@atomic-chat/core 0.18.5 → 0.19.1

package/dist/runtime/litert-lm/check.js

@@ -0,0 +1,113 @@
+/**
+ * `POST /models/litert-lm/check` (openspec change `add-litert-lm-runtime`, task 6.5): whether a
+ * variant would run here. Before a download, from what core already knows — the platform, this
+ * machine's Intel NPU generation, the active build, whether OpenVINO is prepared — never the network.
+ * For a file on disk, also from the file itself (`litert-host inspect`), whose facts come back in the
+ * answer for the client to write into `model.yml`.
+ */
+import { realpath } from 'node:fs/promises';
+import { isAbsolute, relative, resolve } from 'node:path';
+import { AtomicCoreError, LITERT_LM_BACKENDS, LITERT_LM_NPUS } from '../../contracts/index.js';
+import { compareVersions } from '../../backend/index.js';
+import { readLiteRtLmBlock } from './model-yml.js';
+import { backendRunsOn } from './runtime.js';
+const verdict = (v, reason, metadata) => ({ verdict: v, reason, ...(metadata ? { metadata } : {}) });
+/** The verdict for a variant (`backend`, `npu`, `min_runtime_version`), and for a file its facts. */
+export function liteRtLmVerdict(facts, request, file) {
+    const backend = request.backend ?? '';
+    if (!LITERT_LM_BACKENDS.some((known) => known === backend))
+        return verdict('platform-unsupported', `This core runs no LiteRT-LM variant on the "${backend}" backend.`, file);
+    if (!backendRunsOn(facts.platform, backend))
+        return verdict('platform-unsupported', `LiteRT-LM runs ${backend.toUpperCase()} variants on Windows x64 only.`, file);
+    if (backend === 'npu') {
+        if (!LITERT_LM_NPUS.some((known) => known === request.npu))
+            return verdict('platform-unsupported', `This core runs no LiteRT-LM variant for the "${request.npu ?? ''}" NPU.`, file);
+        if (facts.npu === null)
+            return verdict('accelerator-unavailable', 'This computer has no Intel NPU LiteRT-LM has models for.', file);
+        if (facts.npu !== request.npu)
+            return verdict('accelerator-unavailable', `This variant is built for the ${request.npu} NPU; this computer has ${facts.npu}.`, file);
+    }
+    if (!facts.build)
+        return verdict('engine-not-installed', 'LiteRT-LM is not installed on this computer.', file);
+    if (file && !file.backends.includes(backend))
+        return verdict('platform-unsupported', `This file runs on ${file.backends.join(', ') || 'no backend'}, not ${backend}.`, file);
+    const needs = file?.min_runtime_version ?? request.min_runtime_version ?? null;
+    if (needs && compareVersions(needs, facts.build.version) > 0)
+        return verdict('requires-newer-engine', `This model needs LiteRT-LM ${needs}; the installed one is ${facts.build.version}.`, file);
+    if (backend === 'npu' && !facts.openVinoPrepared)
+        return verdict('accelerator-not-prepared', 'OpenVINO is not prepared yet; prepare it on the LiteRT-LM page.', file);
+    return verdict('ok', null, file);
+}
+const VERSION_RE = /^\d+\.\d+(\.\d+)?$/;
+const FIELDS = ['backend', 'npu', 'min_runtime_version', 'path', 'model_id'];
+/** The request body, strictly: a variant, a path with its backend, or a model id. */
+export function parseLiteRtLmCheckRequest(body) {
+    const invalid = (why) => {
+        throw new AtomicCoreError('INVALID_ARGUMENT', why);
+    };
+    if (typeof body !== 'object' || body === null || Array.isArray(body))
+        return invalid('The request body must be a JSON object.');
+    const record = body;
+    const extra = Object.keys(record).filter((key) => !FIELDS.some((field) => field === key));
+    if (extra.length > 0)
+        invalid(`The request has fields this core does not know: ${extra.sort().join(', ')}.`);
+    const out = {};
+    for (const key of FIELDS) {
+        const value = record[key];
+        if (value === undefined)
+            continue;
+        if (typeof value !== 'string' || value === '')
+            invalid(`${key} must be text.`);
+        out[key] = value;
+    }
+    if (out.min_runtime_version !== undefined && !VERSION_RE.test(out.min_runtime_version))
+        invalid('min_runtime_version must be X.Y or X.Y.Z.');
+    if (out.model_id !== undefined) {
+        if (Object.keys(out).length > 1)
+            invalid('model_id is checked alone: its variant comes from model.yml.');
+        return out;
+    }
+    if (out.backend === undefined)
+        invalid('backend is required for a variant or a file.');
+    return out;
+}
+export class LiteRtLmCheckService {
+    deps;
+    constructor(deps) {
+        this.deps = deps;
+    }
+    /** A path inside the models root, links resolved, or `INVALID_ARGUMENT`. */
+    async inRoot(path) {
+        const refuse = () => new AtomicCoreError('INVALID_ARGUMENT', 'path must be a file inside the LiteRT-LM models folder.');
+        if (!isAbsolute(path))
+            throw new AtomicCoreError('INVALID_ARGUMENT', 'path must be absolute.');
+        const root = await realpath(this.deps.modelsDir).catch(() => resolve(this.deps.modelsDir));
+        // A file that is not there yet is judged by its spelling; one that is, by where its links lead.
+        const full = await realpath(path).catch(() => resolve(path));
+        const rel = relative(root, full);
+        if (rel === '' || rel.startsWith('..') || isAbsolute(rel))
+            throw refuse();
+        return resolve(path);
+    }
+    async check(request) {
+        let variant = request;
+        let file = request.path !== undefined ? await this.inRoot(request.path) : null;
+        if (request.model_id !== undefined) {
+            const yml = await this.deps.registry.read(request.model_id);
+            const block = readLiteRtLmBlock(yml);
+            variant = { backend: block.backend, ...(block.npu ? { npu: block.npu } : {}) };
+            // The registry's own path: an installed model may be a link, which the runtime loads as it is.
+            file = this.deps.registry.resolvePaths(yml).modelPath;
+        }
+        const build = (await this.deps.resolveBuild()) ?? null;
+        const facts = {
+            platform: this.deps.platform,
+            build,
+            npu: await this.deps.npuGeneration(),
+            openVinoPrepared: (await this.deps.openVinoLibs()) !== undefined,
+        };
+        const metadata = file && build ? await this.deps.inspect(build.dir, file) : null;
+        return liteRtLmVerdict(facts, variant, metadata);
+    }
+}
+//# sourceMappingURL=check.js.map

package/dist/runtime/litert-lm/check.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"check.js","sourceRoot":"","sources":["../../../src/runtime/litert-lm/check.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,EAAE,QAAQ,EAAE,MAAM,kBAAkB,CAAA;AAC3C,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,WAAW,CAAA;AACzD,OAAO,EAAE,eAAe,EAAE,kBAAkB,EAAE,cAAc,EAAE,MAAM,0BAA0B,CAAA;AAS9F,OAAO,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAA;AACxD,OAAO,EAAE,iBAAiB,EAAE,MAAM,gBAAgB,CAAA;AAClD,OAAO,EAAE,aAAa,EAAE,MAAM,cAAc,CAAA;AAS5C,MAAM,OAAO,GAAG,CACd,CAAiC,EACjC,MAAqB,EACrB,QAAqC,EAChB,EAAE,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,MAAM,EAAE,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,CAAA;AAErF,qGAAqG;AACrG,MAAM,UAAU,eAAe,CAC7B,KAAyB,EACzB,OAA8E,EAC9E,IAAiC;IAEjC,MAAM,OAAO,GAAG,OAAO,CAAC,OAAO,IAAI,EAAE,CAAA;IACrC,IAAI,CAAC,kBAAkB,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,KAAK,OAAO,CAAC;QACxD,OAAO,OAAO,CACZ,sBAAsB,EACtB,+CAA+C,OAAO,YAAY,EAClE,IAAI,CACL,CAAA;IACH,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,QAAQ,EAAE,OAA8C,CAAC;QAChF,OAAO,OAAO,CACZ,sBAAsB,EACtB,kBAAkB,OAAO,CAAC,WAAW,EAAE,gCAAgC,EACvE,IAAI,CACL,CAAA;IACH,IAAI,OAAO,KAAK,KAAK,EAAE,CAAC;QACtB,IAAI,CAAC,cAAc,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,KAAK,OAAO,CAAC,GAAG,CAAC;YACxD,OAAO,OAAO,CACZ,sBAAsB,EACtB,gDAAgD,OAAO,CAAC,GAAG,IAAI,EAAE,QAAQ,EACzE,IAAI,CACL,CAAA;QACH,IAAI,KAAK,CAAC,GAAG,KAAK,IAAI;YACpB,OAAO,OAAO,CACZ,yBAAyB,EACzB,0DAA0D,EAC1D,IAAI,CACL,CAAA;QACH,IAAI,KAAK,CAAC,GAAG,KAAK,OAAO,CAAC,GAAG;YAC3B,OAAO,OAAO,CACZ,yBAAyB,EACzB,iCAAiC,OAAO,CAAC,GAAG,2BAA2B,KAAK,CAAC,GAAG,GAAG,EACnF,IAAI,CACL,CAAA;IACL,CAAC;IACD,IAAI,CAAC,KAAK,CAAC,KAAK;QACd,OAAO,OAAO,CAAC,sBAAsB,EAAE,8CAA8C,EAAE,IAAI,CAAC,CAAA;IAC9F,IAAI,IAAI,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,QAAQ,CAAC,OAA8C,CAAC;QACjF,OAAO,OAAO,CACZ,sBAAsB,EACtB,qBAAqB,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,YAAY,SAAS,OAAO,GAAG,EAChF,IAAI,CACL,CAAA;IACH,MAAM,KAAK,GAAG,IAAI,EAAE,mBAAmB,IAAI,OAAO,CAAC,mBAAmB,IAAI,IAAI,CAAA;IAC9E,IAAI,KAAK,IAAI,eAAe,CAAC,KAAK,EAAE,KAAK,CAAC,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC;QAC1D,OAAO,OAAO,CACZ,uBAAuB,EACvB,8BAA8B,KAAK,0BAA0B,KAAK,CAAC,KAAK,CAAC,OAAO,GAAG,EACnF,IAAI,CACL,CAAA;IACH,IAAI,OAAO,KAAK,KAAK,IAAI,CAAC,KAAK,CAAC,gBAAgB;QAC9C,OAAO,OAAO,CACZ,0BAA0B,EAC1B,iEAAiE,EACjE,IAAI,CACL,CAAA;IACH,OAAO,OAAO,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,CAAA;AAClC,CAAC;AAED,MAAM,UAAU,GAAG,oBAAoB,CAAA;AACvC,MAAM,MAAM,GAAG,CAAC,SAAS,EAAE,KAAK,EAAE,qBAAqB,EAAE,MAAM,EAAE,UAAU,CAAU,CAAA;AAErF,qFAAqF;AACrF,MAAM,UAAU,yBAAyB,CAAC,IAAa;IACrD,MAAM,OAAO,GAAG,CAAC,GAAW,EAAS,EAAE;QACrC,MAAM,IAAI,eAAe,CAAC,kBAAkB,EAAE,GAAG,CAAC,CAAA;IACpD,CAAC,CAAA;IACD,IAAI,OAAO,IAAI,KAAK,QAAQ,IAAI,IAAI,KAAK,IAAI,IAAI,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC;QAClE,OAAO,OAAO,CAAC,yCAAyC,CAAC,CAAA;IAC3D,MAAM,MAAM,GAAG,IAA+B,CAAA;IAC9C,MAAM,KAAK,GAAG,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,KAAK,GAAG,CAAC,CAAC,CAAA;IACzF,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,CAAC,mDAAmD,KAAK,CAAC,IAAI,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAA;IAC5G,MAAM,GAAG,GAAyB,EAAE,CAAA;IACpC,KAAK,MAAM,GAAG,IAAI,MAAM,EAAE,CAAC;QACzB,MAAM,KAAK,GAAG,MAAM,CAAC,GAAG,CAAC,CAAA;QACzB,IAAI,KAAK,KAAK,SAAS;YAAE,SAAQ;QACjC,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,EAAE;YAAE,OAAO,CAAC,GAAG,GAAG,gBAAgB,CAAC,CAAA;QAC9E,GAAG,CAAC,GAAG,CAAC,GAAG,KAAe,CAAA;IAC5B,CAAC;IACD,IAAI,GAAG,CAAC,mBAAmB,KAAK,SAAS,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,GAAG,CAAC,mBAAmB,CAAC;QACpF,OAAO,CAAC,2CAA2C,CAAC,CAAA;IACtD,IAAI,GAAG,CAAC,QAAQ,KAAK,SAAS,EAAE,CAAC;QAC/B,IAAI,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,MAAM,GAAG,CAAC;YAAE,OAAO,CAAC,8DAA8D,CAAC,CAAA;QACxG,OAAO,GAAG,CAAA;IACZ,CAAC;IACD,IAAI,GAAG,CAAC,OAAO,KAAK,SAAS;QAAE,OAAO,CAAC,8CAA8C,CAAC,CAAA;IACtF,OAAO,GAAG,CAAA;AACZ,CAAC;AAcD,MAAM,OAAO,oBAAoB;IACF;IAA7B,YAA6B,IAAuB;QAAvB,SAAI,GAAJ,IAAI,CAAmB;IAAG,CAAC;IAExD,4EAA4E;IACpE,KAAK,CAAC,MAAM,CAAC,IAAY;QAC/B,MAAM,MAAM,GAAG,GAAG,EAAE,CAClB,IAAI,eAAe,CAAC,kBAAkB,EAAE,yDAAyD,CAAC,CAAA;QACpG,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC;YAAE,MAAM,IAAI,eAAe,CAAC,kBAAkB,EAAE,wBAAwB,CAAC,CAAA;QAC9F,MAAM,IAAI,GAAG,MAAM,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAA;QAC1F,gGAAgG;QAChG,MAAM,IAAI,GAAG,MAAM,QAAQ,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAA;QAC5D,MAAM,GAAG,GAAG,QAAQ,CAAC,IAAI,EAAE,IAAI,CAAC,CAAA;QAChC,IAAI,GAAG,KAAK,EAAE,IAAI,GAAG,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,UAAU,CAAC,GAAG,CAAC;YAAE,MAAM,MAAM,EAAE,CAAA;QACzE,OAAO,OAAO,CAAC,IAAI,CAAC,CAAA;IACtB,CAAC;IAED,KAAK,CAAC,KAAK,CAAC,OAA6B;QACvC,IAAI,OAAO,GAA0E,OAAO,CAAA;QAC5F,IAAI,IAAI,GAAkB,OAAO,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,MAAM,IAAI,CAAC,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAA;QAC7F,IAAI,OAAO,CAAC,QAAQ,KAAK,SAAS,EAAE,CAAC;YACnC,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAA;YAC3D,MAAM,KAAK,GAAG,iBAAiB,CAAC,GAAG,CAAC,CAAA;YACpC,OAAO,GAAG,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,EAAE,KAAK,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAA;YAC9E,+FAA+F;YAC/F,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC,SAAS,CAAA;QACvD,CAAC;QACD,MAAM,KAAK,GAAG,CAAC,MAAM,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,CAAC,IAAI,IAAI,CAAA;QACtD,MAAM,KAAK,GAAuB;YAChC,QAAQ,EAAE,IAAI,CAAC,IAAI,CAAC,QAAQ;YAC5B,KAAK;YACL,GAAG,EAAE,MAAM,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE;YACpC,gBAAgB,EAAE,CAAC,MAAM,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,CAAC,KAAK,SAAS;SACjE,CAAA;QACD,MAAM,QAAQ,GAAG,IAAI,IAAI,KAAK,CAAC,CAAC,CAAC,MAAM,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAA;QAChF,OAAO,eAAe,CAAC,KAAK,EAAE,OAAO,EAAE,QAAQ,CAAC,CAAA;IAClD,CAAC;CACF"}

package/dist/runtime/litert-lm/host/args.d.ts

@@ -0,0 +1,47 @@
+/**
+ * The arguments of the hidden `litert-host` mode (openspec change `add-litert-lm-runtime`, task 5.1).
+ * Core is the only caller, so the parse is strict: anything unknown is a bug on core's side.
+ *
+ *   litert-host probe   --lib <build>
+ *   litert-host inspect --lib <build> <file>
+ *   litert-host serve   --lib <build> --engine-version <v> --model <file> --backend cpu|gpu|npu
+ *                       --max-tokens <n> --port <p, 0 for any> --cache <dir> --id <model id>
+ *                       [--dispatch <dir> --ov-libs <dir>] [--speculative] [--reasoning]
+ */
+export declare const LITERT_LM_BACKENDS: readonly ["cpu", "gpu", "npu"];
+export type LiteRtLmBackend = (typeof LITERT_LM_BACKENDS)[number];
+export type LiteRtHostArgs = {
+    mode: 'probe';
+    lib: string;
+} | {
+    mode: 'inspect';
+    lib: string;
+    file: string;
+} | {
+    mode: 'serve';
+    /** The build directory, holding `lib/`. */
+    lib: string;
+    /** The build's LiteRT-LM version, checked against the file's `min_runtime_version`. */
+    engineVersion: string;
+    model: string;
+    backend: LiteRtLmBackend;
+    /** The engine's `max_num_tokens`, already cut to what the file allows. */
+    maxTokens: number;
+    port: number;
+    /** The library's `cache_dir`; created when missing. */
+    cache: string;
+    /** The id `/v1/models` lists: core's session id. */
+    id: string;
+    /** NPU only: the build's `vendors/intel_openvino/dispatch`. */
+    dispatch?: string;
+    /** NPU only: the prepared OpenVINO libraries. */
+    ovLibs?: string;
+    /** Use the file's draft model when it has one. */
+    speculative: boolean;
+    /** Think by default (`model.yml` `reasoning`); a request may still turn it off. */
+    reasoning: boolean;
+};
+export declare class LiteRtHostUsageError extends Error {
+    constructor(message: string);
+}
+export declare function parseLiteRtHostArgs(argv: readonly string[]): LiteRtHostArgs;

package/dist/runtime/litert-lm/host/args.js

@@ -0,0 +1,108 @@
+/**
+ * The arguments of the hidden `litert-host` mode (openspec change `add-litert-lm-runtime`, task 5.1).
+ * Core is the only caller, so the parse is strict: anything unknown is a bug on core's side.
+ *
+ *   litert-host probe   --lib <build>
+ *   litert-host inspect --lib <build> <file>
+ *   litert-host serve   --lib <build> --engine-version <v> --model <file> --backend cpu|gpu|npu
+ *                       --max-tokens <n> --port <p, 0 for any> --cache <dir> --id <model id>
+ *                       [--dispatch <dir> --ov-libs <dir>] [--speculative] [--reasoning]
+ */
+import { parseArgs } from 'node:util';
+export const LITERT_LM_BACKENDS = ['cpu', 'gpu', 'npu'];
+export class LiteRtHostUsageError extends Error {
+    constructor(message) {
+        super(message);
+        this.name = 'LiteRtHostUsageError';
+    }
+}
+const required = (values, key) => {
+    const value = values[key];
+    if (typeof value !== 'string' || value === '')
+        throw new LiteRtHostUsageError(`--${key} is required`);
+    return value;
+};
+const VERSION_RE = /^\d+\.\d+\.\d+$/;
+function wholeNumber(values, key, min, max) {
+    const raw = required(values, key);
+    const value = Number(raw);
+    if (!Number.isSafeInteger(value) || value < min || value > max)
+        throw new LiteRtHostUsageError(`--${key} ${raw} is not a whole number from ${min} to ${max}`);
+    return value;
+}
+export function parseLiteRtHostArgs(argv) {
+    const [mode, ...rest] = argv;
+    try {
+        if (mode === 'probe') {
+            const { values } = parseArgs({ args: rest, options: { lib: { type: 'string' } }, strict: true });
+            return { mode, lib: required(values, 'lib') };
+        }
+        if (mode === 'inspect') {
+            const { values, positionals } = parseArgs({
+                args: rest,
+                options: { lib: { type: 'string' } },
+                strict: true,
+                allowPositionals: true,
+            });
+            const lib = required(values, 'lib');
+            if (positionals.length !== 1 || !positionals[0])
+                throw new LiteRtHostUsageError('inspect takes one model file');
+            return { mode, lib, file: positionals[0] };
+        }
+        if (mode === 'serve') {
+            const { values } = parseArgs({
+                args: rest,
+                options: {
+                    'lib': { type: 'string' },
+                    'engine-version': { type: 'string' },
+                    'model': { type: 'string' },
+                    'backend': { type: 'string' },
+                    'max-tokens': { type: 'string' },
+                    'port': { type: 'string' },
+                    'cache': { type: 'string' },
+                    'id': { type: 'string' },
+                    'dispatch': { type: 'string' },
+                    'ov-libs': { type: 'string' },
+                    'speculative': { type: 'boolean' },
+                    'reasoning': { type: 'boolean' },
+                },
+                strict: true,
+            });
+            const lib = required(values, 'lib');
+            const engineVersion = required(values, 'engine-version');
+            if (!VERSION_RE.test(engineVersion))
+                throw new LiteRtHostUsageError(`--engine-version ${engineVersion} is not X.Y.Z`);
+            const backend = required(values, 'backend');
+            if (!LITERT_LM_BACKENDS.some((known) => known === backend))
+                throw new LiteRtHostUsageError(`--backend ${backend} is not cpu, gpu or npu`);
+            const dispatch = values['dispatch'];
+            const ovLibs = values['ov-libs'];
+            if (backend === 'npu' && (!dispatch || !ovLibs))
+                throw new LiteRtHostUsageError('--backend npu needs --dispatch and --ov-libs');
+            if (backend !== 'npu' && (dispatch !== undefined || ovLibs !== undefined))
+                throw new LiteRtHostUsageError('--dispatch and --ov-libs are only for --backend npu');
+            return {
+                mode,
+                lib,
+                engineVersion,
+                model: required(values, 'model'),
+                backend: backend,
+                maxTokens: wholeNumber(values, 'max-tokens', 1, 16_777_216),
+                port: wholeNumber(values, 'port', 0, 65535),
+                cache: required(values, 'cache'),
+                id: required(values, 'id'),
+                ...(dispatch ? { dispatch } : {}),
+                ...(ovLibs ? { ovLibs } : {}),
+                speculative: values['speculative'] === true,
+                reasoning: values['reasoning'] === true,
+            };
+        }
+    }
+    catch (error) {
+        if (error instanceof LiteRtHostUsageError)
+            throw error;
+        throw new LiteRtHostUsageError(error instanceof Error ? error.message : String(error));
+    }
+    throw new LiteRtHostUsageError(`litert-host: unknown mode ${JSON.stringify(mode ?? '')} (probe, inspect, serve)`);
+}
+//# sourceMappingURL=args.js.map

package/dist/runtime/litert-lm/host/args.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"args.js","sourceRoot":"","sources":["../../../../src/runtime/litert-lm/host/args.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,OAAO,EAAE,SAAS,EAAE,MAAM,WAAW,CAAA;AAErC,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC,KAAK,EAAE,KAAK,EAAE,KAAK,CAAU,CAAA;AA+BhE,MAAM,OAAO,oBAAqB,SAAQ,KAAK;IAC7C,YAAY,OAAe;QACzB,KAAK,CAAC,OAAO,CAAC,CAAA;QACd,IAAI,CAAC,IAAI,GAAG,sBAAsB,CAAA;IACpC,CAAC;CACF;AAED,MAAM,QAAQ,GAAG,CAAC,MAA+B,EAAE,GAAW,EAAU,EAAE;IACxE,MAAM,KAAK,GAAG,MAAM,CAAC,GAAG,CAAC,CAAA;IACzB,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,EAAE;QAAE,MAAM,IAAI,oBAAoB,CAAC,KAAK,GAAG,cAAc,CAAC,CAAA;IACrG,OAAO,KAAK,CAAA;AACd,CAAC,CAAA;AAED,MAAM,UAAU,GAAG,iBAAiB,CAAA;AAEpC,SAAS,WAAW,CAAC,MAA+B,EAAE,GAAW,EAAE,GAAW,EAAE,GAAW;IACzF,MAAM,GAAG,GAAG,QAAQ,CAAC,MAAM,EAAE,GAAG,CAAC,CAAA;IACjC,MAAM,KAAK,GAAG,MAAM,CAAC,GAAG,CAAC,CAAA;IACzB,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,KAAK,GAAG,GAAG,IAAI,KAAK,GAAG,GAAG;QAC5D,MAAM,IAAI,oBAAoB,CAAC,KAAK,GAAG,IAAI,GAAG,+BAA+B,GAAG,OAAO,GAAG,EAAE,CAAC,CAAA;IAC/F,OAAO,KAAK,CAAA;AACd,CAAC;AAED,MAAM,UAAU,mBAAmB,CAAC,IAAuB;IACzD,MAAM,CAAC,IAAI,EAAE,GAAG,IAAI,CAAC,GAAG,IAAI,CAAA;IAC5B,IAAI,CAAC;QACH,IAAI,IAAI,KAAK,OAAO,EAAE,CAAC;YACrB,MAAM,EAAE,MAAM,EAAE,GAAG,SAAS,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,OAAO,EAAE,EAAE,GAAG,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC,CAAA;YAChG,OAAO,EAAE,IAAI,EAAE,GAAG,EAAE,QAAQ,CAAC,MAAM,EAAE,KAAK,CAAC,EAAE,CAAA;QAC/C,CAAC;QACD,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;YACvB,MAAM,EAAE,MAAM,EAAE,WAAW,EAAE,GAAG,SAAS,CAAC;gBACxC,IAAI,EAAE,IAAI;gBACV,OAAO,EAAE,EAAE,GAAG,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE;gBACpC,MAAM,EAAE,IAAI;gBACZ,gBAAgB,EAAE,IAAI;aACvB,CAAC,CAAA;YACF,MAAM,GAAG,GAAG,QAAQ,CAAC,MAAM,EAAE,KAAK,CAAC,CAAA;YACnC,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC;gBAC7C,MAAM,IAAI,oBAAoB,CAAC,8BAA8B,CAAC,CAAA;YAChE,OAAO,EAAE,IAAI,EAAE,GAAG,EAAE,IAAI,EAAE,WAAW,CAAC,CAAC,CAAC,EAAE,CAAA;QAC5C,CAAC;QACD,IAAI,IAAI,KAAK,OAAO,EAAE,CAAC;YACrB,MAAM,EAAE,MAAM,EAAE,GAAG,SAAS,CAAC;gBAC3B,IAAI,EAAE,IAAI;gBACV,OAAO,EAAE;oBACP,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;oBACzB,gBAAgB,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;oBACpC,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;oBAC3B,SAAS,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;oBAC7B,YAAY,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;oBAChC,MAAM,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;oBAC1B,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;oBAC3B,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;oBACxB,UAAU,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;oBAC9B,SAAS,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;oBAC7B,aAAa,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE;oBAClC,WAAW,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE;iBACjC;gBACD,MAAM,EAAE,IAAI;aACb,CAAC,CAAA;YACF,MAAM,GAAG,GAAG,QAAQ,CAAC,MAAM,EAAE,KAAK,CAAC,CAAA;YACnC,MAAM,aAAa,GAAG,QAAQ,CAAC,MAAM,EAAE,gBAAgB,CAAC,CAAA;YACxD,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,aAAa,CAAC;gBACjC,MAAM,IAAI,oBAAoB,CAAC,oBAAoB,aAAa,eAAe,CAAC,CAAA;YAClF,MAAM,OAAO,GAAG,QAAQ,CAAC,MAAM,EAAE,SAAS,CAAC,CAAA;YAC3C,IAAI,CAAC,kBAAkB,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,KAAK,OAAO,CAAC;gBACxD,MAAM,IAAI,oBAAoB,CAAC,aAAa,OAAO,yBAAyB,CAAC,CAAA;YAC/E,MAAM,QAAQ,GAAG,MAAM,CAAC,UAAU,CAAC,CAAA;YACnC,MAAM,MAAM,GAAG,MAAM,CAAC,SAAS,CAAC,CAAA;YAChC,IAAI,OAAO,KAAK,KAAK,IAAI,CAAC,CAAC,QAAQ,IAAI,CAAC,MAAM,CAAC;gBAC7C,MAAM,IAAI,oBAAoB,CAAC,8CAA8C,CAAC,CAAA;YAChF,IAAI,OAAO,KAAK,KAAK,IAAI,CAAC,QAAQ,KAAK,SAAS,IAAI,MAAM,KAAK,SAAS,CAAC;gBACvE,MAAM,IAAI,oBAAoB,CAAC,qDAAqD,CAAC,CAAA;YACvF,OAAO;gBACL,IAAI;gBACJ,GAAG;gBACH,aAAa;gBACb,KAAK,EAAE,QAAQ,CAAC,MAAM,EAAE,OAAO,CAAC;gBAChC,OAAO,EAAE,OAA0B;gBACnC,SAAS,EAAE,WAAW,CAAC,MAAM,EAAE,YAAY,EAAE,CAAC,EAAE,UAAU,CAAC;gBAC3D,IAAI,EAAE,WAAW,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,EAAE,KAAK,CAAC;gBAC3C,KAAK,EAAE,QAAQ,CAAC,MAAM,EAAE,OAAO,CAAC;gBAChC,EAAE,EAAE,QAAQ,CAAC,MAAM,EAAE,IAAI,CAAC;gBAC1B,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBACjC,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC7B,WAAW,EAAE,MAAM,CAAC,aAAa,CAAC,KAAK,IAAI;gBAC3C,SAAS,EAAE,MAAM,CAAC,WAAW,CAAC,KAAK,IAAI;aACxC,CAAA;QACH,CAAC;IACH,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,KAAK,YAAY,oBAAoB;YAAE,MAAM,KAAK,CAAA;QACtD,MAAM,IAAI,oBAAoB,CAAC,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAA;IACxF,CAAC;IACD,MAAM,IAAI,oBAAoB,CAC5B,6BAA6B,IAAI,CAAC,SAAS,CAAC,IAAI,IAAI,EAAE,CAAC,0BAA0B,CAClF,CAAA;AACH,CAAC"}

package/dist/runtime/litert-lm/host/ffi-bun.d.ts

@@ -0,0 +1,16 @@
+/**
+ * `LiteRtLibrary` over `bun:ffi` — the second and last file in `src/` that may import `bun:*` (change
+ * `add-litert-lm-runtime`, design D1 and decision 3.6; allow-listed in both runtime-agnostic gates). It
+ * is imported dynamically by the `litert-host` mode alone, so the npm package still loads under Node.
+ *
+ * Streaming is the hard part (spike S2). The library calls a stream callback on a C++ thread of its
+ * own and frees the chunk as soon as the callback returns. A plain `JSCallback` cannot run there; a
+ * `threadsafe` one only queues the call, so the JS thread would read freed memory, and even with the
+ * chunk copied first Bun's threadsafe callbacks crashed about one run in five under a fast stream.
+ * So the callback is C: a shim compiled at start by the TinyCC built into Bun (`cc`) copies each chunk
+ * on the library's thread into a FIFO under a mutex, and JS drains the FIFO on a 2 ms timer that runs
+ * only while a generation is under way. The shim links nothing (`-nostdlib`: TinyCC finds no libc
+ * without development packages on Linux); `malloc` and the mutex come in as pointers from `dlsym`.
+ */
+import type { LoadLiteRtLibrary } from './library.js';
+export declare const loadBunLiteRtLibrary: LoadLiteRtLibrary;