Skip to content

A modern and easy to use dashboard. 80+ integrations. 20K+ icons built in. Authentication out of the box. No YAML, drag and drop configuration.

fix: target llama.cpp stats to displayed model (#7150)

Thomasd46f69654973
diff --git a/packages/integrations/src/llama-cpp/llamacpp-integration.ts b/packages/integrations/src/llama-cpp/llamacpp-integration.ts
index 9640c78ed..fcc3984e9 100644
--- a/packages/integrations/src/llama-cpp/llamacpp-integration.ts
+++ b/packages/integrations/src/llama-cpp/llamacpp-integration.ts
@@ -46,12 +46,9 @@ export class LlamacppIntegration extends Integration {
}
public async getStatsAsync(): Promise<LlamacppStats> {
- const [healthResponse, modelsResponse, metricsResponse, slotsResponse] = await Promise.all([
+ const [healthResponse, modelsResponse] = await Promise.all([
fetchWithTrustedCertificatesAsync(this.url("/health"), { timeout: LLMACPP_REQUEST_TIMEOUT_MS }),
fetchWithTrustedCertificatesAsync(this.url("/v1/models"), { timeout: LLMACPP_REQUEST_TIMEOUT_MS }),
- fetchWithTrustedCertificatesAsync(this.url("/metrics"), { timeout: LLMACPP_REQUEST_TIMEOUT_MS }),
- // /slots is not present on every build; a missing endpoint degrades to no context usage.
- fetchWithTrustedCertificatesAsync(this.url("/slots"), { timeout: LLMACPP_REQUEST_TIMEOUT_MS }).catch(() => null),
]);
if (!healthResponse.ok) {
@@ -61,14 +58,28 @@ export class LlamacppIntegration extends Integration {
throw new ResponseError(modelsResponse);
}
+ const [health, models] = await Promise.all([
+ parseLlamacppHealthAsync(healthResponse),
+ parseLlamacppModelsAsync(modelsResponse),
+ ]);
+ const firstModelId = models.data[0]?.id;
+
+ const [metricsResponse, slotsResponse] = await Promise.all([
+ fetchWithTrustedCertificatesAsync(this.url("/metrics", { model: firstModelId }), {
+ timeout: LLMACPP_REQUEST_TIMEOUT_MS,
+ }),
+ // /slots is not present on every build; a missing endpoint degrades to no context usage.
+ fetchWithTrustedCertificatesAsync(this.url("/slots", { model: firstModelId }), {
+ timeout: LLMACPP_REQUEST_TIMEOUT_MS,
+ }).catch(() => null),
+ ]);
+
// /metrics only exists when llama-server was started with --metrics; a non-OK
// response (typically 501 Not Implemented) degrades to no rate counters instead
// of dropping the health and model data that the other endpoints still provide.
const metricsAvailable = metricsResponse.ok;
- const [health, models, metrics, slots] = await Promise.all([
- parseLlamacppHealthAsync(healthResponse),
- parseLlamacppModelsAsync(modelsResponse),
+ const [metrics, slots] = await Promise.all([
metricsAvailable ? parseLlamacppMetricsAsync(metricsResponse) : Promise.resolve([]),
slotsResponse?.ok ? parseLlamacppSlotsAsync(slotsResponse) : Promise.resolve(null),
]);
diff --git a/packages/integrations/src/llama-cpp/test/llamacpp-integration.spec.ts b/packages/integrations/src/llama-cpp/test/llamacpp-integration.spec.ts
index 966ed9083..88d981170 100644
--- a/packages/integrations/src/llama-cpp/test/llamacpp-integration.spec.ts
+++ b/packages/integrations/src/llama-cpp/test/llamacpp-integration.spec.ts
@@ -342,7 +342,12 @@ describe("LlamacppIntegration getStatsAsync", () => {
const requestedUrls = mockFetch.mock.calls.map((call) => String(call[0])).toSorted();
expect(requestedUrls).toStrictEqual(
- [`${TEST_URL}/health`, `${TEST_URL}/metrics`, `${TEST_URL}/slots`, `${TEST_URL}/v1/models`].toSorted(),
+ [
+ `${TEST_URL}/health`,
+ `${TEST_URL}/metrics?model=%2Froot%2Fmodels%2Funsloth%2FQwen3.8-27B-UD-Q4_K_XL.gguf`,
+ `${TEST_URL}/slots?model=%2Froot%2Fmodels%2Funsloth%2FQwen3.8-27B-UD-Q4_K_XL.gguf`,
+ `${TEST_URL}/v1/models`,
+ ].toSorted(),
);
});