@modular-prompt/driver 0.14.0 → 0.16.0

package/dist/local-inference/process-client.d.ts

@@ -0,0 +1,50 @@
+/**
+ * Local Inference Protocol プロセスクライアント
+ *
+ * stdio JSON-RPC 通信とリクエストキューを統合する。
+ */
+import { Readable } from 'stream';
+import type { RuntimeProfile } from '../runtime/index.js';
+import type { InferenceCapabilities, InferenceFormatTestResult, InferenceRenderResult, InferenceTokenizeResult, InferenceCachePrefillResult, InferenceMessage, InferenceToolDefinition, InferenceSamplingOptions } from './protocol.js';
+import { type SamplingOptionsMapper } from './request-queue.js';
+export interface InferenceProcessClientConfig {
+    modelName: string;
+    /** uv --project に渡す Python プロジェクトディレクトリ */
+    pythonProjectDir: string;
+    /** UV_PROJECT_ENVIRONMENT に設定する venv パス */
+    venvPath: string;
+    /** 指定時は起動前に runtime venv の存在を確認する */
+    runtimeProfile?: RuntimeProfile;
+    extraSpawnArgs?: string[];
+    extraEnv?: Record<string, string>;
+    loggerPrefix?: string;
+    mapSamplingOptions?: SamplingOptionsMapper;
+    processExitErrorMessage?: (code: number | null, signal: string | null) => string;
+}
+export declare class InferenceProcessClient {
+    readonly modelName: string;
+    private requestQueue;
+    private processComm;
+    constructor(config: InferenceProcessClientConfig);
+    ensureInitialized(): Promise<void>;
+    getCapabilities(): Promise<InferenceCapabilities>;
+    formatTest(messages: InferenceMessage[], options?: {
+        primer?: string;
+    }): Promise<InferenceFormatTestResult>;
+    render(messages: InferenceMessage[], options?: InferenceSamplingOptions & {
+        primer?: string;
+    }, tools?: InferenceToolDefinition[], reasoningEffort?: 'low' | 'medium' | 'high'): Promise<InferenceRenderResult>;
+    tokenize(messages: InferenceMessage[], tools?: InferenceToolDefinition[], reasoningEffort?: 'low' | 'medium' | 'high'): Promise<InferenceTokenizeResult>;
+    cachePrefill(cachePath: string, messages: InferenceMessage[], baseCachePath?: string, trimToTokens?: number, prefixOffsets?: number[], prefixHashes?: string[], tools?: InferenceToolDefinition[], reasoningEffort?: 'low' | 'medium' | 'high'): Promise<InferenceCachePrefillResult>;
+    completion(prompt: string, options?: unknown, images?: string[], maxImageSize?: number): Promise<Readable>;
+    generate(prompt: string | number[], options?: unknown, images?: string[], maxImageSize?: number, cachePath?: string, cacheTrimTokens?: number, primer?: string): Promise<Readable>;
+    exit(): Promise<void>;
+    cancelActiveRequest(): void;
+    getStatus(): {
+        modelName: string;
+        queueLength: number;
+        isStreamingActive: boolean;
+        isJsonBuffering: boolean;
+    };
+}
+//# sourceMappingURL=process-client.d.ts.map

package/dist/local-inference/process-client.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"process-client.d.ts","sourceRoot":"","sources":["../../src/local-inference/process-client.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,EAAE,QAAQ,EAAE,MAAM,QAAQ,CAAC;AAElC,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC;AAE1D,OAAO,KAAK,EACV,qBAAqB,EACrB,yBAAyB,EACzB,qBAAqB,EACrB,uBAAuB,EACvB,2BAA2B,EAC3B,gBAAgB,EAChB,uBAAuB,EACvB,wBAAwB,EACzB,MAAM,eAAe,CAAC;AAKvB,OAAO,EAGL,KAAK,qBAAqB,EAC3B,MAAM,oBAAoB,CAAC;AAE5B,MAAM,WAAW,4BAA4B;IAC3C,SAAS,EAAE,MAAM,CAAC;IAClB,2CAA2C;IAC3C,gBAAgB,EAAE,MAAM,CAAC;IACzB,2CAA2C;IAC3C,QAAQ,EAAE,MAAM,CAAC;IACjB,qCAAqC;IACrC,cAAc,CAAC,EAAE,cAAc,CAAC;IAChC,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;IAC1B,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAClC,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,kBAAkB,CAAC,EAAE,qBAAqB,CAAC;IAC3C,uBAAuB,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,GAAG,IAAI,EAAE,MAAM,EAAE,MAAM,GAAG,IAAI,KAAK,MAAM,CAAC;CAClF;AAED,qBAAa,sBAAsB;IACjC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAE3B,OAAO,CAAC,YAAY,CAAwB;IAC5C,OAAO,CAAC,WAAW,CAAuB;gBAE9B,MAAM,EAAE,4BAA4B;IAkD1C,iBAAiB,IAAI,OAAO,CAAC,IAAI,CAAC;IAIlC,eAAe,IAAI,OAAO,CAAC,qBAAqB,CAAC;IAIjD,UAAU,CACd,QAAQ,EAAE,gBAAgB,EAAE,EAC5B,OAAO,CAAC,EAAE;QAAE,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE,GAC5B,OAAO,CAAC,yBAAyB,CAAC;IAI/B,MAAM,CACV,QAAQ,EAAE,gBAAgB,EAAE,EAC5B,OAAO,CAAC,EAAE,wBAAwB,GAAG;QAAE,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE,EACxD,KAAK,CAAC,EAAE,uBAAuB,EAAE,EACjC,eAAe,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,GAC1C,OAAO,CAAC,qBAAqB,CAAC;IAI3B,QAAQ,CACZ,QAAQ,EAAE,gBAAgB,EAAE,EAC5B,KAAK,CAAC,EAAE,uBAAuB,EAAE,EACjC,eAAe,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,GAC1C,OAAO,CAAC,uBAAuB,CAAC;IAI7B,YAAY,CAChB,SAAS,EAAE,MAAM,EACjB,QAAQ,EAAE,gBAAgB,EAAE,EAC5B,aAAa,CAAC,EAAE,MAAM,EACtB,YAAY,CAAC,EAAE,MAAM,EACrB,aAAa,CAAC,EAAE,MAAM,EAAE,EACxB,YAAY,CAAC,EAAE,MAAM,EAAE,EACvB,KAAK,CAAC,EAAE,uBAAuB,EAAE,EACjC,eAAe,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,GAC1C,OAAO,CAAC,2BAA2B,CAAC;IAajC,UAAU,CACd,MAAM,EAAE,MAAM,EACd,OAAO,CAAC,EAAE,OAAO,EACjB,MAAM,CAAC,EAAE,MAAM,EAAE,EACjB,YAAY,CAAC,EAAE,MAAM,GACpB,OAAO,CAAC,QAAQ,CAAC;IAId,QAAQ,CACZ,MAAM,EAAE,MAAM,GAAG,MAAM,EAAE,EACzB,OAAO,CAAC,EAAE,OAAO,EACjB,MAAM,CAAC,EAAE,MAAM,EAAE,EACjB,YAAY,CAAC,EAAE,MAAM,EACrB,SAAS,CAAC,EAAE,MAAM,EAClB,eAAe,CAAC,EAAE,MAAM,EACxB,MAAM,CAAC,EAAE,MAAM,GACd,OAAO,CAAC,QAAQ,CAAC;IAYd,IAAI,IAAI,OAAO,CAAC,IAAI,CAAC;IAI3B,mBAAmB,IAAI,IAAI;IAI3B,SAAS;;;;;;CAQV"}

package/dist/local-inference/process-client.js

@@ -0,0 +1,92 @@
+/**
+ * Local Inference Protocol プロセスクライアント
+ *
+ * stdio JSON-RPC 通信とリクエストキューを統合する。
+ */
+import { Logger } from '@modular-prompt/utils';
+import { assertRuntimeReady } from '../runtime/index.js';
+import { ProcessCommunication, } from './process-communication.js';
+import { InferenceRequestQueue, } from './request-queue.js';
+export class InferenceProcessClient {
+    modelName;
+    requestQueue;
+    processComm;
+    constructor(config) {
+        this.modelName = config.modelName;
+        if (config.runtimeProfile) {
+            assertRuntimeReady(config.runtimeProfile);
+        }
+        const logger = new Logger({
+            prefix: config.loggerPrefix ?? 'LIP',
+            context: 'process',
+        });
+        // processCallbacks は this.requestQueue 代入前に定義するが、
+        // コールバックは子プロセスの stdout/exit など非同期イベント時のみ呼ばれるため安全。
+        const processCallbacks = {
+            onJsonResponse: (jsonData) => this.requestQueue.handleJsonResponse(jsonData),
+            onRequestCompleted: () => this.requestQueue.onRequestCompleted(),
+            onProcessExit: (code, signal) => {
+                if (code !== 0) {
+                    const message = config.processExitErrorMessage?.(code, signal) ??
+                        `Inference process exited unexpectedly (code=${code}, signal=${signal})`;
+                    logger.error(message);
+                    this.requestQueue.rejectAll(new Error(message));
+                }
+            },
+        };
+        const queueCallbacks = {
+            sendToProcess: (data) => this.processComm.sendToProcess(data),
+            createNewStream: () => this.processComm.createNewStream(),
+            cancelActiveStream: () => this.processComm.cancelActiveStream(),
+        };
+        this.processComm = new ProcessCommunication({
+            pythonProjectDir: config.pythonProjectDir,
+            venvPath: config.venvPath,
+            modelName: config.modelName,
+            extraArgs: config.extraSpawnArgs,
+            extraEnv: config.extraEnv,
+            loggerPrefix: config.loggerPrefix,
+            processExitErrorMessage: config.processExitErrorMessage,
+        }, processCallbacks);
+        this.requestQueue = new InferenceRequestQueue(queueCallbacks, config.mapSamplingOptions);
+    }
+    async ensureInitialized() {
+        // 互換性のため維持
+    }
+    async getCapabilities() {
+        return this.requestQueue.addCapabilitiesRequest();
+    }
+    async formatTest(messages, options) {
+        return this.requestQueue.addFormatTestRequest(messages, options);
+    }
+    async render(messages, options, tools, reasoningEffort) {
+        return this.requestQueue.addRenderRequest(messages, options, tools, reasoningEffort);
+    }
+    async tokenize(messages, tools, reasoningEffort) {
+        return this.requestQueue.addTokenizeRequest(messages, tools, reasoningEffort);
+    }
+    async cachePrefill(cachePath, messages, baseCachePath, trimToTokens, prefixOffsets, prefixHashes, tools, reasoningEffort) {
+        return this.requestQueue.addCachePrefillRequest(cachePath, messages, baseCachePath, trimToTokens, prefixOffsets, prefixHashes, tools, reasoningEffort);
+    }
+    async completion(prompt, options, images, maxImageSize) {
+        return this.requestQueue.addCompletionRequest(prompt, options, images, maxImageSize);
+    }
+    async generate(prompt, options, images, maxImageSize, cachePath, cacheTrimTokens, primer) {
+        return this.requestQueue.addGenerateRequest(prompt, options, images, maxImageSize, cachePath, cacheTrimTokens, primer);
+    }
+    async exit() {
+        await this.processComm.exit();
+    }
+    cancelActiveRequest() {
+        this.requestQueue.cancelActiveRequest();
+    }
+    getStatus() {
+        return {
+            modelName: this.modelName,
+            queueLength: this.requestQueue.length,
+            isStreamingActive: this.processComm.isStreamingActive(),
+            isJsonBuffering: this.processComm.isJsonBuffering(),
+        };
+    }
+}
+//# sourceMappingURL=process-client.js.map

package/dist/local-inference/process-client.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"process-client.js","sourceRoot":"","sources":["../../src/local-inference/process-client.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAGH,OAAO,EAAE,MAAM,EAAE,MAAM,uBAAuB,CAAC;AAE/C,OAAO,EAAE,kBAAkB,EAAE,MAAM,qBAAqB,CAAC;AAWzD,OAAO,EACL,oBAAoB,GAErB,MAAM,4BAA4B,CAAC;AACpC,OAAO,EACL,qBAAqB,GAGtB,MAAM,oBAAoB,CAAC;AAiB5B,MAAM,OAAO,sBAAsB;IACxB,SAAS,CAAS;IAEnB,YAAY,CAAwB;IACpC,WAAW,CAAuB;IAE1C,YAAY,MAAoC;QAC9C,IAAI,CAAC,SAAS,GAAG,MAAM,CAAC,SAAS,CAAC;QAElC,IAAI,MAAM,CAAC,cAAc,EAAE,CAAC;YAC1B,kBAAkB,CAAC,MAAM,CAAC,cAAc,CAAC,CAAC;QAC5C,CAAC;QAED,MAAM,MAAM,GAAG,IAAI,MAAM,CAAC;YACxB,MAAM,EAAE,MAAM,CAAC,YAAY,IAAI,KAAK;YACpC,OAAO,EAAE,SAAS;SACnB,CAAC,CAAC;QAEH,kDAAkD;QAClD,kDAAkD;QAClD,MAAM,gBAAgB,GAAkC;YACtD,cAAc,EAAE,CAAC,QAAQ,EAAE,EAAE,CAAC,IAAI,CAAC,YAAY,CAAC,kBAAkB,CAAC,QAAQ,CAAC;YAC5E,kBAAkB,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,YAAY,CAAC,kBAAkB,EAAE;YAChE,aAAa,EAAE,CAAC,IAAI,EAAE,MAAM,EAAE,EAAE;gBAC9B,IAAI,IAAI,KAAK,CAAC,EAAE,CAAC;oBACf,MAAM,OAAO,GACX,MAAM,CAAC,uBAAuB,EAAE,CAAC,IAAI,EAAE,MAAM,CAAC;wBAC9C,+CAA+C,IAAI,YAAY,MAAM,GAAG,CAAC;oBAC3E,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;oBACtB,IAAI,CAAC,YAAY,CAAC,SAAS,CAAC,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC;gBAClD,CAAC;YACH,CAAC;SACF,CAAC;QAEF,MAAM,cAAc,GAA0B;YAC5C,aAAa,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,WAAW,CAAC,aAAa,CAAC,IAAI,CAAC;YAC7D,eAAe,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,WAAW,CAAC,eAAe,EAAE;YACzD,kBAAkB,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,WAAW,CAAC,kBAAkB,EAAE;SAChE,CAAC;QAEF,IAAI,CAAC,WAAW,GAAG,IAAI,oBAAoB,CACzC;YACE,gBAAgB,EAAE,MAAM,CAAC,gBAAgB;YACzC,QAAQ,EAAE,MAAM,CAAC,QAAQ;YACzB,SAAS,EAAE,MAAM,CAAC,SAAS;YAC3B,SAAS,EAAE,MAAM,CAAC,cAAc;YAChC,QAAQ,EAAE,MAAM,CAAC,QAAQ;YACzB,YAAY,EAAE,MAAM,CAAC,YAAY;YACjC,uBAAuB,EAAE,MAAM,CAAC,uBAAuB;SACxD,EACD,gBAAgB,CACjB,CAAC;QAEF,IAAI,CAAC,YAAY,GAAG,IAAI,qBAAqB,CAAC,cAAc,EAAE,MAAM,CAAC,kBAAkB,CAAC,CAAC;IAC3F,CAAC;IAED,KAAK,CAAC,iBAAiB;QACrB,WAAW;IACb,CAAC;IAED,KAAK,CAAC,eAAe;QACnB,OAAO,IAAI,CAAC,YAAY,CAAC,sBAAsB,EAAE,CAAC;IACpD,CAAC;IAED,KAAK,CAAC,UAAU,CACd,QAA4B,EAC5B,OAA6B;QAE7B,OAAO,IAAI,CAAC,YAAY,CAAC,oBAAoB,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;IACnE,CAAC;IAED,KAAK,CAAC,MAAM,CACV,QAA4B,EAC5B,OAAwD,EACxD,KAAiC,EACjC,eAA2C;QAE3C,OAAO,IAAI,CAAC,YAAY,CAAC,gBAAgB,CAAC,QAAQ,EAAE,OAAO,EAAE,KAAK,EAAE,eAAe,CAAC,CAAC;IACvF,CAAC;IAED,KAAK,CAAC,QAAQ,CACZ,QAA4B,EAC5B,KAAiC,EACjC,eAA2C;QAE3C,OAAO,IAAI,CAAC,YAAY,CAAC,kBAAkB,CAAC,QAAQ,EAAE,KAAK,EAAE,eAAe,CAAC,CAAC;IAChF,CAAC;IAED,KAAK,CAAC,YAAY,CAChB,SAAiB,EACjB,QAA4B,EAC5B,aAAsB,EACtB,YAAqB,EACrB,aAAwB,EACxB,YAAuB,EACvB,KAAiC,EACjC,eAA2C;QAE3C,OAAO,IAAI,CAAC,YAAY,CAAC,sBAAsB,CAC7C,SAAS,EACT,QAAQ,EACR,aAAa,EACb,YAAY,EACZ,aAAa,EACb,YAAY,EACZ,KAAK,EACL,eAAe,CAChB,CAAC;IACJ,CAAC;IAED,KAAK,CAAC,UAAU,CACd,MAAc,EACd,OAAiB,EACjB,MAAiB,EACjB,YAAqB;QAErB,OAAO,IAAI,CAAC,YAAY,CAAC,oBAAoB,CAAC,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,YAAY,CAAC,CAAC;IACvF,CAAC;IAED,KAAK,CAAC,QAAQ,CACZ,MAAyB,EACzB,OAAiB,EACjB,MAAiB,EACjB,YAAqB,EACrB,SAAkB,EAClB,eAAwB,EACxB,MAAe;QAEf,OAAO,IAAI,CAAC,YAAY,CAAC,kBAAkB,CACzC,MAAM,EACN,OAAO,EACP,MAAM,EACN,YAAY,EACZ,SAAS,EACT,eAAe,EACf,MAAM,CACP,CAAC;IACJ,CAAC;IAED,KAAK,CAAC,IAAI;QACR,MAAM,IAAI,CAAC,WAAW,CAAC,IAAI,EAAE,CAAC;IAChC,CAAC;IAED,mBAAmB;QACjB,IAAI,CAAC,YAAY,CAAC,mBAAmB,EAAE,CAAC;IAC1C,CAAC;IAED,SAAS;QACP,OAAO;YACL,SAAS,EAAE,IAAI,CAAC,SAAS;YACzB,WAAW,EAAE,IAAI,CAAC,YAAY,CAAC,MAAM;YACrC,iBAAiB,EAAE,IAAI,CAAC,WAAW,CAAC,iBAAiB,EAAE;YACvD,eAAe,EAAE,IAAI,CAAC,WAAW,CAAC,eAAe,EAAE;SACpD,CAAC;IACJ,CAAC;CACF"}

package/dist/local-inference/process-communication.d.ts

@@ -0,0 +1,41 @@
+/**
+ * Thin Python 推論プロセスとの stdio 通信
+ */
+import { Readable } from 'stream';
+export interface ProcessCommunicationCallbacks {
+    onJsonResponse: (jsonData: string) => void;
+    onRequestCompleted: () => void;
+    onProcessExit: (code: number | null, signal: string | null) => void;
+}
+export interface ProcessCommunicationConfig {
+    /** uv --project に渡す Python プロジェクトディレクトリ */
+    pythonProjectDir: string;
+    /** UV_PROJECT_ENVIRONMENT に設定する venv パス */
+    venvPath: string;
+    modelName: string;
+    /** `__main__.py` と modelName の後に付与する追加引数 */
+    extraArgs?: string[];
+    /** 子プロセス spawn 時にマージする追加環境変数 */
+    extraEnv?: Record<string, string>;
+    loggerPrefix?: string;
+    loggerContext?: string;
+    processExitErrorMessage?: (code: number | null, signal: string | null) => string;
+}
+export declare class ProcessCommunication {
+    private process;
+    private decoder;
+    private currentStream;
+    private jsonBuffer;
+    private draining;
+    private callbacks;
+    private readonly exitErrorMessage;
+    constructor(config: ProcessCommunicationConfig, callbacks: ProcessCommunicationCallbacks);
+    private processData;
+    createNewStream(): Readable;
+    cancelActiveStream(): void;
+    sendToProcess(data: string): void;
+    isStreamingActive(): boolean;
+    isJsonBuffering(): boolean;
+    exit(): Promise<void>;
+}
+//# sourceMappingURL=process-communication.d.ts.map

package/dist/local-inference/process-communication.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"process-communication.d.ts","sourceRoot":"","sources":["../../src/local-inference/process-communication.ts"],"names":[],"mappings":"AAAA;;GAEG;AAGH,OAAO,EAAE,QAAQ,EAAE,MAAM,QAAQ,CAAC;AAIlC,MAAM,WAAW,6BAA6B;IAC5C,cAAc,EAAE,CAAC,QAAQ,EAAE,MAAM,KAAK,IAAI,CAAC;IAC3C,kBAAkB,EAAE,MAAM,IAAI,CAAC;IAC/B,aAAa,EAAE,CAAC,IAAI,EAAE,MAAM,GAAG,IAAI,EAAE,MAAM,EAAE,MAAM,GAAG,IAAI,KAAK,IAAI,CAAC;CACrE;AAED,MAAM,WAAW,0BAA0B;IACzC,2CAA2C;IAC3C,gBAAgB,EAAE,MAAM,CAAC;IACzB,2CAA2C;IAC3C,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,4CAA4C;IAC5C,SAAS,CAAC,EAAE,MAAM,EAAE,CAAC;IACrB,gCAAgC;IAChC,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAClC,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,uBAAuB,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,GAAG,IAAI,EAAE,MAAM,EAAE,MAAM,GAAG,IAAI,KAAK,MAAM,CAAC;CAClF;AAED,qBAAa,oBAAoB;IAC/B,OAAO,CAAC,OAAO,CAAiC;IAChD,OAAO,CAAC,OAAO,CAAgB;IAC/B,OAAO,CAAC,aAAa,CAAyB;IAC9C,OAAO,CAAC,UAAU,CAAc;IAChC,OAAO,CAAC,QAAQ,CAAS;IACzB,OAAO,CAAC,SAAS,CAAgC;IACjD,OAAO,CAAC,QAAQ,CAAC,gBAAgB,CAAyD;gBAE9E,MAAM,EAAE,0BAA0B,EAAE,SAAS,EAAE,6BAA6B;IAoDxF,OAAO,CAAC,WAAW;IAwCnB,eAAe,IAAI,QAAQ;IAO3B,kBAAkB,IAAI,IAAI;IAc1B,aAAa,CAAC,IAAI,EAAE,MAAM,GAAG,IAAI;IAIjC,iBAAiB,IAAI,OAAO;IAI5B,eAAe,IAAI,OAAO;IAIpB,IAAI,IAAI,OAAO,CAAC,IAAI,CAAC;CAe5B"}

package/dist/{mlx-ml/process → local-inference}/process-communication.js

@@ -1,20 +1,10 @@
 /**
- * MLX Driver プロセス通信管理
- *
- * Pythonプロセスとの通信、データの送受信、ストリーミング処理を管理
+ * Thin Python 推論プロセスとの stdio 通信
  */
 import { spawn } from 'child_process';
 import { Readable } from 'stream';
 import { StringDecoder } from 'string_decoder';
-import path from "path";
-import { fileURLToPath } from "url";
 import { Logger } from '@modular-prompt/utils';
-const logger = new Logger({ prefix: 'MLX', context: 'process' });
-// Get the mlx-ml/python directory
-// From dist/mlx-ml/process/ -> go up 3 levels to package root, then to src/mlx-ml/python
-const packageRoot = path.resolve(path.dirname(fileURLToPath(import.meta.url)), '..', '..', '..' // dist/mlx-ml/process -> dist/mlx-ml -> dist -> package root
-);
-const mlxDriverDir = path.join(packageRoot, 'src', 'mlx-ml', 'python');
 export class ProcessCommunication {
     process;
     decoder;
@@ -22,32 +12,34 @@ export class ProcessCommunication {
     jsonBuffer = '';
     draining = false;
     callbacks;
-    constructor(modelName, callbacks, options) {
+    exitErrorMessage;
+    constructor(config, callbacks) {
         this.callbacks = callbacks;
         this.decoder = new StringDecoder('utf8');
+        this.exitErrorMessage =
+            config.processExitErrorMessage ??
+                ((code, signal) => `Inference process exited unexpectedly (code=${code}, signal=${signal})`);
+        const logger = new Logger({
+            prefix: config.loggerPrefix ?? 'LIP',
+            context: config.loggerContext ?? 'process',
+        });
         const args = [
             '--project',
-            mlxDriverDir,
+            config.pythonProjectDir,
             'run',
             'python',
             '__main__.py',
-            modelName
+            config.modelName,
+            ...(config.extraArgs ?? []),
         ];
-        if (options?.textOnly) {
-            args.push('--text-only');
-        }
-        if (options?.drafterModel) {
-            args.push('--drafter', options.drafterModel);
-        }
-        if (options?.draftBlockSize !== undefined) {
-            args.push('--draft-block-size', options.draftBlockSize.toString());
-        }
         this.process = spawn('uv', args, {
-            cwd: mlxDriverDir
+            cwd: config.pythonProjectDir,
+            env: {
+                ...process.env,
+                UV_PROJECT_ENVIRONMENT: config.venvPath,
+                ...config.extraEnv,
+            },
         });
-        this.setupProcessHandlers();
-    }
-    setupProcessHandlers() {
         this.process.stderr.on('data', (data) => {
             logger.debug(data.toString());
         });
@@ -59,36 +51,25 @@ export class ProcessCommunication {
         });
         this.process.on('exit', (code, signal) => {
             if (this.currentStream) {
-                this.currentStream.destroy(new Error(`MLX process exited unexpectedly (code=${code}, signal=${signal})`));
+                this.currentStream.destroy(new Error(this.exitErrorMessage(code, signal)));
                 this.currentStream = null;
             }
             this.callbacks.onProcessExit(code, signal);
         });
     }
-    /**
-     * stdoutから受信したデータを処理する
-     * null文字をレスポンス区切りとして解釈し、
-     * 1つのdataチャンクに複数のレスポンスが含まれる場合も正しく処理する
-     */
     processData(data) {
         let remaining = data;
         while (remaining.length > 0) {
             const nullIndex = remaining.indexOf('\0');
             if (nullIndex !== -1) {
-                // null文字が見つかった場合、レスポンス終了
                 const chunk = this.decoder.write(remaining.slice(0, nullIndex));
                 this.decoder = new StringDecoder('utf8');
                 if (this.currentStream) {
-                    // ストリーミングレスポンスの場合
                     this.currentStream.push(chunk);
-                    this.currentStream.push(null); // ストリーム終了
+                    this.currentStream.push(null);
                     this.currentStream = null;
                 }
-                else if (this.draining) {
-                    // キャンセル後のドレイン: データを破棄
-                }
-                else {
-                    // JSONレスポンスの場合
+                else if (!this.draining) {
                     this.jsonBuffer += chunk;
                     this.callbacks.onJsonResponse(this.jsonBuffer);
                     this.jsonBuffer = '';
@@ -97,21 +78,14 @@ export class ProcessCommunication {
                 if (this.draining) {
                     this.draining = false;
                 }
-                // null文字以降の残りデータを続けて処理
                 remaining = remaining.slice(nullIndex + 1);
             }
             else {
-                // null文字がない場合、データを蓄積
                 const chunk = this.decoder.write(remaining);
                 if (this.currentStream) {
-                    // ストリーミング中
                     this.currentStream.push(chunk);
                 }
-                else if (this.draining) {
-                    // キャンセル後のドレイン: データを破棄
-                }
-                else {
-                    // JSONレスポンス蓄積中
+                else if (!this.draining) {
                     this.jsonBuffer += chunk;
                 }
                 break;
@@ -120,15 +94,10 @@ export class ProcessCommunication {
     }
     createNewStream() {
         this.currentStream = new Readable({
-            read() { } // 空のreadメソッド
+            read() { },
         });
         return this.currentStream;
     }
-    /**
-     * Cancel the active streaming request.
-     * Destroys the current stream, drains stdout until the response terminator,
-     * and sends a cancel command to the Python process.
-     */
     cancelActiveStream() {
         if (!this.currentStream && !this.draining) {
             return;
@@ -140,9 +109,6 @@ export class ProcessCommunication {
         this.draining = true;
         this.sendToProcess(JSON.stringify({ method: 'cancel' }) + '\n');
     }
-    isDraining() {
-        return this.draining;
-    }
     sendToProcess(data) {
         this.process.stdin.write(data);
     }
@@ -155,15 +121,13 @@ export class ProcessCommunication {
     async exit() {
         return new Promise((resolve) => {
             const timeout = setTimeout(() => {
-                // タイムアウト後は強制終了
                 this.process.kill('SIGTERM');
                 resolve();
-            }, 5000); // 5秒でタイムアウト
+            }, 5000);
             this.process.once('exit', () => {
                 clearTimeout(timeout);
                 resolve();
             });
-            // stdinを閉じてプロセスに終了を通知
             this.process.stdin.end();
         });
     }

package/dist/local-inference/process-communication.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"process-communication.js","sourceRoot":"","sources":["../../src/local-inference/process-communication.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAkC,KAAK,EAAE,MAAM,eAAe,CAAC;AACtE,OAAO,EAAE,QAAQ,EAAE,MAAM,QAAQ,CAAC;AAClC,OAAO,EAAE,aAAa,EAAE,MAAM,gBAAgB,CAAC;AAC/C,OAAO,EAAE,MAAM,EAAE,MAAM,uBAAuB,CAAC;AAuB/C,MAAM,OAAO,oBAAoB;IACvB,OAAO,CAAiC;IACxC,OAAO,CAAgB;IACvB,aAAa,GAAoB,IAAI,CAAC;IACtC,UAAU,GAAW,EAAE,CAAC;IACxB,QAAQ,GAAG,KAAK,CAAC;IACjB,SAAS,CAAgC;IAChC,gBAAgB,CAAyD;IAE1F,YAAY,MAAkC,EAAE,SAAwC;QACtF,IAAI,CAAC,SAAS,GAAG,SAAS,CAAC;QAC3B,IAAI,CAAC,OAAO,GAAG,IAAI,aAAa,CAAC,MAAM,CAAC,CAAC;QACzC,IAAI,CAAC,gBAAgB;YACnB,MAAM,CAAC,uBAAuB;gBAC9B,CAAC,CAAC,IAAI,EAAE,MAAM,EAAE,EAAE,CAAC,+CAA+C,IAAI,YAAY,MAAM,GAAG,CAAC,CAAC;QAE/F,MAAM,MAAM,GAAG,IAAI,MAAM,CAAC;YACxB,MAAM,EAAE,MAAM,CAAC,YAAY,IAAI,KAAK;YACpC,OAAO,EAAE,MAAM,CAAC,aAAa,IAAI,SAAS;SAC3C,CAAC,CAAC;QAEH,MAAM,IAAI,GAAG;YACX,WAAW;YACX,MAAM,CAAC,gBAAgB;YACvB,KAAK;YACL,QAAQ;YACR,aAAa;YACb,MAAM,CAAC,SAAS;YAChB,GAAG,CAAC,MAAM,CAAC,SAAS,IAAI,EAAE,CAAC;SAC5B,CAAC;QAEF,IAAI,CAAC,OAAO,GAAG,KAAK,CAAC,IAAI,EAAE,IAAI,EAAE;YAC/B,GAAG,EAAE,MAAM,CAAC,gBAAgB;YAC5B,GAAG,EAAE;gBACH,GAAG,OAAO,CAAC,GAAG;gBACd,sBAAsB,EAAE,MAAM,CAAC,QAAQ;gBACvC,GAAG,MAAM,CAAC,QAAQ;aACnB;SACF,CAAC,CAAC;QAEH,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,EAAE;YACtC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC;QAChC,CAAC,CAAC,CAAC;QAEH,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,EAAE;YACtC,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC;QACzB,CAAC,CAAC,CAAC;QAEH,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,GAAG,EAAE,EAAE;YAC/B,MAAM,CAAC,KAAK,CAAC,sBAAsB,EAAE,GAAG,CAAC,CAAC;QAC5C,CAAC,CAAC,CAAC;QAEH,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,MAAM,EAAE,EAAE;YACvC,IAAI,IAAI,CAAC,aAAa,EAAE,CAAC;gBACvB,IAAI,CAAC,aAAa,CAAC,OAAO,CAAC,IAAI,KAAK,CAAC,IAAI,CAAC,gBAAgB,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC;gBAC3E,IAAI,CAAC,aAAa,GAAG,IAAI,CAAC;YAC5B,CAAC;YACD,IAAI,CAAC,SAAS,CAAC,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;QAC7C,CAAC,CAAC,CAAC;IACL,CAAC;IAEO,WAAW,CAAC,IAAY;QAC9B,IAAI,SAAS,GAAW,IAAI,CAAC;QAE7B,OAAO,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC5B,MAAM,SAAS,GAAG,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;YAE1C,IAAI,SAAS,KAAK,CAAC,CAAC,EAAE,CAAC;gBACrB,MAAM,KAAK,GAAG,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC,CAAC,CAAC;gBAChE,IAAI,CAAC,OAAO,GAAG,IAAI,aAAa,CAAC,MAAM,CAAC,CAAC;gBAEzC,IAAI,IAAI,CAAC,aAAa,EAAE,CAAC;oBACvB,IAAI,CAAC,aAAa,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;oBAC/B,IAAI,CAAC,aAAa,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;oBAC9B,IAAI,CAAC,aAAa,GAAG,IAAI,CAAC;gBAC5B,CAAC;qBAAM,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,CAAC;oBAC1B,IAAI,CAAC,UAAU,IAAI,KAAK,CAAC;oBACzB,IAAI,CAAC,SAAS,CAAC,cAAc,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;oBAC/C,IAAI,CAAC,UAAU,GAAG,EAAE,CAAC;gBACvB,CAAC;gBAED,IAAI,CAAC,SAAS,CAAC,kBAAkB,EAAE,CAAC;gBAEpC,IAAI,IAAI,CAAC,QAAQ,EAAE,CAAC;oBAClB,IAAI,CAAC,QAAQ,GAAG,KAAK,CAAC;gBACxB,CAAC;gBAED,SAAS,GAAG,SAAS,CAAC,KAAK,CAAC,SAAS,GAAG,CAAC,CAAC,CAAC;YAC7C,CAAC;iBAAM,CAAC;gBACN,MAAM,KAAK,GAAG,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;gBAE5C,IAAI,IAAI,CAAC,aAAa,EAAE,CAAC;oBACvB,IAAI,CAAC,aAAa,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;gBACjC,CAAC;qBAAM,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,CAAC;oBAC1B,IAAI,CAAC,UAAU,IAAI,KAAK,CAAC;gBAC3B,CAAC;gBACD,MAAM;YACR,CAAC;QACH,CAAC;IACH,CAAC;IAED,eAAe;QACb,IAAI,CAAC,aAAa,GAAG,IAAI,QAAQ,CAAC;YAChC,IAAI,KAAI,CAAC;SACV,CAAC,CAAC;QACH,OAAO,IAAI,CAAC,aAAa,CAAC;IAC5B,CAAC;IAED,kBAAkB;QAChB,IAAI,CAAC,IAAI,CAAC,aAAa,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,CAAC;YAC1C,OAAO;QACT,CAAC;QAED,IAAI,IAAI,CAAC,aAAa,EAAE,CAAC;YACvB,IAAI,CAAC,aAAa,CAAC,OAAO,EAAE,CAAC;YAC7B,IAAI,CAAC,aAAa,GAAG,IAAI,CAAC;QAC5B,CAAC;QAED,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC;QACrB,IAAI,CAAC,aAAa,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC,GAAG,IAAI,CAAC,CAAC;IAClE,CAAC;IAED,aAAa,CAAC,IAAY;QACxB,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IACjC,CAAC;IAED,iBAAiB;QACf,OAAO,IAAI,CAAC,aAAa,KAAK,IAAI,CAAC;IACrC,CAAC;IAED,eAAe;QACb,OAAO,IAAI,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC;IACpC,CAAC;IAED,KAAK,CAAC,IAAI;QACR,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE;YAC7B,MAAM,OAAO,GAAG,UAAU,CAAC,GAAG,EAAE;gBAC9B,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;gBAC7B,OAAO,EAAE,CAAC;YACZ,CAAC,EAAE,IAAI,CAAC,CAAC;YAET,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,MAAM,EAAE,GAAG,EAAE;gBAC7B,YAAY,CAAC,OAAO,CAAC,CAAC;gBACtB,OAAO,EAAE,CAAC;YACZ,CAAC,CAAC,CAAC;YAEH,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC;QAC3B,CAAC,CAAC,CAAC;IACL,CAAC;CACF"}

package/dist/local-inference/process-port.d.ts

@@ -0,0 +1,12 @@
+import type { Readable } from 'stream';
+import type { InferenceCapabilities, InferenceMessage, InferenceRenderResult, InferenceToolDefinition } from './protocol.js';
+/** LIP プロセスへの最小ポート（MlxProcess / 将来 PyTorchProcess が実装） */
+export interface InferenceProcessPort {
+    ensureInitialized(): Promise<void>;
+    getCapabilities(): Promise<InferenceCapabilities>;
+    render(messages: InferenceMessage[], options?: Record<string, unknown>, tools?: InferenceToolDefinition[], reasoningEffort?: 'low' | 'medium' | 'high'): Promise<InferenceRenderResult>;
+    generate(prompt: string | number[], options?: Record<string, unknown>, images?: string[], maxImageSize?: number, cachePath?: string, cacheTrimTokens?: number, primer?: string): Promise<Readable>;
+    cancelActiveRequest(): void;
+    exit(): Promise<void>;
+}
+//# sourceMappingURL=process-port.d.ts.map

package/dist/local-inference/process-port.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"process-port.d.ts","sourceRoot":"","sources":["../../src/local-inference/process-port.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,QAAQ,CAAC;AACvC,OAAO,KAAK,EACV,qBAAqB,EACrB,gBAAgB,EAChB,qBAAqB,EACrB,uBAAuB,EACxB,MAAM,eAAe,CAAC;AAEvB,0DAA0D;AAC1D,MAAM,WAAW,oBAAoB;IACnC,iBAAiB,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC;IACnC,eAAe,IAAI,OAAO,CAAC,qBAAqB,CAAC,CAAC;IAClD,MAAM,CACJ,QAAQ,EAAE,gBAAgB,EAAE,EAC5B,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EACjC,KAAK,CAAC,EAAE,uBAAuB,EAAE,EACjC,eAAe,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,GAC1C,OAAO,CAAC,qBAAqB,CAAC,CAAC;IAClC,QAAQ,CACN,MAAM,EAAE,MAAM,GAAG,MAAM,EAAE,EACzB,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EACjC,MAAM,CAAC,EAAE,MAAM,EAAE,EACjB,YAAY,CAAC,EAAE,MAAM,EACrB,SAAS,CAAC,EAAE,MAAM,EAClB,eAAe,CAAC,EAAE,MAAM,EACxB,MAAM,CAAC,EAAE,MAAM,GACd,OAAO,CAAC,QAAQ,CAAC,CAAC;IACrB,mBAAmB,IAAI,IAAI,CAAC;IAC5B,IAAI,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC;CACvB"}

package/dist/local-inference/process-port.js

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=process-port.js.map

package/dist/local-inference/process-port.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"process-port.js","sourceRoot":"","sources":["../../src/local-inference/process-port.ts"],"names":[],"mappings":""}

package/dist/local-inference/prompt-utils.d.ts

@@ -0,0 +1,6 @@
+import type { CompiledPrompt } from '@modular-prompt/core';
+/**
+ * Check if the prompt contains MessageElement
+ */
+export declare function hasMessageElement(prompt: CompiledPrompt): boolean;
+//# sourceMappingURL=prompt-utils.d.ts.map

package/dist/local-inference/prompt-utils.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"prompt-utils.d.ts","sourceRoot":"","sources":["../../src/local-inference/prompt-utils.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AAE3D;;GAEG;AACH,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,cAAc,GAAG,OAAO,CAcjE"}

package/dist/local-inference/prompt-utils.js

@@ -0,0 +1,17 @@
+/**
+ * Check if the prompt contains MessageElement
+ */
+export function hasMessageElement(prompt) {
+    const checkElements = (elements) => {
+        if (!elements)
+            return false;
+        return elements.some((element) => {
+            const el = element;
+            return el?.type === 'message';
+        });
+    };
+    return (checkElements(prompt.instructions) ||
+        checkElements(prompt.data) ||
+        checkElements(prompt.output));
+}
+//# sourceMappingURL=prompt-utils.js.map

package/dist/local-inference/prompt-utils.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"prompt-utils.js","sourceRoot":"","sources":["../../src/local-inference/prompt-utils.ts"],"names":[],"mappings":"AAEA;;GAEG;AACH,MAAM,UAAU,iBAAiB,CAAC,MAAsB;IACtD,MAAM,aAAa,GAAG,CAAC,QAAmB,EAAW,EAAE;QACrD,IAAI,CAAC,QAAQ;YAAE,OAAO,KAAK,CAAC;QAC5B,OAAO,QAAQ,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,EAAE;YAC/B,MAAM,EAAE,GAAG,OAA4B,CAAC;YACxC,OAAO,EAAE,EAAE,IAAI,KAAK,SAAS,CAAC;QAChC,CAAC,CAAC,CAAC;IACL,CAAC,CAAC;IAEF,OAAO,CACL,aAAa,CAAC,MAAM,CAAC,YAAY,CAAC;QAClC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC;QAC1B,aAAa,CAAC,MAAM,CAAC,MAAM,CAAC,CAC7B,CAAC;AACJ,CAAC"}

package/dist/local-inference/protocol.d.ts

@@ -0,0 +1,192 @@
+/**
+ * Local Inference Protocol (LIP) 型定義
+ *
+ * Thin Python 推論型ドライバ（MLX / PyTorch 等）が共有する stdio JSON-RPC プロトコル。
+ * Python プロセスからの応答は snake_case のまま受け取る。
+ */
+import type { SpecialToken, SpecialTokenPair } from '../formatter/types.js';
+/** LIP で利用可能な JSON-RPC メソッド */
+export type InferenceMethod = 'capabilities' | 'format_test' | 'render' | 'chat' | 'completion' | 'generate' | 'cache_prefill' | 'tokenize';
+/** VLM content part for structured message content */
+export type InferenceContentPart = {
+    type: 'text';
+    text: string;
+} | {
+    type: 'image';
+};
+/** 標準メッセージ（system / user / assistant） */
+export interface InferenceStandardMessage {
+    role: 'user' | 'assistant' | 'system';
+    content: string | InferenceContentPart[];
+}
+/** tool_calls 付き assistant メッセージ（HuggingFace 互換形式） */
+export interface InferenceAssistantToolCallMessage {
+    role: 'assistant';
+    content: string;
+    tool_calls: Array<{
+        id: string;
+        type: 'function';
+        function: {
+            name: string;
+            arguments: string;
+        };
+    }>;
+}
+/** tool result メッセージ（HuggingFace 互換形式） */
+export interface InferenceToolResultMessage {
+    role: 'tool';
+    content: string;
+    tool_call_id: string;
+    name: string;
+}
+export type InferenceMessage = InferenceStandardMessage | InferenceAssistantToolCallMessage | InferenceToolResultMessage;
+/** HuggingFace apply_chat_template 互換の tool 定義 */
+export interface InferenceToolDefinition {
+    type: 'function';
+    function: {
+        name: string;
+        description?: string;
+        parameters?: Record<string, unknown>;
+    };
+}
+/** サンプリング等の推論オプション（バックエンド共通の最小集合） */
+export interface InferenceSamplingOptions {
+    max_tokens?: number;
+    temperature?: number;
+    top_p?: number;
+    top_k?: number;
+    repetition_penalty?: number;
+    repetition_context_size?: number;
+    trust_remote_code?: boolean;
+}
+export interface InferenceBaseRequest {
+    method: InferenceMethod;
+}
+export interface InferenceCapabilitiesRequest extends InferenceBaseRequest {
+    method: 'capabilities';
+}
+export interface InferenceFormatTestRequest extends InferenceBaseRequest {
+    method: 'format_test';
+    messages: InferenceMessage[];
+    options?: {
+        primer?: string;
+    };
+}
+export interface InferenceTokenizeRequest extends InferenceBaseRequest {
+    method: 'tokenize';
+    messages: InferenceMessage[];
+    tools?: InferenceToolDefinition[];
+    reasoning_effort?: 'low' | 'medium' | 'high';
+}
+export interface InferenceRenderRequest extends InferenceBaseRequest {
+    method: 'render';
+    messages: InferenceMessage[];
+    options?: InferenceSamplingOptions & {
+        primer?: string;
+    };
+    tools?: InferenceToolDefinition[];
+    reasoning_effort?: 'low' | 'medium' | 'high';
+}
+/** @deprecated Python chat ハンドラ廃止。render + generate を使用すること。 */
+export interface InferenceChatRequest extends InferenceBaseRequest {
+    method: 'chat';
+    messages: InferenceMessage[];
+    primer?: string;
+    tools?: InferenceToolDefinition[];
+    options?: InferenceSamplingOptions;
+    images?: string[];
+    maxImageSize?: number;
+    reasoning_effort?: 'low' | 'medium' | 'high';
+    cache_path?: string;
+    cache_trim_tokens?: number;
+}
+export interface InferenceRenderResult {
+    formatted_prompt: string | null;
+    error: string | null;
+}
+export interface InferenceCompletionRequest extends InferenceBaseRequest {
+    method: 'completion';
+    prompt: string;
+    options?: InferenceSamplingOptions;
+    images?: string[];
+    maxImageSize?: number;
+}
+/** 整形済み prompt（文字列 or トークン ID）のストリーム推論 */
+export interface InferenceGenerateRequest extends InferenceBaseRequest {
+    method: 'generate';
+    prompt: string | number[];
+    primer?: string;
+    options?: InferenceSamplingOptions;
+    images?: string[];
+    maxImageSize?: number;
+    cache_path?: string;
+    cache_trim_tokens?: number;
+}
+export interface InferenceCachePrefillRequest extends InferenceBaseRequest {
+    method: 'cache_prefill';
+    cache_path: string;
+    messages: InferenceMessage[];
+    base_cache_path?: string;
+    trim_to_tokens?: number;
+    prefix_offsets?: number[];
+    prefix_hashes?: string[];
+    tools?: InferenceToolDefinition[];
+    reasoning_effort?: 'low' | 'medium' | 'high';
+}
+export interface InferenceCachePrefillResult {
+    cache_path: string;
+    token_count?: number;
+    prefix_offsets?: number[];
+    prefix_hashes?: string[];
+}
+export type InferenceRequest = InferenceCapabilitiesRequest | InferenceFormatTestRequest | InferenceRenderRequest | InferenceTokenizeRequest | InferenceChatRequest | InferenceCompletionRequest | InferenceGenerateRequest | InferenceCachePrefillRequest;
+/** MLX-LM が認識する tool_parser_type（capabilities 経由で参照）。
+ * 現時点では MLX 固有の知識。PyTorch バックエンド追加時（#302 Phase 6）に
+ * バックエンド固有モジュールへ移動・一般化する可能性あり。 */
+export type KnownToolParserType = 'json_tools' | 'pythonic' | 'function_gemma' | 'mistral' | 'kimi_k2' | 'longcat' | 'glm47' | 'qwen3_coder' | 'minimax_m2' | 'gemma4' | 'harmony';
+export interface ToolCallFormat {
+    tool_parser_type?: string;
+    call_start?: string;
+    call_end?: string;
+    response_start?: string;
+    response_end?: string;
+}
+export interface ChatTemplateInfo {
+    supported_roles: string[];
+    preview?: string;
+    constraints: Record<string, unknown>;
+    tool_call_format?: ToolCallFormat;
+}
+/**
+ * Python プロセスから取得するランタイム能力情報（snake_case）
+ */
+export interface InferenceCapabilities {
+    methods: string[];
+    special_tokens: Record<string, SpecialToken | SpecialTokenPair>;
+    features: {
+        apply_chat_template: boolean;
+        vocab_size?: number;
+        model_max_length?: number;
+        chat_template?: ChatTemplateInfo;
+    };
+    chat_restrictions?: {
+        single_system_at_start?: boolean;
+        max_system_messages?: number;
+        alternating_turns?: boolean;
+        requires_user_last?: boolean;
+        allow_empty_messages?: boolean;
+    };
+    model_kind?: 'lm' | 'vlm';
+}
+export interface InferenceFormatTestResult {
+    formatted_prompt: string | null;
+    template_applied: boolean;
+    model_specific_processing: InferenceMessage[] | null;
+    error: string | null;
+}
+export interface InferenceTokenizeResult {
+    token_ids: number[] | null;
+    token_count: number;
+    error: string | null;
+}
+//# sourceMappingURL=protocol.d.ts.map