+ assert.match(hint, /lag/);
+ assert.match(hint, /performance/);
+ for (const locale of LOCALES) {
+ const module = (await import(`../src/i18n/locales/${locale}.ts`)) as Record<
+ string,
+ unknown
+ >;
+ const catalog = Object.values(module)[0] as {
+ settings: { chat: Record<string, unknown> };
+ };
+ assert.ok(
+ typeof catalog.settings.chat.generationStatsUpdatesHint === "string",
+ `${locale} hint`,
+ );
+ }
+});
diff --git a/studio/frontend/tests/realtime-meter-settled.test.ts b/studio/frontend/tests/realtime-meter-settled.test.ts
new file mode 100644
index 00000000000..0b7bdd5a569
--- /dev/null
+++ b/studio/frontend/tests/realtime-meter-settled.test.ts
@@ -0,0 +1,63 @@
+// SPDX-License-Identifier: AGPL-3.0-only
+// Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+// The realtime context meter steps aside once a pass's own context_usage lands. The mark lives on the
+// thread's live progress in the store, so a chat switch, a remount or a pass with no decode frame keeps it.
+
+import assert from "node:assert/strict";
+import { register } from "node:module";
+import test from "node:test";
+
+import { installLocalStorageFake, readSrc } from "./helpers/kit.ts";
+
+const { store: localStorageFake } = installLocalStorageFake();
+localStorageFake.set("unsloth_chat_settings_imported_to_studio_db", "true");
+register("./thread-sampling-resolver.mjs", import.meta.url);
+await import("./helpers/store-stubs/settings-http.ts");
+await import("./helpers/store-stubs/chat-history-storage.ts");
+
+const { useChatRuntimeStore } = await import("../src/features/chat/stores/chat-runtime-store.ts");
+const { liveContextUsage } = await import("../src/features/chat/lib/live-generation-stats.ts");
+
+const prefill = { processed: 5316, total: 5316, cached: 0 } as never;
+const decode = { tokens: 12 } as never;
+const live = (key: string) => useChatRuntimeStore.getState().liveGenerationByThreadId[key];
+
+test("a pass's own context_usage settles its live progress until the next pass starts", () => {
+ const store = useChatRuntimeStore.getState();
+ store.setLivePrefill("t1", prefill);
+ assert.ok(liveContextUsage(live("t1")), "prefill counts live");
+ // A short tool-call pass: its usage lands before any decode frame.
+ store.settleLiveGeneration("t1");
+ assert.equal(live("t1")?.settled, true);
+ assert.equal(liveContextUsage(live("t1")), null, "the exact settled count shows");
+ store.setLiveDecode("t1", decode);
+ assert.equal(live("t1")?.settled, true, "a trailing decode frame keeps it settled");
+ store.settleLiveGeneration("t1");
+ store.setLivePrefill("t1", prefill);
+ assert.equal(live("t1")?.settled, undefined);
+ assert.ok(liveContextUsage(live("t1")));
+ store.clearLiveGenerationForThread("t1");
+});
+
+test("settling is per thread and survives a remount", () => {
+ const store = useChatRuntimeStore.getState();
+ store.setLivePrefill("a", prefill);
+ store.setLivePrefill("b", prefill);
+ store.settleLiveGeneration("a");
+ assert.equal(live("a")?.settled, true);
+ assert.equal(live("b")?.settled, undefined, "the other chat keeps counting live");
+ store.clearLiveGenerationForThread("a");
+ const before = useChatRuntimeStore.getState().liveGenerationByThreadId;
+ store.settleLiveGeneration("a");
+ assert.equal(useChatRuntimeStore.getState().liveGenerationByThreadId, before);
+ store.clearLiveGenerationForThread("b");
+});
+
+test("the adapter settles the run's own thread when it publishes that pass's context_usage", () => {
+ const adapter = readSrc("features/chat/api/chat-adapter.ts");
+ assert.match(
+ adapter,
+ /const usageKey = liveThreadKey\(serverCancel\);[\s\S]{0,900}store\.settleLiveGeneration\(usageKey\);/,
+ );
+});