@akon-labs/nexus-cli 0.1.0

package/dist/extensions/llama/client.js

@@ -0,0 +1,302 @@
+function errorMessage(payload, fallback) {
+    if (typeof payload !== "object" || payload === null)
+        return fallback;
+    const error = payload.error;
+    if (typeof error !== "object" || error === null)
+        return fallback;
+    const message = error.message;
+    return typeof message === "string" && message ? message : fallback;
+}
+function isModelInfo(value) {
+    if (typeof value !== "object" || value === null)
+        return false;
+    const candidate = value;
+    return typeof candidate.id === "string" && typeof candidate.status?.value === "string";
+}
+function linkSignal(source, target) {
+    if (!source)
+        return () => { };
+    if (source.aborted) {
+        target.abort(source.reason);
+        return () => { };
+    }
+    const abort = () => target.abort(source.reason);
+    source.addEventListener("abort", abort, { once: true });
+    return () => source.removeEventListener("abort", abort);
+}
+function sleep(ms, signal) {
+    return new Promise((resolve, reject) => {
+        if (signal?.aborted) {
+            reject(signal.reason ?? new Error("Cancelled"));
+            return;
+        }
+        const abort = () => {
+            clearTimeout(timeout);
+            reject(signal?.reason ?? new Error("Cancelled"));
+        };
+        const timeout = setTimeout(() => {
+            signal?.removeEventListener("abort", abort);
+            resolve();
+        }, ms);
+        signal?.addEventListener("abort", abort, { once: true });
+    });
+}
+function parseLoadProgress(data) {
+    if (typeof data !== "object" || data === null)
+        return undefined;
+    const progress = data.progress;
+    if (typeof progress !== "object" || progress === null)
+        return undefined;
+    const value = progress;
+    const stage = typeof value.current === "string" ? value.current : typeof value.stage === "string" ? value.stage : undefined;
+    const stages = Array.isArray(value.stages)
+        ? value.stages.filter((entry) => typeof entry === "string")
+        : [];
+    const stageRatio = typeof value.value === "number" ? Math.max(0, Math.min(1, value.value)) : undefined;
+    let ratio = stageRatio;
+    if (stage && stages.length > 0) {
+        const index = stages.indexOf(stage);
+        if (index >= 0)
+            ratio = (index + (stageRatio ?? 0)) / stages.length;
+    }
+    return {
+        message: stage ? `Loading ${stage.replaceAll("_", " ")}` : "Loading model",
+        ratio,
+    };
+}
+function parseDownloadProgress(data) {
+    if (typeof data !== "object" || data === null)
+        return undefined;
+    const nested = data.progress;
+    const files = typeof nested === "object" && nested !== null ? nested : data;
+    let done = 0;
+    let total = 0;
+    for (const value of Object.values(files)) {
+        if (typeof value !== "object" || value === null)
+            continue;
+        const entry = value;
+        if (typeof entry.done !== "number" || typeof entry.total !== "number")
+            continue;
+        done += entry.done;
+        total += entry.total;
+    }
+    if (total <= 0)
+        return undefined;
+    return {
+        message: "Downloading model",
+        ratio: done / total,
+        detail: `${formatBytes(done)} / ${formatBytes(total)}`,
+    };
+}
+export function formatBytes(bytes) {
+    if (bytes < 1024)
+        return `${bytes} B`;
+    const units = ["KiB", "MiB", "GiB", "TiB"];
+    let value = bytes / 1024;
+    let unit = units[0];
+    for (let index = 1; index < units.length && value >= 1024; index++) {
+        value /= 1024;
+        unit = units[index];
+    }
+    return `${value >= 10 ? value.toFixed(1) : value.toFixed(2)} ${unit}`;
+}
+export function normalizeLlamaServerUrl(value) {
+    const url = new URL(value.trim());
+    if (url.protocol !== "http:" && url.protocol !== "https:") {
+        throw new Error("Server URL must use http or https");
+    }
+    url.hash = "";
+    url.search = "";
+    url.pathname = url.pathname.replace(/\/+$/u, "").replace(/\/v1$/u, "") || "/";
+    return url.toString().replace(/\/$/u, "");
+}
+export function llamaInferenceUrl(serverUrl) {
+    return `${normalizeLlamaServerUrl(serverUrl)}/v1`;
+}
+export class LlamaClient {
+    serverUrl;
+    apiKey;
+    constructor(serverUrl, apiKey) {
+        this.serverUrl = normalizeLlamaServerUrl(serverUrl);
+        this.apiKey = apiKey;
+    }
+    async request(path, init = {}) {
+        const headers = new Headers(init.headers);
+        if (init.body !== undefined)
+            headers.set("Content-Type", "application/json");
+        if (this.apiKey)
+            headers.set("Authorization", `Bearer ${this.apiKey}`);
+        const timeout = AbortSignal.timeout(15_000);
+        const signal = init.signal ? AbortSignal.any([init.signal, timeout]) : timeout;
+        const response = await fetch(`${this.serverUrl}${path}`, { ...init, headers, signal });
+        let payload;
+        try {
+            payload = await response.json();
+        }
+        catch {
+            payload = undefined;
+        }
+        if (!response.ok)
+            throw new Error(errorMessage(payload, `llama.cpp returned HTTP ${response.status}`));
+        return payload;
+    }
+    async list(options = {}) {
+        const payload = await this.request(`/models${options.reload ? "?reload=1" : ""}`, { signal: options.signal });
+        if (typeof payload !== "object" || payload === null || !Array.isArray(payload.data)) {
+            throw new Error("llama.cpp returned an invalid model catalog");
+        }
+        const data = payload.data;
+        if (!data.every(isModelInfo))
+            throw new Error("Server is not running in llama.cpp router mode");
+        return data;
+    }
+    async load(model, signal) {
+        await this.request("/models/load", { method: "POST", body: JSON.stringify({ model }), signal });
+    }
+    async unload(model, signal) {
+        await this.request("/models/unload", { method: "POST", body: JSON.stringify({ model }), signal });
+    }
+    async unloadAndWait(model, signal) {
+        await this.unload(model, signal);
+        while (true) {
+            const entry = (await this.list({ signal })).find((candidate) => candidate.id === model);
+            if (!entry || entry.status.value === "unloaded")
+                return;
+            await sleep(100, signal);
+        }
+    }
+    async download(model, signal) {
+        await this.request("/models", { method: "POST", body: JSON.stringify({ model }), signal });
+    }
+    async watch(onEvent, signal) {
+        const headers = new Headers();
+        if (this.apiKey)
+            headers.set("Authorization", `Bearer ${this.apiKey}`);
+        const response = await fetch(`${this.serverUrl}/models/sse`, { headers, signal });
+        if (!response.ok || !response.body)
+            throw new Error(`llama.cpp SSE returned HTTP ${response.status}`);
+        const reader = response.body.getReader();
+        const decoder = new TextDecoder();
+        let buffer = "";
+        while (true) {
+            const chunk = await reader.read();
+            if (chunk.done)
+                break;
+            buffer += decoder.decode(chunk.value, { stream: true }).replaceAll("\r\n", "\n");
+            let boundary = buffer.indexOf("\n\n");
+            while (boundary >= 0) {
+                const frame = buffer.slice(0, boundary);
+                buffer = buffer.slice(boundary + 2);
+                const data = frame
+                    .split("\n")
+                    .filter((line) => line.startsWith("data:"))
+                    .map((line) => line.slice(5).trimStart())
+                    .join("\n");
+                if (data) {
+                    try {
+                        const event = JSON.parse(data);
+                        if (event && typeof event.model === "string" && typeof event.event === "string")
+                            onEvent(event);
+                    }
+                    catch {
+                        // Ignore malformed events; catalog polling remains authoritative.
+                    }
+                }
+                boundary = buffer.indexOf("\n\n");
+            }
+        }
+    }
+    async loadAndWait(model, onProgress, signal) {
+        const watcher = new AbortController();
+        const unlink = linkSignal(signal, watcher);
+        let eventLoaded = false;
+        let eventError;
+        void this.watch((event) => {
+            if (event.model !== model)
+                return;
+            if (event.event !== "model_status" && event.event !== "status_change")
+                return;
+            const data = event.data;
+            if (data?.status === "loaded")
+                eventLoaded = true;
+            if (data?.status === "unloaded")
+                eventError = "Model failed to load";
+            const progress = parseLoadProgress(event.data);
+            if (progress)
+                onProgress(progress);
+        }, watcher.signal).catch(() => { });
+        try {
+            await this.load(model, signal);
+            onProgress({ message: "Loading model" });
+            while (true) {
+                if (signal?.aborted)
+                    throw signal.reason ?? new Error("Cancelled");
+                const entry = (await this.list({ signal })).find((candidate) => candidate.id === model);
+                if (entry?.status.value === "loaded")
+                    return entry;
+                if (eventLoaded && !entry)
+                    return { id: model, status: { value: "loaded" } };
+                if (entry?.status.failed || eventError) {
+                    throw new Error(entry?.status.exit_code === undefined
+                        ? (eventError ?? "Model failed to load")
+                        : `Model exited with code ${entry.status.exit_code}`);
+                }
+                await sleep(250, signal);
+            }
+        }
+        finally {
+            unlink();
+            watcher.abort();
+        }
+    }
+    async downloadAndWait(model, onProgress, signal) {
+        const watcher = new AbortController();
+        const unlink = linkSignal(signal, watcher);
+        let finished = false;
+        let failure;
+        let sawDownloading = false;
+        let polls = 0;
+        void this.watch((event) => {
+            if (event.model !== model)
+                return;
+            if (event.event === "download_finished")
+                finished = true;
+            if (event.event === "download_failed")
+                failure = errorMessage(event.data, "Download failed");
+            if (event.event === "download_progress") {
+                sawDownloading = true;
+                const progress = parseDownloadProgress(event.data);
+                if (progress)
+                    onProgress(progress);
+            }
+        }, watcher.signal).catch(() => { });
+        try {
+            await this.download(model, signal);
+            onProgress({ message: "Downloading model" });
+            while (true) {
+                if (signal?.aborted)
+                    throw signal.reason ?? new Error("Cancelled");
+                if (failure)
+                    throw new Error(failure);
+                const models = await this.list({ signal });
+                polls++;
+                const entry = models.find((candidate) => candidate.id === model);
+                if (entry?.status.value === "downloading") {
+                    sawDownloading = true;
+                    const progress = parseDownloadProgress(entry.status.progress);
+                    if (progress)
+                        onProgress(progress);
+                }
+                else if (finished || (entry && (sawDownloading || polls >= 2))) {
+                    return this.list({ reload: true, signal });
+                }
+                await sleep(500, signal);
+            }
+        }
+        finally {
+            unlink();
+            watcher.abort();
+        }
+    }
+}
+//# sourceMappingURL=client.js.map

package/dist/extensions/llama/client.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"client.js","sourceRoot":"","sources":["../../../src/extensions/llama/client.ts"],"names":[],"mappings":"AA0CA,SAAS,YAAY,CAAC,OAAgB,EAAE,QAAgB,EAAU;IACjE,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,IAAI;QAAE,OAAO,QAAQ,CAAC;IACrE,MAAM,KAAK,GAAI,OAA+B,CAAC,KAAK,CAAC;IACrD,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI;QAAE,OAAO,QAAQ,CAAC;IACjE,MAAM,OAAO,GAAI,KAA+B,CAAC,OAAO,CAAC;IACzD,OAAO,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,QAAQ,CAAC;AAAA,CACnE;AAED,SAAS,WAAW,CAAC,KAAc,EAA2B;IAC7D,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI;QAAE,OAAO,KAAK,CAAC;IAC9D,MAAM,SAAS,GAAG,KAAuD,CAAC;IAC1E,OAAO,OAAO,SAAS,CAAC,EAAE,KAAK,QAAQ,IAAI,OAAO,SAAS,CAAC,MAAM,EAAE,KAAK,KAAK,QAAQ,CAAC;AAAA,CACvF;AAED,SAAS,UAAU,CAAC,MAA+B,EAAE,MAAuB,EAAc;IACzF,IAAI,CAAC,MAAM;QAAE,OAAO,GAAG,EAAE,CAAC,EAAC,CAAC,CAAC;IAC7B,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;QACpB,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;QAC5B,OAAO,GAAG,EAAE,CAAC,EAAC,CAAC,CAAC;IACjB,CAAC;IACD,MAAM,KAAK,GAAG,GAAG,EAAE,CAAC,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;IAChD,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;IACxD,OAAO,GAAG,EAAE,CAAC,MAAM,CAAC,mBAAmB,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;AAAA,CACxD;AAED,SAAS,KAAK,CAAC,EAAU,EAAE,MAAoB,EAAiB;IAC/D,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE,CAAC;QACvC,IAAI,MAAM,EAAE,OAAO,EAAE,CAAC;YACrB,MAAM,CAAC,MAAM,CAAC,MAAM,IAAI,IAAI,KAAK,CAAC,WAAW,CAAC,CAAC,CAAC;YAChD,OAAO;QACR,CAAC;QACD,MAAM,KAAK,GAAG,GAAG,EAAE,CAAC;YACnB,YAAY,CAAC,OAAO,CAAC,CAAC;YACtB,MAAM,CAAC,MAAM,EAAE,MAAM,IAAI,IAAI,KAAK,CAAC,WAAW,CAAC,CAAC,CAAC;QAAA,CACjD,CAAC;QACF,MAAM,OAAO,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC;YAChC,MAAM,EAAE,mBAAmB,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;YAC5C,OAAO,EAAE,CAAC;QAAA,CACV,EAAE,EAAE,CAAC,CAAC;QACP,MAAM,EAAE,gBAAgB,CAAC,OAAO,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;IAAA,CACzD,CAAC,CAAC;AAAA,CACH;AAED,SAAS,iBAAiB,CAAC,IAAa,EAA6B;IACpE,IAAI,OAAO,IAAI,KAAK,QAAQ,IAAI,IAAI,KAAK,IAAI;QAAE,OAAO,SAAS,CAAC;IAChE,MAAM,QAAQ,GAAI,IAA+B,CAAC,QAAQ,CAAC;IAC3D,IAAI,OAAO,QAAQ,KAAK,QAAQ,IAAI,QAAQ,KAAK,IAAI;QAAE,OAAO,SAAS,CAAC;IACxE,MAAM,KAAK,GAAG,QAAqF,CAAC;IACpG,MAAM,KAAK,GACV,OAAO,KAAK,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,KAAK,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC;IAC/G,MAAM,MAAM,GAAG,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,MAAM,CAAC;QACzC,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAmB,EAAE,CAAC,OAAO,KAAK,KAAK,QAAQ,CAAC;QAC5E,CAAC,CAAC,EAAE,CAAC;IACN,MAAM,UAAU,GAAG,OAAO,KAAK,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACvG,IAAI,KAAK,GAAG,UAAU,CAAC;IACvB,IAAI,KAAK,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAChC,MAAM,KAAK,GAAG,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;QACpC,IAAI,KAAK,IAAI,CAAC;YAAE,KAAK,GAAG,CAAC,KAAK,GAAG,CAAC,UAAU,IAAI,CAAC,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;IACrE,CAAC;IACD,OAAO;QACN,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,WAAW,KAAK,CAAC,UAAU,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC,eAAe;QAC1E,KAAK;KACL,CAAC;AAAA,CACF;AAED,SAAS,qBAAqB,CAAC,IAAa,EAA6B;IACxE,IAAI,OAAO,IAAI,KAAK,QAAQ,IAAI,IAAI,KAAK,IAAI;QAAE,OAAO,SAAS,CAAC;IAChE,MAAM,MAAM,GAAI,IAA+B,CAAC,QAAQ,CAAC;IACzD,MAAM,KAAK,GAAG,OAAO,MAAM,KAAK,QAAQ,IAAI,MAAM,KAAK,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC;IAC5E,IAAI,IAAI,GAAG,CAAC,CAAC;IACb,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,MAAM,CAAC,KAAgC,CAAC,EAAE,CAAC;QACrE,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI;YAAE,SAAS;QAC1D,MAAM,KAAK,GAAG,KAA4C,CAAC;QAC3D,IAAI,OAAO,KAAK,CAAC,IAAI,KAAK,QAAQ,IAAI,OAAO,KAAK,CAAC,KAAK,KAAK,QAAQ;YAAE,SAAS;QAChF,IAAI,IAAI,KAAK,CAAC,IAAI,CAAC;QACnB,KAAK,IAAI,KAAK,CAAC,KAAK,CAAC;IACtB,CAAC;IACD,IAAI,KAAK,IAAI,CAAC;QAAE,OAAO,SAAS,CAAC;IACjC,OAAO;QACN,OAAO,EAAE,mBAAmB;QAC5B,KAAK,EAAE,IAAI,GAAG,KAAK;QACnB,MAAM,EAAE,GAAG,WAAW,CAAC,IAAI,CAAC,MAAM,WAAW,CAAC,KAAK,CAAC,EAAE;KACtD,CAAC;AAAA,CACF;AAED,MAAM,UAAU,WAAW,CAAC,KAAa,EAAU;IAClD,IAAI,KAAK,GAAG,IAAI;QAAE,OAAO,GAAG,KAAK,IAAI,CAAC;IACtC,MAAM,KAAK,GAAG,CAAC,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,CAAC,CAAC;IAC3C,IAAI,KAAK,GAAG,KAAK,GAAG,IAAI,CAAC;IACzB,IAAI,IAAI,GAAG,KAAK,CAAC,CAAC,CAAE,CAAC;IACrB,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,KAAK,CAAC,MAAM,IAAI,KAAK,IAAI,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;QACpE,KAAK,IAAI,IAAI,CAAC;QACd,IAAI,GAAG,KAAK,CAAC,KAAK,CAAE,CAAC;IACtB,CAAC;IACD,OAAO,GAAG,KAAK,IAAI,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,IAAI,EAAE,CAAC;AAAA,CACtE;AAED,MAAM,UAAU,uBAAuB,CAAC,KAAa,EAAU;IAC9D,MAAM,GAAG,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC;IAClC,IAAI,GAAG,CAAC,QAAQ,KAAK,OAAO,IAAI,GAAG,CAAC,QAAQ,KAAK,QAAQ,EAAE,CAAC;QAC3D,MAAM,IAAI,KAAK,CAAC,mCAAmC,CAAC,CAAC;IACtD,CAAC;IACD,GAAG,CAAC,IAAI,GAAG,EAAE,CAAC;IACd,GAAG,CAAC,MAAM,GAAG,EAAE,CAAC;IAChB,GAAG,CAAC,QAAQ,GAAG,GAAG,CAAC,QAAQ,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,QAAQ,EAAE,EAAE,CAAC,IAAI,GAAG,CAAC;IAC9E,OAAO,GAAG,CAAC,QAAQ,EAAE,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;AAAA,CAC1C;AAED,MAAM,UAAU,iBAAiB,CAAC,SAAiB,EAAU;IAC5D,OAAO,GAAG,uBAAuB,CAAC,SAAS,CAAC,KAAK,CAAC;AAAA,CAClD;AAED,MAAM,OAAO,WAAW;IACd,SAAS,CAAS;IACV,MAAM,CAAqB;IAE5C,YAAY,SAAiB,EAAE,MAAe,EAAE;QAC/C,IAAI,CAAC,SAAS,GAAG,uBAAuB,CAAC,SAAS,CAAC,CAAC;QACpD,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;IAAA,CACrB;IAEO,KAAK,CAAC,OAAO,CAAC,IAAY,EAAE,IAAI,GAAgB,EAAE,EAAoB;QAC7E,MAAM,OAAO,GAAG,IAAI,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAC1C,IAAI,IAAI,CAAC,IAAI,KAAK,SAAS;YAAE,OAAO,CAAC,GAAG,CAAC,cAAc,EAAE,kBAAkB,CAAC,CAAC;QAC7E,IAAI,IAAI,CAAC,MAAM;YAAE,OAAO,CAAC,GAAG,CAAC,eAAe,EAAE,UAAU,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC;QACvE,MAAM,OAAO,GAAG,WAAW,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QAC5C,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC;QAC/E,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,IAAI,CAAC,SAAS,GAAG,IAAI,EAAE,EAAE,EAAE,GAAG,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;QACvF,IAAI,OAAgB,CAAC;QACrB,IAAI,CAAC;YACJ,OAAO,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;QACjC,CAAC;QAAC,MAAM,CAAC;YACR,OAAO,GAAG,SAAS,CAAC;QACrB,CAAC;QACD,IAAI,CAAC,QAAQ,CAAC,EAAE;YAAE,MAAM,IAAI,KAAK,CAAC,YAAY,CAAC,OAAO,EAAE,2BAA2B,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;QACvG,OAAO,OAAO,CAAC;IAAA,CACf;IAED,KAAK,CAAC,IAAI,CAAC,OAAO,GAA+C,EAAE,EAA6B;QAC/F,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,UAAU,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,EAAE,EAAE,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC;QAC9G,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,IAAI,IAAI,CAAC,KAAK,CAAC,OAAO,CAAE,OAA8B,CAAC,IAAI,CAAC,EAAE,CAAC;YAC7G,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAC;QAChE,CAAC;QACD,MAAM,IAAI,GAAI,OAA+B,CAAC,IAAI,CAAC;QACnD,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,WAAW,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,gDAAgD,CAAC,CAAC;QAChG,OAAO,IAAI,CAAC;IAAA,CACZ;IAED,KAAK,CAAC,IAAI,CAAC,KAAa,EAAE,MAAoB,EAAiB;QAC9D,MAAM,IAAI,CAAC,OAAO,CAAC,cAAc,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC;IAAA,CAChG;IAED,KAAK,CAAC,MAAM,CAAC,KAAa,EAAE,MAAoB,EAAiB;QAChE,MAAM,IAAI,CAAC,OAAO,CAAC,gBAAgB,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC;IAAA,CAClG;IAED,KAAK,CAAC,aAAa,CAAC,KAAa,EAAE,MAAoB,EAAiB;QACvE,MAAM,IAAI,CAAC,MAAM,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;QACjC,OAAO,IAAI,EAAE,CAAC;YACb,MAAM,KAAK,GAAG,CAAC,MAAM,IAAI,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,EAAE,KAAK,KAAK,CAAC,CAAC;YACxF,IAAI,CAAC,KAAK,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,KAAK,UAAU;gBAAE,OAAO;YACxD,MAAM,KAAK,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;QAC1B,CAAC;IAAA,CACD;IAED,KAAK,CAAC,QAAQ,CAAC,KAAa,EAAE,MAAoB,EAAiB;QAClE,MAAM,IAAI,CAAC,OAAO,CAAC,SAAS,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC;IAAA,CAC3F;IAED,KAAK,CAAC,KAAK,CAAC,OAAyC,EAAE,MAAoB,EAAiB;QAC3F,MAAM,OAAO,GAAG,IAAI,OAAO,EAAE,CAAC;QAC9B,IAAI,IAAI,CAAC,MAAM;YAAE,OAAO,CAAC,GAAG,CAAC,eAAe,EAAE,UAAU,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC;QACvE,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,IAAI,CAAC,SAAS,aAAa,EAAE,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;QAClF,IAAI,CAAC,QAAQ,CAAC,EAAE,IAAI,CAAC,QAAQ,CAAC,IAAI;YAAE,MAAM,IAAI,KAAK,CAAC,+BAA+B,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC;QACtG,MAAM,MAAM,GAAG,QAAQ,CAAC,IAAI,CAAC,SAAS,EAAE,CAAC;QACzC,MAAM,OAAO,GAAG,IAAI,WAAW,EAAE,CAAC;QAClC,IAAI,MAAM,GAAG,EAAE,CAAC;QAChB,OAAO,IAAI,EAAE,CAAC;YACb,MAAM,KAAK,GAAG,MAAM,MAAM,CAAC,IAAI,EAAE,CAAC;YAClC,IAAI,KAAK,CAAC,IAAI;gBAAE,MAAM;YACtB,MAAM,IAAI,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,KAAK,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,UAAU,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC;YACjF,IAAI,QAAQ,GAAG,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;YACtC,OAAO,QAAQ,IAAI,CAAC,EAAE,CAAC;gBACtB,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC;gBACxC,MAAM,GAAG,MAAM,CAAC,KAAK,CAAC,QAAQ,GAAG,CAAC,CAAC,CAAC;gBACpC,MAAM,IAAI,GAAG,KAAK;qBAChB,KAAK,CAAC,IAAI,CAAC;qBACX,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC;qBAC1C,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC;qBACxC,IAAI,CAAC,IAAI,CAAC,CAAC;gBACb,IAAI,IAAI,EAAE,CAAC;oBACV,IAAI,CAAC;wBACJ,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAoB,CAAC;wBAClD,IAAI,KAAK,IAAI,OAAO,KAAK,CAAC,KAAK,KAAK,QAAQ,IAAI,OAAO,KAAK,CAAC,KAAK,KAAK,QAAQ;4BAAE,OAAO,CAAC,KAAK,CAAC,CAAC;oBACjG,CAAC;oBAAC,MAAM,CAAC;wBACR,kEAAkE;oBACnE,CAAC;gBACF,CAAC;gBACD,QAAQ,GAAG,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;YACnC,CAAC;QACF,CAAC;IAAA,CACD;IAED,KAAK,CAAC,WAAW,CAChB,KAAa,EACb,UAA6C,EAC7C,MAAoB,EACM;QAC1B,MAAM,OAAO,GAAG,IAAI,eAAe,EAAE,CAAC;QACtC,MAAM,MAAM,GAAG,UAAU,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;QAC3C,IAAI,WAAW,GAAG,KAAK,CAAC;QACxB,IAAI,UAA8B,CAAC;QACnC,KAAK,IAAI,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC;YAC1B,IAAI,KAAK,CAAC,KAAK,KAAK,KAAK;gBAAE,OAAO;YAClC,IAAI,KAAK,CAAC,KAAK,KAAK,cAAc,IAAI,KAAK,CAAC,KAAK,KAAK,eAAe;gBAAE,OAAO;YAC9E,MAAM,IAAI,GAAG,KAAK,CAAC,IAAwC,CAAC;YAC5D,IAAI,IAAI,EAAE,MAAM,KAAK,QAAQ;gBAAE,WAAW,GAAG,IAAI,CAAC;YAClD,IAAI,IAAI,EAAE,MAAM,KAAK,UAAU;gBAAE,UAAU,GAAG,sBAAsB,CAAC;YACrE,MAAM,QAAQ,GAAG,iBAAiB,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;YAC/C,IAAI,QAAQ;gBAAE,UAAU,CAAC,QAAQ,CAAC,CAAC;QAAA,CACnC,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,EAAC,CAAC,CAAC,CAAC;QACnC,IAAI,CAAC;YACJ,MAAM,IAAI,CAAC,IAAI,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;YAC/B,UAAU,CAAC,EAAE,OAAO,EAAE,eAAe,EAAE,CAAC,CAAC;YACzC,OAAO,IAAI,EAAE,CAAC;gBACb,IAAI,MAAM,EAAE,OAAO;oBAAE,MAAM,MAAM,CAAC,MAAM,IAAI,IAAI,KAAK,CAAC,WAAW,CAAC,CAAC;gBACnE,MAAM,KAAK,GAAG,CAAC,MAAM,IAAI,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,EAAE,KAAK,KAAK,CAAC,CAAC;gBACxF,IAAI,KAAK,EAAE,MAAM,CAAC,KAAK,KAAK,QAAQ;oBAAE,OAAO,KAAK,CAAC;gBACnD,IAAI,WAAW,IAAI,CAAC,KAAK;oBAAE,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,EAAE,KAAK,EAAE,QAAQ,EAAE,EAAE,CAAC;gBAC7E,IAAI,KAAK,EAAE,MAAM,CAAC,MAAM,IAAI,UAAU,EAAE,CAAC;oBACxC,MAAM,IAAI,KAAK,CACd,KAAK,EAAE,MAAM,CAAC,SAAS,KAAK,SAAS;wBACpC,CAAC,CAAC,CAAC,UAAU,IAAI,sBAAsB,CAAC;wBACxC,CAAC,CAAC,0BAA0B,KAAK,CAAC,MAAM,CAAC,SAAS,EAAE,CACrD,CAAC;gBACH,CAAC;gBACD,MAAM,KAAK,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;YAC1B,CAAC;QACF,CAAC;gBAAS,CAAC;YACV,MAAM,EAAE,CAAC;YACT,OAAO,CAAC,KAAK,EAAE,CAAC;QACjB,CAAC;IAAA,CACD;IAED,KAAK,CAAC,eAAe,CACpB,KAAa,EACb,UAA6C,EAC7C,MAAoB,EACQ;QAC5B,MAAM,OAAO,GAAG,IAAI,eAAe,EAAE,CAAC;QACtC,MAAM,MAAM,GAAG,UAAU,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;QAC3C,IAAI,QAAQ,GAAG,KAAK,CAAC;QACrB,IAAI,OAA2B,CAAC;QAChC,IAAI,cAAc,GAAG,KAAK,CAAC;QAC3B,IAAI,KAAK,GAAG,CAAC,CAAC;QACd,KAAK,IAAI,CAAC,KAAK,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC;YAC1B,IAAI,KAAK,CAAC,KAAK,KAAK,KAAK;gBAAE,OAAO;YAClC,IAAI,KAAK,CAAC,KAAK,KAAK,mBAAmB;gBAAE,QAAQ,GAAG,IAAI,CAAC;YACzD,IAAI,KAAK,CAAC,KAAK,KAAK,iBAAiB;gBAAE,OAAO,GAAG,YAAY,CAAC,KAAK,CAAC,IAAI,EAAE,iBAAiB,CAAC,CAAC;YAC7F,IAAI,KAAK,CAAC,KAAK,KAAK,mBAAmB,EAAE,CAAC;gBACzC,cAAc,GAAG,IAAI,CAAC;gBACtB,MAAM,QAAQ,GAAG,qBAAqB,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;gBACnD,IAAI,QAAQ;oBAAE,UAAU,CAAC,QAAQ,CAAC,CAAC;YACpC,CAAC;QAAA,CACD,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,EAAC,CAAC,CAAC,CAAC;QACnC,IAAI,CAAC;YACJ,MAAM,IAAI,CAAC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;YACnC,UAAU,CAAC,EAAE,OAAO,EAAE,mBAAmB,EAAE,CAAC,CAAC;YAC7C,OAAO,IAAI,EAAE,CAAC;gBACb,IAAI,MAAM,EAAE,OAAO;oBAAE,MAAM,MAAM,CAAC,MAAM,IAAI,IAAI,KAAK,CAAC,WAAW,CAAC,CAAC;gBACnE,IAAI,OAAO;oBAAE,MAAM,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC;gBACtC,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC;gBAC3C,KAAK,EAAE,CAAC;gBACR,MAAM,KAAK,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,EAAE,KAAK,KAAK,CAAC,CAAC;gBACjE,IAAI,KAAK,EAAE,MAAM,CAAC,KAAK,KAAK,aAAa,EAAE,CAAC;oBAC3C,cAAc,GAAG,IAAI,CAAC;oBACtB,MAAM,QAAQ,GAAG,qBAAqB,CAAC,KAAK,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;oBAC9D,IAAI,QAAQ;wBAAE,UAAU,CAAC,QAAQ,CAAC,CAAC;gBACpC,CAAC;qBAAM,IAAI,QAAQ,IAAI,CAAC,KAAK,IAAI,CAAC,cAAc,IAAI,KAAK,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC;oBAClE,OAAO,IAAI,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC,CAAC;gBAC5C,CAAC;gBACD,MAAM,KAAK,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;YAC1B,CAAC;QACF,CAAC;gBAAS,CAAC;YACV,MAAM,EAAE,CAAC;YACT,OAAO,CAAC,KAAK,EAAE,CAAC;QACjB,CAAC;IAAA,CACD;CACD","sourcesContent":["export type LlamaModelStatus = \"unloaded\" | \"loading\" | \"loaded\" | \"downloading\" | \"sleeping\";\n\nexport interface LlamaModelInfo {\n\tid: string;\n\taliases?: string[];\n\tstatus: {\n\t\tvalue: LlamaModelStatus;\n\t\targs?: string[];\n\t\tfailed?: boolean;\n\t\texit_code?: number;\n\t\tprogress?: Record<string, { done: number; total: number }>;\n\t};\n\tarchitecture?: {\n\t\tinput_modalities?: string[];\n\t\toutput_modalities?: string[];\n\t};\n\tsource?: string;\n\tmeta?: {\n\t\tn_ctx?: number;\n\t\tn_ctx_train?: number;\n\t\tsize?: number;\n\t\tftype?: string;\n\t};\n}\n\nexport interface LlamaModelsResponse {\n\tdata: LlamaModelInfo[];\n\tobject?: string;\n}\n\nexport interface LlamaModelEvent {\n\tmodel: string;\n\tevent: string;\n\tdata?: unknown;\n}\n\nexport interface LlamaProgress {\n\tmessage: string;\n\tratio?: number;\n\tdetail?: string;\n}\n\nfunction errorMessage(payload: unknown, fallback: string): string {\n\tif (typeof payload !== \"object\" || payload === null) return fallback;\n\tconst error = (payload as { error?: unknown }).error;\n\tif (typeof error !== \"object\" || error === null) return fallback;\n\tconst message = (error as { message?: unknown }).message;\n\treturn typeof message === \"string\" && message ? message : fallback;\n}\n\nfunction isModelInfo(value: unknown): value is LlamaModelInfo {\n\tif (typeof value !== \"object\" || value === null) return false;\n\tconst candidate = value as { id?: unknown; status?: { value?: unknown } };\n\treturn typeof candidate.id === \"string\" && typeof candidate.status?.value === \"string\";\n}\n\nfunction linkSignal(source: AbortSignal | undefined, target: AbortController): () => void {\n\tif (!source) return () => {};\n\tif (source.aborted) {\n\t\ttarget.abort(source.reason);\n\t\treturn () => {};\n\t}\n\tconst abort = () => target.abort(source.reason);\n\tsource.addEventListener(\"abort\", abort, { once: true });\n\treturn () => source.removeEventListener(\"abort\", abort);\n}\n\nfunction sleep(ms: number, signal?: AbortSignal): Promise<void> {\n\treturn new Promise((resolve, reject) => {\n\t\tif (signal?.aborted) {\n\t\t\treject(signal.reason ?? new Error(\"Cancelled\"));\n\t\t\treturn;\n\t\t}\n\t\tconst abort = () => {\n\t\t\tclearTimeout(timeout);\n\t\t\treject(signal?.reason ?? new Error(\"Cancelled\"));\n\t\t};\n\t\tconst timeout = setTimeout(() => {\n\t\t\tsignal?.removeEventListener(\"abort\", abort);\n\t\t\tresolve();\n\t\t}, ms);\n\t\tsignal?.addEventListener(\"abort\", abort, { once: true });\n\t});\n}\n\nfunction parseLoadProgress(data: unknown): LlamaProgress | undefined {\n\tif (typeof data !== \"object\" || data === null) return undefined;\n\tconst progress = (data as { progress?: unknown }).progress;\n\tif (typeof progress !== \"object\" || progress === null) return undefined;\n\tconst value = progress as { stages?: unknown; current?: unknown; stage?: unknown; value?: unknown };\n\tconst stage =\n\t\ttypeof value.current === \"string\" ? value.current : typeof value.stage === \"string\" ? value.stage : undefined;\n\tconst stages = Array.isArray(value.stages)\n\t\t? value.stages.filter((entry): entry is string => typeof entry === \"string\")\n\t\t: [];\n\tconst stageRatio = typeof value.value === \"number\" ? Math.max(0, Math.min(1, value.value)) : undefined;\n\tlet ratio = stageRatio;\n\tif (stage && stages.length > 0) {\n\t\tconst index = stages.indexOf(stage);\n\t\tif (index >= 0) ratio = (index + (stageRatio ?? 0)) / stages.length;\n\t}\n\treturn {\n\t\tmessage: stage ? `Loading ${stage.replaceAll(\"_\", \" \")}` : \"Loading model\",\n\t\tratio,\n\t};\n}\n\nfunction parseDownloadProgress(data: unknown): LlamaProgress | undefined {\n\tif (typeof data !== \"object\" || data === null) return undefined;\n\tconst nested = (data as { progress?: unknown }).progress;\n\tconst files = typeof nested === \"object\" && nested !== null ? nested : data;\n\tlet done = 0;\n\tlet total = 0;\n\tfor (const value of Object.values(files as Record<string, unknown>)) {\n\t\tif (typeof value !== \"object\" || value === null) continue;\n\t\tconst entry = value as { done?: unknown; total?: unknown };\n\t\tif (typeof entry.done !== \"number\" || typeof entry.total !== \"number\") continue;\n\t\tdone += entry.done;\n\t\ttotal += entry.total;\n\t}\n\tif (total <= 0) return undefined;\n\treturn {\n\t\tmessage: \"Downloading model\",\n\t\tratio: done / total,\n\t\tdetail: `${formatBytes(done)} / ${formatBytes(total)}`,\n\t};\n}\n\nexport function formatBytes(bytes: number): string {\n\tif (bytes < 1024) return `${bytes} B`;\n\tconst units = [\"KiB\", \"MiB\", \"GiB\", \"TiB\"];\n\tlet value = bytes / 1024;\n\tlet unit = units[0]!;\n\tfor (let index = 1; index < units.length && value >= 1024; index++) {\n\t\tvalue /= 1024;\n\t\tunit = units[index]!;\n\t}\n\treturn `${value >= 10 ? value.toFixed(1) : value.toFixed(2)} ${unit}`;\n}\n\nexport function normalizeLlamaServerUrl(value: string): string {\n\tconst url = new URL(value.trim());\n\tif (url.protocol !== \"http:\" && url.protocol !== \"https:\") {\n\t\tthrow new Error(\"Server URL must use http or https\");\n\t}\n\turl.hash = \"\";\n\turl.search = \"\";\n\turl.pathname = url.pathname.replace(/\\/+$/u, \"\").replace(/\\/v1$/u, \"\") || \"/\";\n\treturn url.toString().replace(/\\/$/u, \"\");\n}\n\nexport function llamaInferenceUrl(serverUrl: string): string {\n\treturn `${normalizeLlamaServerUrl(serverUrl)}/v1`;\n}\n\nexport class LlamaClient {\n\treadonly serverUrl: string;\n\tprivate readonly apiKey: string | undefined;\n\n\tconstructor(serverUrl: string, apiKey?: string) {\n\t\tthis.serverUrl = normalizeLlamaServerUrl(serverUrl);\n\t\tthis.apiKey = apiKey;\n\t}\n\n\tprivate async request(path: string, init: RequestInit = {}): Promise<unknown> {\n\t\tconst headers = new Headers(init.headers);\n\t\tif (init.body !== undefined) headers.set(\"Content-Type\", \"application/json\");\n\t\tif (this.apiKey) headers.set(\"Authorization\", `Bearer ${this.apiKey}`);\n\t\tconst timeout = AbortSignal.timeout(15_000);\n\t\tconst signal = init.signal ? AbortSignal.any([init.signal, timeout]) : timeout;\n\t\tconst response = await fetch(`${this.serverUrl}${path}`, { ...init, headers, signal });\n\t\tlet payload: unknown;\n\t\ttry {\n\t\t\tpayload = await response.json();\n\t\t} catch {\n\t\t\tpayload = undefined;\n\t\t}\n\t\tif (!response.ok) throw new Error(errorMessage(payload, `llama.cpp returned HTTP ${response.status}`));\n\t\treturn payload;\n\t}\n\n\tasync list(options: { reload?: boolean; signal?: AbortSignal } = {}): Promise<LlamaModelInfo[]> {\n\t\tconst payload = await this.request(`/models${options.reload ? \"?reload=1\" : \"\"}`, { signal: options.signal });\n\t\tif (typeof payload !== \"object\" || payload === null || !Array.isArray((payload as { data?: unknown }).data)) {\n\t\t\tthrow new Error(\"llama.cpp returned an invalid model catalog\");\n\t\t}\n\t\tconst data = (payload as { data: unknown[] }).data;\n\t\tif (!data.every(isModelInfo)) throw new Error(\"Server is not running in llama.cpp router mode\");\n\t\treturn data;\n\t}\n\n\tasync load(model: string, signal?: AbortSignal): Promise<void> {\n\t\tawait this.request(\"/models/load\", { method: \"POST\", body: JSON.stringify({ model }), signal });\n\t}\n\n\tasync unload(model: string, signal?: AbortSignal): Promise<void> {\n\t\tawait this.request(\"/models/unload\", { method: \"POST\", body: JSON.stringify({ model }), signal });\n\t}\n\n\tasync unloadAndWait(model: string, signal?: AbortSignal): Promise<void> {\n\t\tawait this.unload(model, signal);\n\t\twhile (true) {\n\t\t\tconst entry = (await this.list({ signal })).find((candidate) => candidate.id === model);\n\t\t\tif (!entry || entry.status.value === \"unloaded\") return;\n\t\t\tawait sleep(100, signal);\n\t\t}\n\t}\n\n\tasync download(model: string, signal?: AbortSignal): Promise<void> {\n\t\tawait this.request(\"/models\", { method: \"POST\", body: JSON.stringify({ model }), signal });\n\t}\n\n\tasync watch(onEvent: (event: LlamaModelEvent) => void, signal?: AbortSignal): Promise<void> {\n\t\tconst headers = new Headers();\n\t\tif (this.apiKey) headers.set(\"Authorization\", `Bearer ${this.apiKey}`);\n\t\tconst response = await fetch(`${this.serverUrl}/models/sse`, { headers, signal });\n\t\tif (!response.ok || !response.body) throw new Error(`llama.cpp SSE returned HTTP ${response.status}`);\n\t\tconst reader = response.body.getReader();\n\t\tconst decoder = new TextDecoder();\n\t\tlet buffer = \"\";\n\t\twhile (true) {\n\t\t\tconst chunk = await reader.read();\n\t\t\tif (chunk.done) break;\n\t\t\tbuffer += decoder.decode(chunk.value, { stream: true }).replaceAll(\"\\r\\n\", \"\\n\");\n\t\t\tlet boundary = buffer.indexOf(\"\\n\\n\");\n\t\t\twhile (boundary >= 0) {\n\t\t\t\tconst frame = buffer.slice(0, boundary);\n\t\t\t\tbuffer = buffer.slice(boundary + 2);\n\t\t\t\tconst data = frame\n\t\t\t\t\t.split(\"\\n\")\n\t\t\t\t\t.filter((line) => line.startsWith(\"data:\"))\n\t\t\t\t\t.map((line) => line.slice(5).trimStart())\n\t\t\t\t\t.join(\"\\n\");\n\t\t\t\tif (data) {\n\t\t\t\t\ttry {\n\t\t\t\t\t\tconst event = JSON.parse(data) as LlamaModelEvent;\n\t\t\t\t\t\tif (event && typeof event.model === \"string\" && typeof event.event === \"string\") onEvent(event);\n\t\t\t\t\t} catch {\n\t\t\t\t\t\t// Ignore malformed events; catalog polling remains authoritative.\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tboundary = buffer.indexOf(\"\\n\\n\");\n\t\t\t}\n\t\t}\n\t}\n\n\tasync loadAndWait(\n\t\tmodel: string,\n\t\tonProgress: (progress: LlamaProgress) => void,\n\t\tsignal?: AbortSignal,\n\t): Promise<LlamaModelInfo> {\n\t\tconst watcher = new AbortController();\n\t\tconst unlink = linkSignal(signal, watcher);\n\t\tlet eventLoaded = false;\n\t\tlet eventError: string | undefined;\n\t\tvoid this.watch((event) => {\n\t\t\tif (event.model !== model) return;\n\t\t\tif (event.event !== \"model_status\" && event.event !== \"status_change\") return;\n\t\t\tconst data = event.data as { status?: unknown } | undefined;\n\t\t\tif (data?.status === \"loaded\") eventLoaded = true;\n\t\t\tif (data?.status === \"unloaded\") eventError = \"Model failed to load\";\n\t\t\tconst progress = parseLoadProgress(event.data);\n\t\t\tif (progress) onProgress(progress);\n\t\t}, watcher.signal).catch(() => {});\n\t\ttry {\n\t\t\tawait this.load(model, signal);\n\t\t\tonProgress({ message: \"Loading model\" });\n\t\t\twhile (true) {\n\t\t\t\tif (signal?.aborted) throw signal.reason ?? new Error(\"Cancelled\");\n\t\t\t\tconst entry = (await this.list({ signal })).find((candidate) => candidate.id === model);\n\t\t\t\tif (entry?.status.value === \"loaded\") return entry;\n\t\t\t\tif (eventLoaded && !entry) return { id: model, status: { value: \"loaded\" } };\n\t\t\t\tif (entry?.status.failed || eventError) {\n\t\t\t\t\tthrow new Error(\n\t\t\t\t\t\tentry?.status.exit_code === undefined\n\t\t\t\t\t\t\t? (eventError ?? \"Model failed to load\")\n\t\t\t\t\t\t\t: `Model exited with code ${entry.status.exit_code}`,\n\t\t\t\t\t);\n\t\t\t\t}\n\t\t\t\tawait sleep(250, signal);\n\t\t\t}\n\t\t} finally {\n\t\t\tunlink();\n\t\t\twatcher.abort();\n\t\t}\n\t}\n\n\tasync downloadAndWait(\n\t\tmodel: string,\n\t\tonProgress: (progress: LlamaProgress) => void,\n\t\tsignal?: AbortSignal,\n\t): Promise<LlamaModelInfo[]> {\n\t\tconst watcher = new AbortController();\n\t\tconst unlink = linkSignal(signal, watcher);\n\t\tlet finished = false;\n\t\tlet failure: string | undefined;\n\t\tlet sawDownloading = false;\n\t\tlet polls = 0;\n\t\tvoid this.watch((event) => {\n\t\t\tif (event.model !== model) return;\n\t\t\tif (event.event === \"download_finished\") finished = true;\n\t\t\tif (event.event === \"download_failed\") failure = errorMessage(event.data, \"Download failed\");\n\t\t\tif (event.event === \"download_progress\") {\n\t\t\t\tsawDownloading = true;\n\t\t\t\tconst progress = parseDownloadProgress(event.data);\n\t\t\t\tif (progress) onProgress(progress);\n\t\t\t}\n\t\t}, watcher.signal).catch(() => {});\n\t\ttry {\n\t\t\tawait this.download(model, signal);\n\t\t\tonProgress({ message: \"Downloading model\" });\n\t\t\twhile (true) {\n\t\t\t\tif (signal?.aborted) throw signal.reason ?? new Error(\"Cancelled\");\n\t\t\t\tif (failure) throw new Error(failure);\n\t\t\t\tconst models = await this.list({ signal });\n\t\t\t\tpolls++;\n\t\t\t\tconst entry = models.find((candidate) => candidate.id === model);\n\t\t\t\tif (entry?.status.value === \"downloading\") {\n\t\t\t\t\tsawDownloading = true;\n\t\t\t\t\tconst progress = parseDownloadProgress(entry.status.progress);\n\t\t\t\t\tif (progress) onProgress(progress);\n\t\t\t\t} else if (finished || (entry && (sawDownloading || polls >= 2))) {\n\t\t\t\t\treturn this.list({ reload: true, signal });\n\t\t\t\t}\n\t\t\t\tawait sleep(500, signal);\n\t\t\t}\n\t\t} finally {\n\t\t\tunlink();\n\t\t\twatcher.abort();\n\t\t}\n\t}\n}\n"]}

package/dist/extensions/llama/huggingface.d.ts

@@ -0,0 +1,23 @@
+export interface HuggingFaceModel {
+    id: string;
+    downloads: number;
+}
+export interface HuggingFaceQuantization {
+    name: string;
+    size?: number;
+}
+export interface HuggingFaceModelDetails {
+    id: string;
+    gated: false | "auto" | "manual";
+    quantizations: HuggingFaceQuantization[];
+}
+export declare function findHuggingFaceToken(env?: NodeJS.ProcessEnv): Promise<string | undefined>;
+export declare class HuggingFaceClient {
+    private readonly token;
+    private readonly baseUrl;
+    constructor(token?: string, baseUrl?: string);
+    private request;
+    search(query: string, signal?: AbortSignal): Promise<HuggingFaceModel[]>;
+    details(id: string, signal?: AbortSignal): Promise<HuggingFaceModelDetails>;
+}
+//# sourceMappingURL=huggingface.d.ts.map

package/dist/extensions/llama/huggingface.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"huggingface.d.ts","sourceRoot":"","sources":["../../../src/extensions/llama/huggingface.ts"],"names":[],"mappings":"AASA,MAAM,WAAW,gBAAgB;IAChC,EAAE,EAAE,MAAM,CAAC;IACX,SAAS,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,uBAAuB;IACvC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,CAAC,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,uBAAuB;IACvC,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,KAAK,GAAG,MAAM,GAAG,QAAQ,CAAC;IACjC,aAAa,EAAE,uBAAuB,EAAE,CAAC;CACzC;AAsBD,wBAAsB,oBAAoB,CAAC,GAAG,GAAE,MAAM,CAAC,UAAwB,GAAG,OAAO,CAAC,MAAM,GAAG,SAAS,CAAC,CAe5G;AAED,qBAAa,iBAAiB;IAC7B,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAqB;IAC3C,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAS;IAEjC,YAAY,KAAK,CAAC,EAAE,MAAM,EAAE,OAAO,SAA2B,EAG7D;YAEa,OAAO;IA4Bf,MAAM,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,WAAW,GAAG,OAAO,CAAC,gBAAgB,EAAE,CAAC,CAgB7E;IAEK,OAAO,CAAC,EAAE,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,WAAW,GAAG,OAAO,CAAC,uBAAuB,CAAC,CAuChF;CACD","sourcesContent":["import { readFile } from \"node:fs/promises\";\nimport { homedir } from \"node:os\";\nimport { join } from \"node:path\";\n\nconst DEFAULT_HUGGING_FACE_URL = \"https://huggingface.co\";\nconst QUANTIZATION_PATTERN =\n\t/(?:^|[-_.])((?:UD-)?(?:IQ\\d(?:_[A-Z0-9]+)+|Q\\d(?:_[A-Z0-9]+)+|BF16|F16|F32|MXFP\\d(?:_[A-Z0-9]+)*))$/iu;\nconst SHARD_SUFFIX_PATTERN = /-\\d{5}-of-\\d{5}$/u;\n\nexport interface HuggingFaceModel {\n\tid: string;\n\tdownloads: number;\n}\n\nexport interface HuggingFaceQuantization {\n\tname: string;\n\tsize?: number;\n}\n\nexport interface HuggingFaceModelDetails {\n\tid: string;\n\tgated: false | \"auto\" | \"manual\";\n\tquantizations: HuggingFaceQuantization[];\n}\n\nfunction payloadError(payload: unknown, fallback: string): string {\n\tif (typeof payload !== \"object\" || payload === null) return fallback;\n\tconst error = (payload as { error?: unknown }).error;\n\treturn typeof error === \"string\" && error ? error : fallback;\n}\n\nfunction parseRateLimitDelay(value: string | null): number | undefined {\n\tconst match = value?.match(/(?:^|;)t=(\\d+)/u);\n\treturn match ? Number(match[1]) : undefined;\n}\n\nasync function readToken(path: string): Promise<string | undefined> {\n\ttry {\n\t\tconst token = (await readFile(path, \"utf8\")).trim();\n\t\treturn token || undefined;\n\t} catch {\n\t\treturn undefined;\n\t}\n}\n\nexport async function findHuggingFaceToken(env: NodeJS.ProcessEnv = process.env): Promise<string | undefined> {\n\tconst fromEnvironment = env.HF_TOKEN?.trim();\n\tif (fromEnvironment) return fromEnvironment;\n\n\tconst paths = [\n\t\tenv.HF_TOKEN_PATH,\n\t\tenv.HF_HOME ? join(env.HF_HOME, \"token\") : undefined,\n\t\tenv.XDG_CACHE_HOME ? join(env.XDG_CACHE_HOME, \"huggingface\", \"token\") : undefined,\n\t\tjoin(homedir(), \".cache\", \"huggingface\", \"token\"),\n\t].filter((path): path is string => Boolean(path));\n\tfor (const path of new Set(paths)) {\n\t\tconst token = await readToken(path);\n\t\tif (token) return token;\n\t}\n\treturn undefined;\n}\n\nexport class HuggingFaceClient {\n\tprivate readonly token: string | undefined;\n\tprivate readonly baseUrl: string;\n\n\tconstructor(token?: string, baseUrl = DEFAULT_HUGGING_FACE_URL) {\n\t\tthis.token = token;\n\t\tthis.baseUrl = baseUrl.replace(/\\/+$/u, \"\");\n\t}\n\n\tprivate async request(path: string, signal?: AbortSignal): Promise<unknown> {\n\t\tconst headers = new Headers();\n\t\tif (this.token) headers.set(\"Authorization\", `Bearer ${this.token}`);\n\t\tconst timeout = AbortSignal.timeout(15_000);\n\t\tconst response = await fetch(`${this.baseUrl}${path}`, {\n\t\t\theaders,\n\t\t\tsignal: signal ? AbortSignal.any([signal, timeout]) : timeout,\n\t\t});\n\t\tlet payload: unknown;\n\t\ttry {\n\t\t\tpayload = await response.json();\n\t\t} catch {\n\t\t\tpayload = undefined;\n\t\t}\n\t\tif (!response.ok) {\n\t\t\tconst fallback = `Hugging Face returned HTTP ${response.status}`;\n\t\t\tif (response.status === 429) {\n\t\t\t\tconst delay =\n\t\t\t\t\tNumber(response.headers.get(\"retry-after\")) || parseRateLimitDelay(response.headers.get(\"ratelimit\"));\n\t\t\t\tthrow new Error(\n\t\t\t\t\tdelay ? `Hugging Face rate limit reached; retry in ${delay}s` : \"Hugging Face rate limit reached\",\n\t\t\t\t);\n\t\t\t}\n\t\t\tthrow new Error(payloadError(payload, fallback));\n\t\t}\n\t\treturn payload;\n\t}\n\n\tasync search(query: string, signal?: AbortSignal): Promise<HuggingFaceModel[]> {\n\t\tconst params = new URLSearchParams({\n\t\t\tsearch: query,\n\t\t\tfilter: \"gguf\",\n\t\t\tsort: \"downloads\",\n\t\t\tdirection: \"-1\",\n\t\t\tlimit: \"20\",\n\t\t});\n\t\tconst payload = await this.request(`/api/models?${params}`, signal);\n\t\tif (!Array.isArray(payload)) throw new Error(\"Hugging Face returned invalid search results\");\n\t\treturn payload.flatMap((value) => {\n\t\t\tif (typeof value !== \"object\" || value === null || typeof (value as { id?: unknown }).id !== \"string\")\n\t\t\t\treturn [];\n\t\t\tconst model = value as { id: string; downloads?: unknown };\n\t\t\treturn [{ id: model.id, downloads: typeof model.downloads === \"number\" ? model.downloads : 0 }];\n\t\t});\n\t}\n\n\tasync details(id: string, signal?: AbortSignal): Promise<HuggingFaceModelDetails> {\n\t\tconst encodedId = id.split(\"/\").map(encodeURIComponent).join(\"/\");\n\t\tconst payload = await this.request(`/api/models/${encodedId}?blobs=true`, signal);\n\t\tif (typeof payload !== \"object\" || payload === null) {\n\t\t\tthrow new Error(\"Hugging Face returned invalid model details\");\n\t\t}\n\t\tconst model = payload as { id?: unknown; gated?: unknown; siblings?: unknown };\n\t\tconst sizes = new Map<string, { total: number; complete: boolean }>();\n\t\tif (Array.isArray(model.siblings)) {\n\t\t\tfor (const value of model.siblings) {\n\t\t\t\tif (typeof value !== \"object\" || value === null) continue;\n\t\t\t\tconst file = value as { rfilename?: unknown; size?: unknown };\n\t\t\t\tif (typeof file.rfilename !== \"string\" || !file.rfilename.toLowerCase().endsWith(\".gguf\")) continue;\n\t\t\t\tconst filename = file.rfilename.split(\"/\").at(-1)!;\n\t\t\t\tif (filename.toLowerCase().startsWith(\"mmproj\")) continue;\n\t\t\t\tconst stem = filename.slice(0, -5).replace(SHARD_SUFFIX_PATTERN, \"\");\n\t\t\t\tconst quantization = stem.match(QUANTIZATION_PATTERN)?.[1]?.toUpperCase();\n\t\t\t\tif (!quantization) continue;\n\t\t\t\tconst current = sizes.get(quantization) ?? { total: 0, complete: true };\n\t\t\t\tif (typeof file.size === \"number\") current.total += file.size;\n\t\t\t\telse current.complete = false;\n\t\t\t\tsizes.set(quantization, current);\n\t\t\t}\n\t\t}\n\t\tconst quantizations = [...sizes]\n\t\t\t.map(([name, size]) => ({ name, size: size.complete ? size.total : undefined }))\n\t\t\t.sort((left, right) => {\n\t\t\t\tif (left.name === \"Q4_K_M\") return -1;\n\t\t\t\tif (right.name === \"Q4_K_M\") return 1;\n\t\t\t\treturn (\n\t\t\t\t\t(left.size ?? Number.MAX_SAFE_INTEGER) - (right.size ?? Number.MAX_SAFE_INTEGER) ||\n\t\t\t\t\tleft.name.localeCompare(right.name)\n\t\t\t\t);\n\t\t\t});\n\t\treturn {\n\t\t\tid: typeof model.id === \"string\" ? model.id : id,\n\t\t\tgated: model.gated === \"auto\" || model.gated === \"manual\" ? model.gated : false,\n\t\t\tquantizations,\n\t\t};\n\t}\n}\n"]}

package/dist/extensions/llama/huggingface.js

@@ -0,0 +1,141 @@
+import { readFile } from "node:fs/promises";
+import { homedir } from "node:os";
+import { join } from "node:path";
+const DEFAULT_HUGGING_FACE_URL = "https://huggingface.co";
+const QUANTIZATION_PATTERN = /(?:^|[-_.])((?:UD-)?(?:IQ\d(?:_[A-Z0-9]+)+|Q\d(?:_[A-Z0-9]+)+|BF16|F16|F32|MXFP\d(?:_[A-Z0-9]+)*))$/iu;
+const SHARD_SUFFIX_PATTERN = /-\d{5}-of-\d{5}$/u;
+function payloadError(payload, fallback) {
+    if (typeof payload !== "object" || payload === null)
+        return fallback;
+    const error = payload.error;
+    return typeof error === "string" && error ? error : fallback;
+}
+function parseRateLimitDelay(value) {
+    const match = value?.match(/(?:^|;)t=(\d+)/u);
+    return match ? Number(match[1]) : undefined;
+}
+async function readToken(path) {
+    try {
+        const token = (await readFile(path, "utf8")).trim();
+        return token || undefined;
+    }
+    catch {
+        return undefined;
+    }
+}
+export async function findHuggingFaceToken(env = process.env) {
+    const fromEnvironment = env.HF_TOKEN?.trim();
+    if (fromEnvironment)
+        return fromEnvironment;
+    const paths = [
+        env.HF_TOKEN_PATH,
+        env.HF_HOME ? join(env.HF_HOME, "token") : undefined,
+        env.XDG_CACHE_HOME ? join(env.XDG_CACHE_HOME, "huggingface", "token") : undefined,
+        join(homedir(), ".cache", "huggingface", "token"),
+    ].filter((path) => Boolean(path));
+    for (const path of new Set(paths)) {
+        const token = await readToken(path);
+        if (token)
+            return token;
+    }
+    return undefined;
+}
+export class HuggingFaceClient {
+    token;
+    baseUrl;
+    constructor(token, baseUrl = DEFAULT_HUGGING_FACE_URL) {
+        this.token = token;
+        this.baseUrl = baseUrl.replace(/\/+$/u, "");
+    }
+    async request(path, signal) {
+        const headers = new Headers();
+        if (this.token)
+            headers.set("Authorization", `Bearer ${this.token}`);
+        const timeout = AbortSignal.timeout(15_000);
+        const response = await fetch(`${this.baseUrl}${path}`, {
+            headers,
+            signal: signal ? AbortSignal.any([signal, timeout]) : timeout,
+        });
+        let payload;
+        try {
+            payload = await response.json();
+        }
+        catch {
+            payload = undefined;
+        }
+        if (!response.ok) {
+            const fallback = `Hugging Face returned HTTP ${response.status}`;
+            if (response.status === 429) {
+                const delay = Number(response.headers.get("retry-after")) || parseRateLimitDelay(response.headers.get("ratelimit"));
+                throw new Error(delay ? `Hugging Face rate limit reached; retry in ${delay}s` : "Hugging Face rate limit reached");
+            }
+            throw new Error(payloadError(payload, fallback));
+        }
+        return payload;
+    }
+    async search(query, signal) {
+        const params = new URLSearchParams({
+            search: query,
+            filter: "gguf",
+            sort: "downloads",
+            direction: "-1",
+            limit: "20",
+        });
+        const payload = await this.request(`/api/models?${params}`, signal);
+        if (!Array.isArray(payload))
+            throw new Error("Hugging Face returned invalid search results");
+        return payload.flatMap((value) => {
+            if (typeof value !== "object" || value === null || typeof value.id !== "string")
+                return [];
+            const model = value;
+            return [{ id: model.id, downloads: typeof model.downloads === "number" ? model.downloads : 0 }];
+        });
+    }
+    async details(id, signal) {
+        const encodedId = id.split("/").map(encodeURIComponent).join("/");
+        const payload = await this.request(`/api/models/${encodedId}?blobs=true`, signal);
+        if (typeof payload !== "object" || payload === null) {
+            throw new Error("Hugging Face returned invalid model details");
+        }
+        const model = payload;
+        const sizes = new Map();
+        if (Array.isArray(model.siblings)) {
+            for (const value of model.siblings) {
+                if (typeof value !== "object" || value === null)
+                    continue;
+                const file = value;
+                if (typeof file.rfilename !== "string" || !file.rfilename.toLowerCase().endsWith(".gguf"))
+                    continue;
+                const filename = file.rfilename.split("/").at(-1);
+                if (filename.toLowerCase().startsWith("mmproj"))
+                    continue;
+                const stem = filename.slice(0, -5).replace(SHARD_SUFFIX_PATTERN, "");
+                const quantization = stem.match(QUANTIZATION_PATTERN)?.[1]?.toUpperCase();
+                if (!quantization)
+                    continue;
+                const current = sizes.get(quantization) ?? { total: 0, complete: true };
+                if (typeof file.size === "number")
+                    current.total += file.size;
+                else
+                    current.complete = false;
+                sizes.set(quantization, current);
+            }
+        }
+        const quantizations = [...sizes]
+            .map(([name, size]) => ({ name, size: size.complete ? size.total : undefined }))
+            .sort((left, right) => {
+            if (left.name === "Q4_K_M")
+                return -1;
+            if (right.name === "Q4_K_M")
+                return 1;
+            return ((left.size ?? Number.MAX_SAFE_INTEGER) - (right.size ?? Number.MAX_SAFE_INTEGER) ||
+                left.name.localeCompare(right.name));
+        });
+        return {
+            id: typeof model.id === "string" ? model.id : id,
+            gated: model.gated === "auto" || model.gated === "manual" ? model.gated : false,
+            quantizations,
+        };
+    }
+}
+//# sourceMappingURL=huggingface.js.map

package/dist/extensions/llama/huggingface.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"huggingface.js","sourceRoot":"","sources":["../../../src/extensions/llama/huggingface.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAC;AAC5C,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAClC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAEjC,MAAM,wBAAwB,GAAG,wBAAwB,CAAC;AAC1D,MAAM,oBAAoB,GACzB,uGAAuG,CAAC;AACzG,MAAM,oBAAoB,GAAG,mBAAmB,CAAC;AAkBjD,SAAS,YAAY,CAAC,OAAgB,EAAE,QAAgB,EAAU;IACjE,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,IAAI;QAAE,OAAO,QAAQ,CAAC;IACrE,MAAM,KAAK,GAAI,OAA+B,CAAC,KAAK,CAAC;IACrD,OAAO,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC;AAAA,CAC7D;AAED,SAAS,mBAAmB,CAAC,KAAoB,EAAsB;IACtE,MAAM,KAAK,GAAG,KAAK,EAAE,KAAK,CAAC,iBAAiB,CAAC,CAAC;IAC9C,OAAO,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAAA,CAC5C;AAED,KAAK,UAAU,SAAS,CAAC,IAAY,EAA+B;IACnE,IAAI,CAAC;QACJ,MAAM,KAAK,GAAG,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QACpD,OAAO,KAAK,IAAI,SAAS,CAAC;IAC3B,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,SAAS,CAAC;IAClB,CAAC;AAAA,CACD;AAED,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,GAAG,GAAsB,OAAO,CAAC,GAAG,EAA+B;IAC7G,MAAM,eAAe,GAAG,GAAG,CAAC,QAAQ,EAAE,IAAI,EAAE,CAAC;IAC7C,IAAI,eAAe;QAAE,OAAO,eAAe,CAAC;IAE5C,MAAM,KAAK,GAAG;QACb,GAAG,CAAC,aAAa;QACjB,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS;QACpD,GAAG,CAAC,cAAc,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,cAAc,EAAE,aAAa,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,SAAS;QACjF,IAAI,CAAC,OAAO,EAAE,EAAE,QAAQ,EAAE,aAAa,EAAE,OAAO,CAAC;KACjD,CAAC,MAAM,CAAC,CAAC,IAAI,EAAkB,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC;IAClD,KAAK,MAAM,IAAI,IAAI,IAAI,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC;QACnC,MAAM,KAAK,GAAG,MAAM,SAAS,CAAC,IAAI,CAAC,CAAC;QACpC,IAAI,KAAK;YAAE,OAAO,KAAK,CAAC;IACzB,CAAC;IACD,OAAO,SAAS,CAAC;AAAA,CACjB;AAED,MAAM,OAAO,iBAAiB;IACZ,KAAK,CAAqB;IAC1B,OAAO,CAAS;IAEjC,YAAY,KAAc,EAAE,OAAO,GAAG,wBAAwB,EAAE;QAC/D,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;IAAA,CAC5C;IAEO,KAAK,CAAC,OAAO,CAAC,IAAY,EAAE,MAAoB,EAAoB;QAC3E,MAAM,OAAO,GAAG,IAAI,OAAO,EAAE,CAAC;QAC9B,IAAI,IAAI,CAAC,KAAK;YAAE,OAAO,CAAC,GAAG,CAAC,eAAe,EAAE,UAAU,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;QACrE,MAAM,OAAO,GAAG,WAAW,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;QAC5C,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,IAAI,CAAC,OAAO,GAAG,IAAI,EAAE,EAAE;YACtD,OAAO;YACP,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO;SAC7D,CAAC,CAAC;QACH,IAAI,OAAgB,CAAC;QACrB,IAAI,CAAC;YACJ,OAAO,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;QACjC,CAAC;QAAC,MAAM,CAAC;YACR,OAAO,GAAG,SAAS,CAAC;QACrB,CAAC;QACD,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;YAClB,MAAM,QAAQ,GAAG,8BAA8B,QAAQ,CAAC,MAAM,EAAE,CAAC;YACjE,IAAI,QAAQ,CAAC,MAAM,KAAK,GAAG,EAAE,CAAC;gBAC7B,MAAM,KAAK,GACV,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC,IAAI,mBAAmB,CAAC,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC;gBACvG,MAAM,IAAI,KAAK,CACd,KAAK,CAAC,CAAC,CAAC,6CAA6C,KAAK,GAAG,CAAC,CAAC,CAAC,iCAAiC,CACjG,CAAC;YACH,CAAC;YACD,MAAM,IAAI,KAAK,CAAC,YAAY,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC,CAAC;QAClD,CAAC;QACD,OAAO,OAAO,CAAC;IAAA,CACf;IAED,KAAK,CAAC,MAAM,CAAC,KAAa,EAAE,MAAoB,EAA+B;QAC9E,MAAM,MAAM,GAAG,IAAI,eAAe,CAAC;YAClC,MAAM,EAAE,KAAK;YACb,MAAM,EAAE,MAAM;YACd,IAAI,EAAE,WAAW;YACjB,SAAS,EAAE,IAAI;YACf,KAAK,EAAE,IAAI;SACX,CAAC,CAAC;QACH,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,eAAe,MAAM,EAAE,EAAE,MAAM,CAAC,CAAC;QACpE,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,8CAA8C,CAAC,CAAC;QAC7F,OAAO,OAAO,CAAC,OAAO,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC;YACjC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,IAAI,OAAQ,KAA0B,CAAC,EAAE,KAAK,QAAQ;gBACpG,OAAO,EAAE,CAAC;YACX,MAAM,KAAK,GAAG,KAA4C,CAAC;YAC3D,OAAO,CAAC,EAAE,EAAE,EAAE,KAAK,CAAC,EAAE,EAAE,SAAS,EAAE,OAAO,KAAK,CAAC,SAAS,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QAAA,CAChG,CAAC,CAAC;IAAA,CACH;IAED,KAAK,CAAC,OAAO,CAAC,EAAU,EAAE,MAAoB,EAAoC;QACjF,MAAM,SAAS,GAAG,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,kBAAkB,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAClE,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,eAAe,SAAS,aAAa,EAAE,MAAM,CAAC,CAAC;QAClF,IAAI,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,IAAI,EAAE,CAAC;YACrD,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAC;QAChE,CAAC;QACD,MAAM,KAAK,GAAG,OAAgE,CAAC;QAC/E,MAAM,KAAK,GAAG,IAAI,GAAG,EAAgD,CAAC;QACtE,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,QAAQ,CAAC,EAAE,CAAC;YACnC,KAAK,MAAM,KAAK,IAAI,KAAK,CAAC,QAAQ,EAAE,CAAC;gBACpC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI;oBAAE,SAAS;gBAC1D,MAAM,IAAI,GAAG,KAAgD,CAAC;gBAC9D,IAAI,OAAO,IAAI,CAAC,SAAS,KAAK,QAAQ,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,OAAO,CAAC;oBAAE,SAAS;gBACpG,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAE,CAAC;gBACnD,IAAI,QAAQ,CAAC,WAAW,EAAE,CAAC,UAAU,CAAC,QAAQ,CAAC;oBAAE,SAAS;gBAC1D,MAAM,IAAI,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,oBAAoB,EAAE,EAAE,CAAC,CAAC;gBACrE,MAAM,YAAY,GAAG,IAAI,CAAC,KAAK,CAAC,oBAAoB,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC;gBAC1E,IAAI,CAAC,YAAY;oBAAE,SAAS;gBAC5B,MAAM,OAAO,GAAG,KAAK,CAAC,GAAG,CAAC,YAAY,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;gBACxE,IAAI,OAAO,IAAI,CAAC,IAAI,KAAK,QAAQ;oBAAE,OAAO,CAAC,KAAK,IAAI,IAAI,CAAC,IAAI,CAAC;;oBACzD,OAAO,CAAC,QAAQ,GAAG,KAAK,CAAC;gBAC9B,KAAK,CAAC,GAAG,CAAC,YAAY,EAAE,OAAO,CAAC,CAAC;YAClC,CAAC;QACF,CAAC;QACD,MAAM,aAAa,GAAG,CAAC,GAAG,KAAK,CAAC;aAC9B,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC;aAC/E,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC;YACtB,IAAI,IAAI,CAAC,IAAI,KAAK,QAAQ;gBAAE,OAAO,CAAC,CAAC,CAAC;YACtC,IAAI,KAAK,CAAC,IAAI,KAAK,QAAQ;gBAAE,OAAO,CAAC,CAAC;YACtC,OAAO,CACN,CAAC,IAAI,CAAC,IAAI,IAAI,MAAM,CAAC,gBAAgB,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,IAAI,MAAM,CAAC,gBAAgB,CAAC;gBAChF,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CACnC,CAAC;QAAA,CACF,CAAC,CAAC;QACJ,OAAO;YACN,EAAE,EAAE,OAAO,KAAK,CAAC,EAAE,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE;YAChD,KAAK,EAAE,KAAK,CAAC,KAAK,KAAK,MAAM,IAAI,KAAK,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,KAAK;YAC/E,aAAa;SACb,CAAC;IAAA,CACF;CACD","sourcesContent":["import { readFile } from \"node:fs/promises\";\nimport { homedir } from \"node:os\";\nimport { join } from \"node:path\";\n\nconst DEFAULT_HUGGING_FACE_URL = \"https://huggingface.co\";\nconst QUANTIZATION_PATTERN =\n\t/(?:^|[-_.])((?:UD-)?(?:IQ\\d(?:_[A-Z0-9]+)+|Q\\d(?:_[A-Z0-9]+)+|BF16|F16|F32|MXFP\\d(?:_[A-Z0-9]+)*))$/iu;\nconst SHARD_SUFFIX_PATTERN = /-\\d{5}-of-\\d{5}$/u;\n\nexport interface HuggingFaceModel {\n\tid: string;\n\tdownloads: number;\n}\n\nexport interface HuggingFaceQuantization {\n\tname: string;\n\tsize?: number;\n}\n\nexport interface HuggingFaceModelDetails {\n\tid: string;\n\tgated: false | \"auto\" | \"manual\";\n\tquantizations: HuggingFaceQuantization[];\n}\n\nfunction payloadError(payload: unknown, fallback: string): string {\n\tif (typeof payload !== \"object\" || payload === null) return fallback;\n\tconst error = (payload as { error?: unknown }).error;\n\treturn typeof error === \"string\" && error ? error : fallback;\n}\n\nfunction parseRateLimitDelay(value: string | null): number | undefined {\n\tconst match = value?.match(/(?:^|;)t=(\\d+)/u);\n\treturn match ? Number(match[1]) : undefined;\n}\n\nasync function readToken(path: string): Promise<string | undefined> {\n\ttry {\n\t\tconst token = (await readFile(path, \"utf8\")).trim();\n\t\treturn token || undefined;\n\t} catch {\n\t\treturn undefined;\n\t}\n}\n\nexport async function findHuggingFaceToken(env: NodeJS.ProcessEnv = process.env): Promise<string | undefined> {\n\tconst fromEnvironment = env.HF_TOKEN?.trim();\n\tif (fromEnvironment) return fromEnvironment;\n\n\tconst paths = [\n\t\tenv.HF_TOKEN_PATH,\n\t\tenv.HF_HOME ? join(env.HF_HOME, \"token\") : undefined,\n\t\tenv.XDG_CACHE_HOME ? join(env.XDG_CACHE_HOME, \"huggingface\", \"token\") : undefined,\n\t\tjoin(homedir(), \".cache\", \"huggingface\", \"token\"),\n\t].filter((path): path is string => Boolean(path));\n\tfor (const path of new Set(paths)) {\n\t\tconst token = await readToken(path);\n\t\tif (token) return token;\n\t}\n\treturn undefined;\n}\n\nexport class HuggingFaceClient {\n\tprivate readonly token: string | undefined;\n\tprivate readonly baseUrl: string;\n\n\tconstructor(token?: string, baseUrl = DEFAULT_HUGGING_FACE_URL) {\n\t\tthis.token = token;\n\t\tthis.baseUrl = baseUrl.replace(/\\/+$/u, \"\");\n\t}\n\n\tprivate async request(path: string, signal?: AbortSignal): Promise<unknown> {\n\t\tconst headers = new Headers();\n\t\tif (this.token) headers.set(\"Authorization\", `Bearer ${this.token}`);\n\t\tconst timeout = AbortSignal.timeout(15_000);\n\t\tconst response = await fetch(`${this.baseUrl}${path}`, {\n\t\t\theaders,\n\t\t\tsignal: signal ? AbortSignal.any([signal, timeout]) : timeout,\n\t\t});\n\t\tlet payload: unknown;\n\t\ttry {\n\t\t\tpayload = await response.json();\n\t\t} catch {\n\t\t\tpayload = undefined;\n\t\t}\n\t\tif (!response.ok) {\n\t\t\tconst fallback = `Hugging Face returned HTTP ${response.status}`;\n\t\t\tif (response.status === 429) {\n\t\t\t\tconst delay =\n\t\t\t\t\tNumber(response.headers.get(\"retry-after\")) || parseRateLimitDelay(response.headers.get(\"ratelimit\"));\n\t\t\t\tthrow new Error(\n\t\t\t\t\tdelay ? `Hugging Face rate limit reached; retry in ${delay}s` : \"Hugging Face rate limit reached\",\n\t\t\t\t);\n\t\t\t}\n\t\t\tthrow new Error(payloadError(payload, fallback));\n\t\t}\n\t\treturn payload;\n\t}\n\n\tasync search(query: string, signal?: AbortSignal): Promise<HuggingFaceModel[]> {\n\t\tconst params = new URLSearchParams({\n\t\t\tsearch: query,\n\t\t\tfilter: \"gguf\",\n\t\t\tsort: \"downloads\",\n\t\t\tdirection: \"-1\",\n\t\t\tlimit: \"20\",\n\t\t});\n\t\tconst payload = await this.request(`/api/models?${params}`, signal);\n\t\tif (!Array.isArray(payload)) throw new Error(\"Hugging Face returned invalid search results\");\n\t\treturn payload.flatMap((value) => {\n\t\t\tif (typeof value !== \"object\" || value === null || typeof (value as { id?: unknown }).id !== \"string\")\n\t\t\t\treturn [];\n\t\t\tconst model = value as { id: string; downloads?: unknown };\n\t\t\treturn [{ id: model.id, downloads: typeof model.downloads === \"number\" ? model.downloads : 0 }];\n\t\t});\n\t}\n\n\tasync details(id: string, signal?: AbortSignal): Promise<HuggingFaceModelDetails> {\n\t\tconst encodedId = id.split(\"/\").map(encodeURIComponent).join(\"/\");\n\t\tconst payload = await this.request(`/api/models/${encodedId}?blobs=true`, signal);\n\t\tif (typeof payload !== \"object\" || payload === null) {\n\t\t\tthrow new Error(\"Hugging Face returned invalid model details\");\n\t\t}\n\t\tconst model = payload as { id?: unknown; gated?: unknown; siblings?: unknown };\n\t\tconst sizes = new Map<string, { total: number; complete: boolean }>();\n\t\tif (Array.isArray(model.siblings)) {\n\t\t\tfor (const value of model.siblings) {\n\t\t\t\tif (typeof value !== \"object\" || value === null) continue;\n\t\t\t\tconst file = value as { rfilename?: unknown; size?: unknown };\n\t\t\t\tif (typeof file.rfilename !== \"string\" || !file.rfilename.toLowerCase().endsWith(\".gguf\")) continue;\n\t\t\t\tconst filename = file.rfilename.split(\"/\").at(-1)!;\n\t\t\t\tif (filename.toLowerCase().startsWith(\"mmproj\")) continue;\n\t\t\t\tconst stem = filename.slice(0, -5).replace(SHARD_SUFFIX_PATTERN, \"\");\n\t\t\t\tconst quantization = stem.match(QUANTIZATION_PATTERN)?.[1]?.toUpperCase();\n\t\t\t\tif (!quantization) continue;\n\t\t\t\tconst current = sizes.get(quantization) ?? { total: 0, complete: true };\n\t\t\t\tif (typeof file.size === \"number\") current.total += file.size;\n\t\t\t\telse current.complete = false;\n\t\t\t\tsizes.set(quantization, current);\n\t\t\t}\n\t\t}\n\t\tconst quantizations = [...sizes]\n\t\t\t.map(([name, size]) => ({ name, size: size.complete ? size.total : undefined }))\n\t\t\t.sort((left, right) => {\n\t\t\t\tif (left.name === \"Q4_K_M\") return -1;\n\t\t\t\tif (right.name === \"Q4_K_M\") return 1;\n\t\t\t\treturn (\n\t\t\t\t\t(left.size ?? Number.MAX_SAFE_INTEGER) - (right.size ?? Number.MAX_SAFE_INTEGER) ||\n\t\t\t\t\tleft.name.localeCompare(right.name)\n\t\t\t\t);\n\t\t\t});\n\t\treturn {\n\t\t\tid: typeof model.id === \"string\" ? model.id : id,\n\t\t\tgated: model.gated === \"auto\" || model.gated === \"manual\" ? model.gated : false,\n\t\t\tquantizations,\n\t\t};\n\t}\n}\n"]}

package/dist/extensions/llama/index.d.ts

@@ -0,0 +1,3 @@
+import type { ExtensionAPI } from "../../core/extensions/types.ts";
+export default function llamaExtension(pi: ExtensionAPI): void;
+//# sourceMappingURL=index.d.ts.map

package/dist/extensions/llama/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/extensions/llama/index.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAA2B,MAAM,gCAAgC,CAAC;AAyC5F,MAAM,CAAC,OAAO,UAAU,cAAc,CAAC,EAAE,EAAE,YAAY,GAAG,IAAI,CA0L7D","sourcesContent":["import type { ExtensionAPI, ExtensionCommandContext } from \"../../core/extensions/types.ts\";\nimport { formatBytes, LlamaClient, type LlamaModelInfo, normalizeLlamaServerUrl } from \"./client.ts\";\nimport { findHuggingFaceToken, HuggingFaceClient } from \"./huggingface.ts\";\nimport { createLlamaProvider, LLAMA_PROVIDER_ID } from \"./provider.ts\";\nimport { type LlamaUi, runWithProgress, showLlamaUi } from \"./ui.ts\";\n\nfunction modelIsLoaded(model: LlamaModelInfo): boolean {\n\treturn model.status.value === \"loaded\" || model.status.value === \"sleeping\";\n}\n\nfunction isConnectionError(error: unknown): boolean {\n\tif (!(error instanceof Error)) return false;\n\tconst message = `${error.name} ${error.message}`.toLowerCase();\n\treturn message.includes(\"fetch failed\") || message.includes(\"timeout\") || message.includes(\"network\");\n}\n\nfunction connectionErrorMessage(error: unknown): string {\n\tif (isConnectionError(error)) return \"Could not connect to the server.\";\n\treturn error instanceof Error ? error.message : String(error);\n}\n\nfunction parseHuggingFaceModel(value: string): { repository: string; quantization?: string } {\n\tconst colon = value.indexOf(\":\", value.indexOf(\"/\") + 1);\n\treturn colon < 0\n\t\t? { repository: value }\n\t\t: { repository: value.slice(0, colon), quantization: value.slice(colon + 1) };\n}\n\nasync function configuredClient(ctx: ExtensionCommandContext): Promise<LlamaClient | undefined> {\n\tconst result = await ctx.modelRegistry.getProviderAuth(LLAMA_PROVIDER_ID);\n\tif (!result) {\n\t\tctx.ui.notify(`Configure llama.cpp with /login ${LLAMA_PROVIDER_ID}`, \"warning\");\n\t\treturn undefined;\n\t}\n\tconst configuredUrl = result.env?.LLAMA_BASE_URL;\n\tconst serverUrl = normalizeLlamaServerUrl(\n\t\ttypeof configuredUrl === \"string\" && configuredUrl ? configuredUrl : (result.auth.baseUrl ?? \"\"),\n\t);\n\treturn new LlamaClient(serverUrl, result.auth.apiKey);\n}\n\nexport default function llamaExtension(pi: ExtensionAPI): void {\n\tconst provider = createLlamaProvider();\n\tpi.registerProvider(provider.provider);\n\n\tconst syncCatalog = async (\n\t\tctx: ExtensionCommandContext,\n\t\tclient: LlamaClient,\n\t\tcatalog?: LlamaModelInfo[],\n\t): Promise<LlamaModelInfo[]> => {\n\t\tconst signal = AbortSignal.timeout(15_000);\n\t\tconst current = catalog ?? (await client.list({ signal }));\n\t\tprovider.setCatalog(current, client.serverUrl);\n\t\tconst result = await ctx.modelRegistry.refresh({\n\t\t\tproviders: [LLAMA_PROVIDER_ID],\n\t\t\tsignal,\n\t\t});\n\t\tif (result.aborted) throw new Error(\"Model catalog refresh timed out.\");\n\t\tconst refreshError = result.errors.get(LLAMA_PROVIDER_ID);\n\t\tif (refreshError) throw refreshError;\n\t\treturn current;\n\t};\n\n\tconst loadModel = async (\n\t\tctx: ExtensionCommandContext,\n\t\tui: LlamaUi,\n\t\tclient: LlamaClient,\n\t\tcatalog: LlamaModelInfo[],\n\t\ttarget: LlamaModelInfo,\n\t): Promise<void> => {\n\t\tconst loaded = catalog.filter((model) => model.id !== target.id && modelIsLoaded(model));\n\t\tlet replace = false;\n\t\tif (loaded.length > 0) {\n\t\t\tconst choice = await ui.select(`${loaded.length} model${loaded.length === 1 ? \" is\" : \"s are\"} loaded`, [\n\t\t\t\t\"Unload all and load\",\n\t\t\t\t\"Keep loaded and load\",\n\t\t\t\t\"Cancel\",\n\t\t\t]);\n\t\t\tif (!choice || choice === \"Cancel\") return;\n\t\t\treplace = choice === \"Unload all and load\";\n\t\t}\n\n\t\tconst restoreLoaded = async (): Promise<void> => {\n\t\t\tctx.ui.notify(\"Restoring previously loaded models\");\n\t\t\tfor (const model of loaded) await client.loadAndWait(model.id, () => {});\n\t\t\tawait syncCatalog(ctx, client);\n\t\t};\n\t\tif (replace) {\n\t\t\tfor (const model of loaded) await client.unloadAndWait(model.id);\n\t\t}\n\n\t\ttry {\n\t\t\tconst result = await runWithProgress(ui, {\n\t\t\t\ttitle: \"Loading model\",\n\t\t\t\tmodel: target.id,\n\t\t\t\tinitialMessage: \"Starting…\",\n\t\t\t\tcancelTitle: \"Stop loading?\",\n\t\t\t\tcancelMessage: target.id,\n\t\t\t\trun: (signal, update) => client.loadAndWait(target.id, update, signal),\n\t\t\t\tcancel: () => client.unload(target.id),\n\t\t\t});\n\t\t\tif (result.cancelled) {\n\t\t\t\tif (replace) await restoreLoaded();\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst refreshed = await syncCatalog(ctx, client);\n\t\t\tconst loadedModel = refreshed.find((model) => model.id === target.id);\n\t\t\tctx.ui.notify(\n\t\t\t\tloadedModel?.status.value === \"loaded\" ? `Loaded ${target.id}` : `Load started for ${target.id}`,\n\t\t\t);\n\t\t} catch (error) {\n\t\t\tif (replace) {\n\t\t\t\ttry {\n\t\t\t\t\tawait restoreLoaded();\n\t\t\t\t} catch {\n\t\t\t\t\t// Preserve the original load error.\n\t\t\t\t}\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t};\n\n\tconst unloadModel = async (\n\t\tctx: ExtensionCommandContext,\n\t\tui: LlamaUi,\n\t\tclient: LlamaClient,\n\t\tmodel: LlamaModelInfo,\n\t): Promise<void> => {\n\t\tif (!(await ui.confirm(\"Unload model?\", model.id))) return;\n\t\tawait client.unloadAndWait(model.id);\n\t\tawait syncCatalog(ctx, client);\n\t\tctx.ui.notify(`Unloaded ${model.id}`);\n\t};\n\n\tconst downloadModel = async (ctx: ExtensionCommandContext, ui: LlamaUi, client: LlamaClient): Promise<void> => {\n\t\tconst huggingFace = new HuggingFaceClient(await findHuggingFaceToken());\n\t\tconst selected = await ui.searchModels((query, signal) => huggingFace.search(query, signal));\n\t\tif (!selected) return;\n\t\tconst parsed = parseHuggingFaceModel(selected);\n\t\tui.showStatus(\"Loading model details\", parsed.repository);\n\t\tconst details = await huggingFace.details(parsed.repository);\n\t\tif (details.gated) {\n\t\t\tconst approval = details.gated === \"manual\" ? \"Manual approval is required\" : \"Accept the access terms\";\n\t\t\tconst choice = await ui.select(\n\t\t\t\t`Hugging Face access required\\n${details.id}\\n\\n${approval} at:\\nhttps://huggingface.co/${details.id}\\n\\nThe llama.cpp server needs HF_TOKEN with access.`,\n\t\t\t\t[\"Continue\", \"Back\"],\n\t\t\t);\n\t\t\tif (choice !== \"Continue\") return;\n\t\t}\n\t\tlet quantization = parsed.quantization;\n\t\tif (!quantization && details.quantizations.length > 0) {\n\t\t\tconst options = details.quantizations.map((entry) => {\n\t\t\t\tconst detail = [\n\t\t\t\t\tentry.size === undefined ? undefined : formatBytes(entry.size),\n\t\t\t\t\tentry.name === \"Q4_K_M\" ? \"recommended\" : undefined,\n\t\t\t\t]\n\t\t\t\t\t.filter((value): value is string => Boolean(value))\n\t\t\t\t\t.join(\" · \");\n\t\t\t\treturn detail ? `${entry.name} · ${detail}` : entry.name;\n\t\t\t});\n\t\t\tconst choice = await ui.select(`Select quantization\\n${details.id}`, options);\n\t\t\tif (!choice) return;\n\t\t\tquantization = details.quantizations[options.indexOf(choice)]?.name;\n\t\t\tif (!quantization) return;\n\t\t}\n\t\tconst model = quantization ? `${details.id}:${quantization}` : details.id;\n\t\tconst result = await runWithProgress(ui, {\n\t\t\ttitle: \"Downloading model\",\n\t\t\tmodel,\n\t\t\tinitialMessage: \"Starting…\",\n\t\t\tcancelTitle: \"Stop download?\",\n\t\t\tcancelMessage: model,\n\t\t\trun: (signal, update) => client.downloadAndWait(model, update, signal),\n\t\t\tcancel: () => client.unload(model),\n\t\t});\n\t\tif (result.cancelled) return;\n\t\tawait syncCatalog(ctx, client, result.value);\n\t\tctx.ui.notify(`Downloaded ${model}`);\n\t};\n\n\tpi.registerCommand(\"llama\", {\n\t\tdescription: \"Manage llama.cpp router models\",\n\t\thandler: async (_args, ctx) => {\n\t\t\tif (ctx.mode !== \"tui\") {\n\t\t\t\tctx.ui.notify(\"/llama is available in interactive mode\", \"warning\");\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst client = await configuredClient(ctx);\n\t\t\tif (!client) return;\n\t\t\tawait showLlamaUi(ctx, async (ui) => {\n\t\t\t\tconst readCatalog = async (): Promise<LlamaModelInfo[] | undefined> => {\n\t\t\t\t\twhile (true) {\n\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\treturn await syncCatalog(ctx, client);\n\t\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\t\tif ((await ui.connectionError(client.serverUrl, connectionErrorMessage(error))) === \"close\") {\n\t\t\t\t\t\t\t\treturn undefined;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t};\n\n\t\t\t\tlet catalog = await readCatalog();\n\t\t\t\tif (!catalog) return;\n\t\t\t\twhile (true) {\n\t\t\t\t\tconst action = await ui.showModels(client.serverUrl, catalog);\n\t\t\t\t\tif (action.type === \"close\") return;\n\t\t\t\t\tlet actionError: unknown;\n\t\t\t\t\ttry {\n\t\t\t\t\t\tif (action.type === \"download\") await downloadModel(ctx, ui, client);\n\t\t\t\t\t\telse if (modelIsLoaded(action.model)) await unloadModel(ctx, ui, client, action.model);\n\t\t\t\t\t\telse if (action.model.status.value === \"unloaded\")\n\t\t\t\t\t\t\tawait loadModel(ctx, ui, client, catalog, action.model);\n\t\t\t\t\t\telse ctx.ui.notify(`${action.model.id} is ${action.model.status.value}`, \"warning\");\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\tactionError = error;\n\t\t\t\t\t}\n\t\t\t\t\tconst refreshed = await readCatalog();\n\t\t\t\t\tif (!refreshed) return;\n\t\t\t\t\tcatalog = refreshed;\n\t\t\t\t\tif (actionError && !isConnectionError(actionError)) {\n\t\t\t\t\t\tctx.ui.notify(actionError instanceof Error ? actionError.message : String(actionError), \"error\");\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t});\n\t\t},\n\t});\n}\n"]}