fix(tui): drop ttft from the assistant footer (#63759)

This commit is contained in:
Dax
2026-11-13 07:58:21 -05:00
committed by GitHub
Unverified
parent 191a69ad45
commit 507c939f41
4 changed files with 5 additions and 23 deletions
@@ -2030,7 +2030,6 @@ function AssistantFooter(props: { message: SessionMessageAssistant }) {
const turn = () => ({ index: ctx.messageIndex(props.message.id), legacy: ctx.legacyTurns() })
const duration = createMemo(() => SessionTurn.duration(messages(), props.message, turn()))
const tokensPerSecond = createMemo(() => SessionTurn.tokensPerSecond(messages(), props.message, turn()))
const timeToFirstToken = createMemo(() => SessionTurn.timeToFirstToken(messages(), props.message, turn()))
const interrupted = createMemo(() => props.message.error?.message === "Step interrupted")
const freeLimit = createMemo(() => props.message.error?.type === "provider.free-limit")
const limitNotice = useLimitNotice(() => props.message)
@@ -2065,9 +2064,6 @@ function AssistantFooter(props: { message: SessionMessageAssistant }) {
<Show when={config.data.session.tps && tokensPerSecond()}>
{(value) => <span style={{ fg: theme.text.muted }}> · {value().toFixed(1)} tok/s</span>}
</Show>
<Show when={config.data.session.tps && ctx.terminal.width >= 36 && timeToFirstToken()}>
{(value) => <span style={{ fg: theme.text.muted }}> · {Locale.duration(value())} ttft</span>}
</Show>
<Show when={interrupted()}>
<span style={{ fg: theme.text.muted }}> · interrupted</span>
</Show>
+5 -5
View File
@@ -795,7 +795,8 @@ test("keeps assistant footer metrics current after prepend, same-length refresh,
(frame) => frame.includes("Original answer") && frame.includes("40.0 tok/s"),
)
expect(initial).toContain("Current input")
expect(initial).toContain("4.0s \u00b7 40.0 tok/s \u00b7 500ms ttft")
expect(initial).toContain("4.0s \u00b7 40.0 tok/s")
expect(initial).not.toContain("ttft")
expect(initial).not.toContain("Prepended input")
setup.mockInput.pressKey("g", { ctrl: true })
@@ -810,7 +811,7 @@ test("keeps assistant footer metrics current after prepend, same-length refresh,
)
expect(prepended).toContain("Current input")
expect(prepended).toContain("Original answer")
expect(prepended).toContain("4.0s \u00b7 40.0 tok/s \u00b7 500ms ttft")
expect(prepended).toContain("4.0s \u00b7 40.0 tok/s")
// Refresh the latest page: length stays four, but the retained assistant moves from index three to one.
refresh = true
@@ -826,7 +827,7 @@ test("keeps assistant footer metrics current after prepend, same-length refresh,
)
expect(refreshed).toContain("Current input")
expect(refreshed).toContain("Original answer")
expect(refreshed).toContain("4.0s \u00b7 40.0 tok/s \u00b7 500ms ttft")
expect(refreshed).toContain("4.0s \u00b7 40.0 tok/s")
expect(refreshed).toContain("3.0s \u00b7 50.0 tok/s")
setup.events.emit({
@@ -841,7 +842,7 @@ test("keeps assistant footer metrics current after prepend, same-length refresh,
(frame) => frame.includes("Original answer") && !frame.includes("Later input") && !frame.includes("Later answer"),
)
expect(reverted).toContain("Current input")
expect(reverted).toContain("4.0s \u00b7 40.0 tok/s \u00b7 500ms ttft")
expect(reverted).toContain("4.0s \u00b7 40.0 tok/s")
expect(reverted).not.toContain("50.0 tok/s")
})
@@ -2844,4 +2845,3 @@ test.each([44, 100])(
},
15_000,
)
-4
View File
@@ -22,7 +22,6 @@ test("measures decode throughput from the first streamed token", () => {
// 65 tokens over the 5s the provider spent streaming them, not the 10s that include waiting for the first.
expect(SessionTurn.tokensPerSecond(messages, final)).toBe(13)
expect(SessionTurn.timeToFirstToken(messages, final)).toBe(2_000)
})
test("falls back to the request window for steps recorded without a first token", () => {
@@ -32,8 +31,6 @@ test("falls back to the request window for steps recorded without a first token"
// The old step counts its 4s request window, the new one its 3s decode window.
expect(SessionTurn.tokensPerSecond(messages, final)).toBe(65 / 7)
// Only the step that recorded a first token can report how long it waited for one.
expect(SessionTurn.timeToFirstToken(messages, final)).toBe(1_000)
})
test("measures request throughput including reasoning without tool time", () => {
@@ -42,7 +39,6 @@ test("measures request throughput including reasoning without tool time", () =>
const messages = [input("user-1", 1_000), first, final]
expect(SessionTurn.tokensPerSecond(messages, final)).toBe(6.5)
expect(SessionTurn.timeToFirstToken(messages, final)).toBeUndefined()
first.time.streamed = undefined
expect(SessionTurn.tokensPerSecond(messages, final)).toBeUndefined()
})
-10
View File
@@ -55,16 +55,6 @@ export function tokensPerSecond(messages: readonly Message[], message: Message,
return output / (total / 1_000)
}
/**
* Milliseconds from the dispatch of the turn's first token-producing step to its first streamed token, or undefined
* until a step records one.
*/
export function timeToFirstToken(messages: readonly Message[], message: Message, options: Options = {}) {
const step = turnSteps(messages, message, options).find((item) => item.time.firstToken !== undefined)
if (step?.time.firstToken === undefined) return
return Math.max(0, step.time.firstToken - step.time.created)
}
export function legacy(messages: readonly Message[]) {
return !messages.some((message) => message.type === "idle")
}