feat: model-on-demand lifecycle — retry on 503, live status pill, warming indicator

- whisper.ts: add getModelStatus(); fix submitJob() to retry on 503 using
  Retry-After header instead of throwing; optional onModelWaiting callback
  lets the pipeline surface model state to the UI during the wait
- pipeline.ts: pass onModelWaiting callback → emits model_warming SSE event
  so the job detail page can show 'Warming up model…' while waiting
- types.ts: add ModelStateTag union and ModelStatus interface
- api/model/status: GET route proxies whisper /model/status (falls back to
  {state:'unloaded'} if whisper unreachable)
- api/model/events: GET route relays whisper SSE stream to the browser;
  AbortController tied to request.signal cleans up on disconnect
- layout.svelte: status pill is now live — initial fetch + EventSource on
  /api/model/events; dot colour + label reflect real model state with a
  pulsing animation while loading or waiting_for_gpu
- jobs/[id]/+page.svelte: handle model_warming event type → show a yellow
  'Warming up model…' sub-label with spinner inside the progress card
- whisper.test.ts: update submitJob mocks to status:202 to match real API

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

src/lib/server/pipeline.ts

@@ -129,7 +129,9 @@ async function runJob(
 		emitProgress(jobId, { type: 'status', status: 'transcribing' });
 
 		const webhookUrl = `${WEBHOOK_BASE_URL}/api/webhook/${jobId}`;
-		const whisperJobId = await submitJob(wavPath, webhookUrl, language);
+		const whisperJobId = await submitJob(wavPath, webhookUrl, language, (state, retryAfterSecs) => {
+			emitProgress(jobId, { type: 'model_warming', state, retryAfterSecs });
+		});
 		updateJob({ id: jobId, whisperJobId });
 
 		// ── 5. Open SSE for live progress (non-blocking relay) ───────────────

src/lib/server/whisper.ts

@@ -1,17 +1,34 @@
-import { execFile } from 'child_process';
+import type { ModelStatus } from '$lib/types.js';
-import { promisify } from 'util';
-
-const execFileAsync = promisify(execFile);
 
 function whisperUrl() {
 	return process.env.WHISPER_URL ?? 'http://localhost:8080';
 }
 
-/** Submit an audio file to whisper-rtx2080. Returns the whisper job id. */
+const sleep = (ms: number) => new Promise((r) => setTimeout(r, ms));
+
+/** Get the current model state from whisper-rtx2080. */
+export async function getModelStatus(): Promise<ModelStatus> {
+	const { default: fetch } = await import('node-fetch');
+	const res = await fetch(`${whisperUrl()}/model/status`, {
+		signal: AbortSignal.timeout(5000)
+	});
+	if (!res.ok) throw new Error(`/model/status returned ${res.status}`);
+	return res.json() as Promise<ModelStatus>;
+}
+
+/**
+ * Submit an audio file to whisper-rtx2080. Returns the whisper job id.
+ *
+ * Handles 503 (model not ready) transparently: retries using the
+ * `Retry-After` header until the model loads or maxAttempts is exhausted.
+ * Calls `onModelWaiting` on each 503 so the caller can surface the wait to the user.
+ */
 export async function submitJob(
 	wavPath: string,
 	webhookUrl: string,
-	language?: string
+	language?: string,
+	onModelWaiting?: (state: string, retryAfterSecs: number) => void,
+	maxAttempts = 20
 ): Promise<string> {
 	const FormData = (await import('form-data')).default;
 	const { createReadStream } = await import('fs');
@@ -23,19 +40,35 @@ export async function submitJob(
 	form.append('webhook_url', webhookUrl);
 	if (language) form.append('language', language);
 
-	const res = await fetch(`${whisperUrl()}/jobs`, {
+	for (let attempt = 1; attempt <= maxAttempts; attempt++) {
-		method: 'POST',
+		const res = await fetch(`${whisperUrl()}/jobs`, {
-		body: form,
+			method: 'POST',
-		headers: form.getHeaders()
+			body: form,
-	});
+			headers: form.getHeaders()
+		});
+
+		if (res.status === 202) {
+			const json = (await res.json()) as { job_id: string };
+			return json.job_id;
+		}
+
+		if (res.status === 503) {
+			const body = (await res.json().catch(() => ({}))) as {
+				state?: string;
+				retry_after_secs?: number;
+			};
+			const state = body.state ?? 'unloaded';
+			const waitSecs = body.retry_after_secs ?? parseInt(res.headers.get('Retry-After') ?? '15');
+			onModelWaiting?.(state, waitSecs);
+			await sleep((waitSecs + 1) * 1000);
+			continue;
+		}
 
-	if (!res.ok) {
 		const text = await res.text();
 		throw new Error(`whisper /jobs returned ${res.status}: ${text}`);
 	}
 
-	const json = (await res.json()) as { job_id: string };
+	throw new Error(`Whisper model did not become ready after ${maxAttempts} attempts`);
-	return json.job_id;
 }
 
 /** Open an SSE stream from whisper and call onProgress/onDone callbacks. */

src/lib/types.ts

@@ -1,5 +1,17 @@
 export type AudioMode = 'auto' | 'standard' | 'aggressive' | 'none';
 
+export type ModelStateTag = 'unloaded' | 'loading' | 'waiting_for_gpu' | 'ready';
+
+export interface ModelStatus {
+	state: ModelStateTag;
+	loaded_at?: string;
+	vram_needed_mb?: number;
+	vram_free_mb?: number;
+	retry_in_secs?: number;
+	vram_used_mb?: number;
+	vram_total_mb?: number;
+}
+
 export type JobStatus = 'pending' | 'downloading' | 'preparing' | 'transcribing' | 'processing' | 'done' | 'failed' | 'cancelled';
 
 export interface Segment {

src/routes/+layout.svelte

@@ -1,9 +1,10 @@
 <script lang="ts">
 	import '../app.css';
-	import { onMount } from 'svelte';
+	import { onMount, onDestroy } from 'svelte';
 	import { browser } from '$app/environment';
 	import { page } from '$app/stores';
 	import { accent } from '$lib/accent.js';
+	import type { ModelStatus } from '$lib/types.js';
 
 	let { children } = $props();
 
@@ -11,8 +12,43 @@
 	// The store subscriber handles everything; just subscribing here keeps it alive.
 	$effect(() => { void $accent; });
 
+	// ── Model status ───────────────────────────────────────
+	let modelStatus = $state<ModelStatus>({ state: 'unloaded' });
+	let modelEs: EventSource | null = null;
+
+	function refreshModelStatus() {
+		fetch('/api/model/status')
+			.then((r) => r.json())
+			.then((s) => (modelStatus = s as ModelStatus))
+			.catch(() => {});
+	}
+
+	function subscribeModelEvents() {
+		modelEs?.close();
+		modelEs = new EventSource('/api/model/events');
+		modelEs.addEventListener('model_loading',        () => refreshModelStatus());
+		modelEs.addEventListener('model_ready',          () => refreshModelStatus());
+		modelEs.addEventListener('model_unloaded',       () => refreshModelStatus());
+		modelEs.addEventListener('model_waiting_for_gpu',() => refreshModelStatus());
+		modelEs.onerror = () => { /* browser reconnects automatically */ };
+	}
+
+	const modelStateMeta: Record<string, { dot: string; label: string; pulse: boolean }> = {
+		unloaded:        { dot: 'var(--text-dim)',    label: 'model unloaded',  pulse: false },
+		loading:         { dot: '#f0b429',            label: 'model loading…',  pulse: true  },
+		waiting_for_gpu: { dot: '#f97316',            label: 'waiting for GPU', pulse: true  },
+		ready:           { dot: '#5dd47a',            label: 'whisper-large-v3',pulse: false }
+	};
+
+	const modelMeta = $derived(
+		modelStateMeta[modelStatus.state] ?? modelStateMeta.unloaded
+	);
+
 	// Push notification setup
 	onMount(async () => {
+		refreshModelStatus();
+		subscribeModelEvents();
+
 		if (!browser || !('serviceWorker' in navigator) || !('PushManager' in window)) return;
 		try {
 			const reg = await navigator.serviceWorker.ready;
@@ -42,6 +78,8 @@
 		}
 	});
 
+	onDestroy(() => modelEs?.close());
+
 	function urlBase64ToUint8Array(base64: string): Uint8Array {
 		const pad = '='.repeat((4 - (base64.length % 4)) % 4);
 		const b64 = (base64 + pad).replace(/-/g, '+').replace(/_/g, '/');
@@ -135,8 +173,12 @@
 
 		<!-- Status dot -->
 		<div class="status-pill">
-			<div class="status-dot"></div>
+			<div
-			<span>whisper-large-v3</span>
+				class="status-dot"
+				class:pulse={modelMeta.pulse}
+				style="background: {modelMeta.dot}"
+			></div>
+			<span>{modelMeta.label}</span>
 		</div>
 	</nav>
 
@@ -268,8 +310,15 @@
 		width: 6px;
 		height: 6px;
 		border-radius: 3px;
-		background: #5dd47a;
 		flex-shrink: 0;
+		transition: background 0.4s;
+	}
+	.status-dot.pulse {
+		animation: dot-pulse 1.4s ease-in-out infinite;
+	}
+	@keyframes dot-pulse {
+		0%, 100% { opacity: 1; }
+		50%       { opacity: 0.3; }
 	}
 
 	/* ── Main content ─────────────────────────────────────── */

src/routes/api/model/events/+server.ts

@@ -0,0 +1,43 @@
+const WHISPER_URL = process.env.WHISPER_URL ?? 'http://localhost:8080';
+
+/** Relay the whisper /model/events SSE stream to the browser. */
+export async function GET({ request }) {
+	const { default: fetch } = await import('node-fetch');
+
+	const ac = new AbortController();
+	request.signal.addEventListener('abort', () => ac.abort());
+
+	const stream = new ReadableStream({
+		async start(controller) {
+			try {
+				const upstream = await fetch(`${WHISPER_URL}/model/events`, {
+					signal: ac.signal as AbortSignal
+				});
+				if (!upstream.body) {
+					controller.close();
+					return;
+				}
+				for await (const chunk of upstream.body) {
+					if (ac.signal.aborted) break;
+					controller.enqueue(chunk instanceof Buffer ? chunk : Buffer.from(String(chunk)));
+				}
+			} catch {
+				// upstream closed, client disconnected, or whisper unreachable — all fine
+			} finally {
+				controller.close();
+			}
+		},
+		cancel() {
+			ac.abort();
+		}
+	});
+
+	return new Response(stream, {
+		headers: {
+			'Content-Type': 'text/event-stream',
+			'Cache-Control': 'no-cache',
+			Connection: 'keep-alive',
+			'X-Accel-Buffering': 'no'
+		}
+	});
+}

src/routes/api/model/status/+server.ts

@@ -0,0 +1,14 @@
+import { getModelStatus } from '$lib/server/whisper.js';
+
+export async function GET() {
+	try {
+		const status = await getModelStatus();
+		return new Response(JSON.stringify(status), {
+			headers: { 'Content-Type': 'application/json' }
+		});
+	} catch {
+		return new Response(JSON.stringify({ state: 'unloaded' }), {
+			headers: { 'Content-Type': 'application/json' }
+		});
+	}
+}

src/routes/jobs/[id]/+page.svelte

@@ -13,6 +13,7 @@
 	let segments = $state<Segment[]>([]);
 	let error = $state('');
 	let chunkInfo = $state({ chunk: 0, total: 0 });
+	let modelWarming = $state<{ state: string; retryAfterSecs: number } | null>(null);
 	let eventSource: EventSource | null = null;
 
 	const statusLabel: Record<string, string> = {
@@ -83,8 +84,11 @@
 			try {
 				const data = JSON.parse(e.data);
 				if (data.type === 'progress') {
+					modelWarming = null;
 					chunkInfo = { chunk: data.chunk ?? 0, total: data.total ?? 0 };
 					if (job) job = { ...job, progress: data.progress ?? job.progress, status: 'transcribing' };
+				} else if (data.type === 'model_warming') {
+					modelWarming = { state: data.state ?? 'loading', retryAfterSecs: data.retryAfterSecs ?? 30 };
 				} else if (data.type === 'status') {
 					if (job) job = { ...job, status: data.status, progress: data.progress ?? job.progress };
 				} else if (data.type === 'done') {
@@ -215,6 +219,15 @@
 					{/if}
 				</div>
 
+				{#if modelWarming}
+					<div class="warming-notice mono">
+						<svg width="12" height="12" viewBox="0 0 12 12" fill="none" style="flex-shrink:0; animation: spin 1.5s linear infinite">
+							<circle cx="6" cy="6" r="4.5" stroke="currentColor" stroke-width="1.4" fill="none" stroke-dasharray="14 8"/>
+						</svg>
+						Warming up model ({modelWarming.state.replace(/_/g, ' ')}) — retrying in {modelWarming.retryAfterSecs}s…
+					</div>
+				{/if}
+
 				<!-- Progress bar -->
 				<div class="progress-bar-track">
 					<div
@@ -484,6 +497,16 @@
 		color: var(--text-muted);
 	}
 
+	.warming-notice {
+		display: flex;
+		align-items: center;
+		gap: 8px;
+		margin-top: 10px;
+		font-size: 11.5px;
+		color: #f0b429;
+		opacity: 0.9;
+	}
+
 	.progress-bar-track {
 		height: 4px;
 		border-radius: 2px;

src/tests/whisper.test.ts

@@ -31,6 +31,7 @@ describe('submitJob', () => {
 	it('POSTs to /jobs and returns job_id', async () => {
 		mocks.fetch.mockResolvedValue({
 			ok: true,
+			status: 202,
 			json: () => Promise.resolve({ job_id: 'whisper-job-abc' })
 		});
 		const id = await submitJob('/tmp/audio.wav', 'http://host/api/webhook/job-1');
@@ -41,6 +42,7 @@ describe('submitJob', () => {
 		vi.stubEnv('WHISPER_URL', 'http://localhost:8091');
 		mocks.fetch.mockResolvedValue({
 			ok: true,
+			status: 202,
 			json: () => Promise.resolve({ job_id: 'x' })
 		});
 		await submitJob('/tmp/audio.wav', 'http://host/api/webhook/job-1');
@@ -54,6 +56,7 @@ describe('submitJob', () => {
 	it('includes task=transcribe in the form', async () => {
 		mocks.fetch.mockResolvedValue({
 			ok: true,
+			status: 202,
 			json: () => Promise.resolve({ job_id: 'x' })
 		});
 		await submitJob('/tmp/audio.wav', 'http://host/webhook');
@@ -63,6 +66,7 @@ describe('submitJob', () => {
 	it('includes webhook_url in the form', async () => {
 		mocks.fetch.mockResolvedValue({
 			ok: true,
+			status: 202,
 			json: () => Promise.resolve({ job_id: 'x' })
 		});
 		await submitJob('/tmp/audio.wav', 'http://192.168.1.10:3000/api/webhook/job-99');
@@ -75,6 +79,7 @@ describe('submitJob', () => {
 	it('includes language when provided', async () => {
 		mocks.fetch.mockResolvedValue({
 			ok: true,
+			status: 202,
 			json: () => Promise.resolve({ job_id: 'x' })
 		});
 		await submitJob('/tmp/audio.wav', 'http://host/webhook', 'en');
@@ -84,6 +89,7 @@ describe('submitJob', () => {
 	it('omits language field when not provided', async () => {
 		mocks.fetch.mockResolvedValue({
 			ok: true,
+			status: 202,
 			json: () => Promise.resolve({ job_id: 'x' })
 		});
 		await submitJob('/tmp/audio.wav', 'http://host/webhook');