Skip to content

armgit / ODS

Public Imported from GitHub
0

ODS V3 Pre-Release: Public testing and refinement ahead of the official V3 launch. Turn your PC, Mac, or Linux box into a private AI server.

fix(dashboard): leave out the throughput tile until a rate exists (#8194)

Mike Bradley4bdf3190348b
diff --git a/ods/extensions/services/dashboard/src/pages/Dashboard.jsx b/ods/extensions/services/dashboard/src/pages/Dashboard.jsx
index 789441ccb..f0b997528 100644
--- a/ods/extensions/services/dashboard/src/pages/Dashboard.jsx
+++ b/ods/extensions/services/dashboard/src/pages/Dashboard.jsx
@@ -646,7 +646,7 @@ function buildChartPoints(values, maxValue) {
}
function throughputLabel(inference, stale = false) {
- if (!Number.isFinite(inference?.tokensPerSecond)) return 'Telemetry unavailable'
+ if (!Number.isFinite(inference?.tokensPerSecond)) return 'Not reported'
if (stale || inference?.throughputState === 'unavailable') return 'Last known rate · telemetry unavailable'
if (inference?.throughputMode === 'cloud_request_average') return 'Last cloud request · includes latency'
if (inference?.throughputState === 'retained') return 'Last run'
@@ -906,13 +906,18 @@ export default function Dashboard({ status, loading, compact = false }) {
})
}
- systemMetrics.push(
- {
+ // The API keeps a model's last measured rate. Until it has one (some native
+ // engines never report it), the tile is left out rather than "unavailable".
+ if (Number.isFinite(status?.inference?.tokensPerSecond)) {
+ systemMetrics.push({
icon: Zap,
label: 'Tokens / second',
- value: Number.isFinite(status?.inference?.tokensPerSecond) ? `${status.inference.tokensPerSecond.toFixed(1)} tok/s` : '—',
- subvalue: throughputLabel(status?.inference,status?.clientTelemetry?.stale),
- },
+ value: `${status.inference.tokensPerSecond.toFixed(1)} tok/s`,
+ subvalue: throughputLabel(status.inference,status?.clientTelemetry?.stale),
+ })
+ }
+
+ systemMetrics.push(
{
icon: Brackets,
label: 'Context',
@@ -1189,7 +1194,7 @@ const SystemOverviewPanel = memo(function SystemOverviewPanel({ tokensPerSecond,
? 'Latest Completion'
: tokenCountMode === 'cumulative'
? 'Accumulated Output'
- : 'Waiting for telemetry'}
+ : 'Not reported'}
values={generated.values}
timestamps={generated.timestamps}
range={range}
@@ -1263,9 +1268,9 @@ const OverviewChart = memo(function OverviewChart({
</div>
</div>
- {delta == null ? (
+ {delta == null ? currentDisplay !== '—' && (
<div className="mb-2 whitespace-nowrap text-xs font-medium text-theme-text-muted">
- {currentDisplay === '—' ? 'telemetry unavailable' : 'collecting samples'}
+ collecting samples
</div>
) : (
<div className={`mb-2 whitespace-nowrap text-xs font-semibold ${deltaTone}`}>
@@ -1344,7 +1349,7 @@ const OverviewChart = memo(function OverviewChart({
fontSize="11"
fontWeight="600"
>
- {currentDisplay === '—' ? 'telemetry unavailable' : 'collecting telemetry'}
+ {currentDisplay === '—' ? 'No data' : 'collecting telemetry'}
</text>
)}
diff --git a/ods/extensions/services/dashboard/src/pages/Dashboard.test.jsx b/ods/extensions/services/dashboard/src/pages/Dashboard.test.jsx
index b891e1d46..fc7a64401 100644
--- a/ods/extensions/services/dashboard/src/pages/Dashboard.test.jsx
+++ b/ods/extensions/services/dashboard/src/pages/Dashboard.test.jsx
@@ -134,14 +134,39 @@ describe('Dashboard system overview', () => {
expect(screen.getByText('Accumulated Output')).toBeInTheDocument()
})
- it.each([[8.25, '8.3 tok/s'], [0, '0.0 tok/s'], [null, '—'], [undefined, '—']])('shows a real compact throughput reading for %s', async (tokensPerSecond, expected) => {
+ it.each([[8.25, '8.3 tok/s'], [0, '0.0 tok/s']])('shows a real compact throughput reading for %s', async (tokensPerSecond, expected) => {
render(<Dashboard compact status={{...baseStatus, inference:{...baseStatus.inference, tokensPerSecond}}} loading={false}/>)
const row = screen.getByText('Tokens / second').closest('.dashboard-metric-row')
expect(within(row).getByText(expected)).toBeVisible()
- expect(within(row).getByText(tokensPerSecond == null ? 'Telemetry unavailable' : 'Runtime reading')).toBeVisible()
+ expect(within(row).getByText('Runtime reading')).toBeVisible()
await waitFor(() => expect(fetch).toHaveBeenCalledWith('/api/features'))
})
+ it.each([null, undefined])('leaves out Tokens / second while the runtime has reported no rate (%s)', async tokensPerSecond => {
+ // A native engine that never reports throughput: nothing reads as a fault.
+ render(<Dashboard compact status={{...baseStatus,
+ inference:{...baseStatus.inference, tokensPerSecond, throughputState:'unavailable', throughputMode:'unavailable', throughputSampledAt:null},
+ cpu:{percent:12,scope:'host'},
+ gpu:{name:'NVIDIA GeForce RTX 4090',memoryType:'discrete',utilization:3,vramUsed:10,vramTotal:24,temperature:41,powerDraw:4}}} loading={false}/>)
+ expect(screen.queryByText('Tokens / second')).toBeNull()
+ expect(screen.queryByText(/telemetry unavailable/i)).toBeNull()
+ expect(within(screen.getByText('Context').closest('.dashboard-metric-row')).getByText('32k')).toBeVisible()
+ expect(within(screen.getByText('Model').closest('.dashboard-metric-row')).getByText('qwen')).toBeVisible()
+ await waitFor(() => expect(fetch).toHaveBeenCalledWith('/api/features'))
+ })
+
+ it('names a missing reading neutrally in the full overview charts', async () => {
+ await renderDashboard({...baseStatus,
+ inference:{...baseStatus.inference, tokensPerSecond:null, lifetimeTokens:null, tokenCountMode:'unavailable', throughputState:'unavailable'}})
+ const overview = screen.getByText('System Overview').closest('section')
+ const throughput = screen.getByRole('img', {name:'TOKENS PER SECOND chart'}).parentElement
+ expect(within(throughput).getByText('Not reported')).toBeVisible()
+ expect(within(throughput).getByText('—')).toBeVisible()
+ expect(within(overview).getAllByText('No data')).toHaveLength(2)
+ expect(within(overview).queryByText(/telemetry/i)).toBeNull()
+ expect(screen.queryByText('Tokens / second')).toBeNull()
+ })
+
it.each([[0, '0.0'], [null, '—'], [undefined, '—']])('keeps full throughput reading %s distinct from unavailable history', async (tokensPerSecond, expected) => {
await renderDashboard({...baseStatus, inference:{...baseStatus.inference, tokensPerSecond, lifetimeTokens:null}})
const throughput = screen.getByRole('img', {name:'TOKENS PER SECOND chart'}).parentElement