custom-providers: live loaded-model status + thinking support for llamacpp

- Extend ProviderStatus with loadedModels[] populated from /running; llamacpp panel now shows active model name instead of just "N models" - Return DiscoverResult from discoverProvider (models + loadedModels) - All llamacpp models get reasoning:true + compat.thinkingFormat:"qwen-chat-template" so Ctrl+T / Shift+Tab work for both Qwen3 and Gemma4 (verified live)
2026-06-29 16:28:30 +02:00
parent 336d26bfcd
commit 9d05a67512
5 changed files with 89 additions and 82 deletions
@@ -105,11 +105,12 @@ describe("discoverProvider dispatch", () => {
      apiKey: "ollama",
      defaultCtx: 8192,
    };
-    const models = await discoverProvider(p);
-    expect(models?.length).toBe(2);
+    const result = await discoverProvider(p);
+    expect(result?.models.length).toBe(2);
    expect(calls.some((u) => u.endsWith("/running"))).toBe(false);
    expect(calls.filter((u) => u.endsWith("/api/show")).length).toBe(2);
-    expect(models?.[0].contextWindow).toBe(16384);
+    expect(result?.models[0].contextWindow).toBe(16384);
+    expect(result?.loadedModels).toEqual([]);
  });

  it("llamacpp provider: probes /running once, never /api/show", async () => {
@@ -120,13 +121,14 @@ describe("discoverProvider dispatch", () => {
      apiKey: "llamacpp",
      defaultCtx: 60000,
    };
-    const models = await discoverProvider(p);
-    expect(models?.length).toBe(2);
+    const result = await discoverProvider(p);
+    expect(result?.models.length).toBe(2);
    expect(calls.some((u) => u.endsWith("/api/show"))).toBe(false);
    expect(calls.filter((u) => u.endsWith("/running")).length).toBe(1);
-    const byId = new Map(models?.map((m) => [m.id, m.contextWindow]));
+    const byId = new Map(result?.models.map((m) => [m.id, m.contextWindow]));
    expect(byId.get("llama3")).toBe(32768);
    expect(byId.get("qwen3")).toBe(65536);
+    expect(result?.loadedModels).toEqual(["llama3", "qwen3"]);
  });

  it("strips a trailing /v1 from baseUrl when deriving the native URL", async () => {
@@ -157,7 +159,37 @@ describe("discoverProvider dispatch", () => {
      apiKey: "ollama",
      defaultCtx: 4242,
    };
-    const models = await discoverProvider(p);
-    expect(models?.[0].contextWindow).toBe(4242);
+    const result = await discoverProvider(p);
+    expect(result?.models[0].contextWindow).toBe(4242);
+  });
+
+  it("marks all llamacpp models as reasoning with qwen-chat-template compat", async () => {
+    const p: CustomProvider = {
+      name: "llamacpp",
+      kind: "llamacpp",
+      baseUrl: "http://localhost:9000/v1",
+      apiKey: "llamacpp",
+      defaultCtx: 60000,
+    };
+    const result = await discoverProvider(p);
+    for (const m of result?.models ?? []) {
+      expect(m.reasoning).toBe(true);
+      expect(m.compat?.thinkingFormat).toBe("qwen-chat-template");
+    }
+  });
+
+  it("ollama models are not marked as reasoning", async () => {
+    const p: CustomProvider = {
+      name: "ollama",
+      kind: "ollama",
+      baseUrl: "http://localhost:11434/v1",
+      apiKey: "ollama",
+      defaultCtx: 8192,
+    };
+    const result = await discoverProvider(p);
+    for (const m of result?.models ?? []) {
+      expect(m.reasoning).toBe(false);
+      expect(m.compat).toBeUndefined();
+    }
  });
 });
@@ -4,6 +4,7 @@ export interface DiscoveredModel {
  id: string;
  name: string;
  reasoning: boolean;
+  compat?: { thinkingFormat: "qwen-chat-template" };
  input: Array<"text" | "image">;
  cost: { input: number; output: number; cacheRead: number; cacheWrite: number };
  contextWindow: number;
@@ -13,6 +14,12 @@ export interface DiscoveredModel {
 export interface ProviderStatus {
  ok: boolean;
  modelCount: number;
+  loadedModels: string[];
+}
+
+export interface DiscoverResult {
+  models: DiscoveredModel[];
+  loadedModels: string[];
 }

 async function fetchJson<T>(
@@ -74,21 +81,31 @@ export function parseContextFromCmd(cmd: string): number | null {
  return Number.isFinite(n) && n > 0 ? n : null;
 }

-async function fetchLlamaSwapContexts(nativeUrl: string): Promise<Map<string, number> | null> {
+interface LlamaSwapInfo {
+  contexts: Map<string, number>;
+  loadedModels: string[];
+}
+
+async function fetchLlamaSwapContexts(nativeUrl: string): Promise<LlamaSwapInfo | null> {
  const running = await fetchJson<{ running?: Array<{ model?: string; cmd?: string }> }>(
    `${nativeUrl}/running`,
  );
  if (!running || !Array.isArray(running.running)) return null;
-  const map = new Map<string, number>();
+  const contexts = new Map<string, number>();
+  const loadedModels: string[] = [];
  for (const entry of running.running) {
-    if (!entry.model || !entry.cmd) continue;
-    const ctx = parseContextFromCmd(entry.cmd);
-    if (ctx !== null) map.set(entry.model, ctx);
+    if (!entry.model) continue;
+    loadedModels.push(entry.model);
+    if (entry.cmd) {
+      const ctx = parseContextFromCmd(entry.cmd);
+      if (ctx !== null) contexts.set(entry.model, ctx);
+    }
  }
-  return map;
+  return { contexts, loadedModels };
 }

-export async function discoverProvider(p: CustomProvider): Promise<DiscoveredModel[] | null> {
+
+export async function discoverProvider(p: CustomProvider): Promise<DiscoverResult | null> {
  const baseUrl = p.baseUrl.replace(/\/$/, "");
  const list = await fetchJson<{ data?: Array<{ id: string }> }>(`${baseUrl}/models`);
  if (!list?.data?.length) return null;
@@ -96,22 +113,27 @@ export async function discoverProvider(p: CustomProvider): Promise<DiscoveredMod
  if (!chatModels.length) return null;

  const nativeUrl = baseUrl.replace(/\/v1$/, "");
-  const llamaSwapContexts =
+  const llamaSwapInfo =
    p.kind === "llamacpp" ? await fetchLlamaSwapContexts(nativeUrl) : null;
+  const loadedModels = llamaSwapInfo?.loadedModels ?? [];

-  return Promise.all(
+  const models = await Promise.all(
    chatModels.map(async (m) => {
      let ctx = p.defaultCtx;
-      if (p.kind === "llamacpp" && llamaSwapContexts) {
-        ctx = llamaSwapContexts.get(m.id) ?? p.defaultCtx;
+      if (p.kind === "llamacpp" && llamaSwapInfo) {
+        ctx = llamaSwapInfo.contexts.get(m.id) ?? p.defaultCtx;
      } else if (p.kind === "ollama") {
        const real = await ollamaContext(nativeUrl, m.id);
        if (real) ctx = real;
      }
+      const thinking: { reasoning: true; compat: { thinkingFormat: "qwen-chat-template" } } | { reasoning: false } =
+        p.kind === "llamacpp"
+          ? { reasoning: true, compat: { thinkingFormat: "qwen-chat-template" } }
+          : { reasoning: false };
      return {
        id: m.id,
        name: m.id,
-        reasoning: false,
+        ...thinking,
        input: ["text"] as Array<"text" | "image">,
        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
        contextWindow: ctx,
@@ -119,4 +141,5 @@ export async function discoverProvider(p: CustomProvider): Promise<DiscoveredMod
      };
    }),
  );
+  return { models, loadedModels };
 }
@@ -12,19 +12,19 @@ export default async function (pi: ExtensionAPI) {
      pi.unregisterProvider(p.name);
      registered.delete(p.name);
    }
-    const models = await discoverProvider(p);
-    if (models && models.length > 0) {
+    const result = await discoverProvider(p);
+    if (result && result.models.length > 0) {
      pi.registerProvider(p.name, {
        name: p.name,
        api: "openai-completions",
        baseUrl: p.baseUrl,
        apiKey: p.apiKey || p.name,
-        models,
+        models: result.models,
      });
      registered.add(p.name);
-      statusByName.set(p.name, { ok: true, modelCount: models.length });
+      statusByName.set(p.name, { ok: true, modelCount: result.models.length, loadedModels: result.loadedModels });
    } else {
-      statusByName.set(p.name, { ok: false, modelCount: 0 });
+      statusByName.set(p.name, { ok: false, modelCount: 0, loadedModels: [] });
    }
  }

@@ -24,13 +24,19 @@ type PanelAction =
  | { kind: "remove"; name: string }
  | { kind: "refresh"; name: string };

-function statusLabel(status: ProviderStatus | undefined): string {
+function statusLabel(p: CustomProvider, status: ProviderStatus | undefined): string {
  if (!status) return "not checked";
-  return status.ok ? `${status.modelCount} models` : "unreachable";
+  if (!status.ok) return "unreachable";
+  if (p.kind === "llamacpp" && status.loadedModels.length > 0) {
+    const active = status.loadedModels[0];
+    const rest = status.modelCount - 1;
+    return rest > 0 ? `${active} (+${rest})` : active;
+  }
+  return `${status.modelCount} models`;
 }

 function providerLine(p: CustomProvider, status: ProviderStatus | undefined): string {
-  return `${p.name.padEnd(12)} [${p.kind.padEnd(8)}] ${p.baseUrl}   [${statusLabel(status)}]`;
+  return `${p.name.padEnd(12)} [${p.kind.padEnd(8)}] ${p.baseUrl}   [${statusLabel(p, status)}]`;
 }

 function showList(
@@ -1456,9 +1456,6 @@
      "cpu": [
        "arm64"
      ],
-      "libc": [
-        "glibc"
-      ],
      "license": "MIT",
      "optional": true,
      "os": [
@@ -1476,9 +1473,6 @@
      "cpu": [
        "arm64"
      ],
-      "libc": [
-        "musl"
-      ],
      "license": "MIT",
      "optional": true,
      "os": [
@@ -1496,9 +1490,6 @@
      "cpu": [
        "riscv64"
      ],
-      "libc": [
-        "glibc"
-      ],
      "license": "MIT",
      "optional": true,
      "os": [
@@ -1516,9 +1507,6 @@
      "cpu": [
        "x64"
      ],
-      "libc": [
-        "glibc"
-      ],
      "license": "MIT",
      "optional": true,
      "os": [
@@ -1536,9 +1524,6 @@
      "cpu": [
        "x64"
      ],
-      "libc": [
-        "musl"
-      ],
      "license": "MIT",
      "optional": true,
      "os": [
@@ -1788,9 +1773,6 @@
        "arm"
      ],
      "dev": true,
-      "libc": [
-        "glibc"
-      ],
      "license": "MIT",
      "optional": true,
      "os": [
@@ -1805,9 +1787,6 @@
        "arm"
      ],
      "dev": true,
-      "libc": [
-        "musl"
-      ],
      "license": "MIT",
      "optional": true,
      "os": [
@@ -1822,9 +1801,6 @@
        "arm64"
      ],
      "dev": true,
-      "libc": [
-        "glibc"
-      ],
      "license": "MIT",
      "optional": true,
      "os": [
@@ -1839,9 +1815,6 @@
        "arm64"
      ],
      "dev": true,
-      "libc": [
-        "musl"
-      ],
      "license": "MIT",
      "optional": true,
      "os": [
@@ -1856,9 +1829,6 @@
        "loong64"
      ],
      "dev": true,
-      "libc": [
-        "glibc"
-      ],
      "license": "MIT",
      "optional": true,
      "os": [
@@ -1873,9 +1843,6 @@
        "loong64"
      ],
      "dev": true,
-      "libc": [
-        "musl"
-      ],
      "license": "MIT",
      "optional": true,
      "os": [
@@ -1890,9 +1857,6 @@
        "ppc64"
      ],
      "dev": true,
-      "libc": [
-        "glibc"
-      ],
      "license": "MIT",
      "optional": true,
      "os": [
@@ -1907,9 +1871,6 @@
        "ppc64"
      ],
      "dev": true,
-      "libc": [
-        "musl"
-      ],
      "license": "MIT",
      "optional": true,
      "os": [
@@ -1924,9 +1885,6 @@
        "riscv64"
      ],
      "dev": true,
-      "libc": [
-        "glibc"
-      ],
      "license": "MIT",
      "optional": true,
      "os": [
@@ -1941,9 +1899,6 @@
        "riscv64"
      ],
      "dev": true,
-      "libc": [
-        "musl"
-      ],
      "license": "MIT",
      "optional": true,
      "os": [
@@ -1958,9 +1913,6 @@
        "s390x"
      ],
      "dev": true,
-      "libc": [
-        "glibc"
-      ],
      "license": "MIT",
      "optional": true,
      "os": [
@@ -1975,9 +1927,6 @@
        "x64"
      ],
      "dev": true,
-      "libc": [
-        "glibc"
-      ],
      "license": "MIT",
      "optional": true,
      "os": [
@@ -1992,9 +1941,6 @@
        "x64"
      ],
      "dev": true,
-      "libc": [
-        "musl"
-      ],
      "license": "MIT",
      "optional": true,
      "os": [